philosophy phd artificial intelligence

Machine Learning - CMU

Phd program in machine learning.

Carnegie Mellon University's doctoral program in Machine Learning is designed to train students to become tomorrow's leaders through a combination of interdisciplinary coursework, hands-on applications, and cutting-edge research. Graduates of the Ph.D. program in Machine Learning will be uniquely positioned to pioneer new developments in the field, and to be leaders in both industry and academia.

Understanding the most effective ways of using the vast amounts of data that are now being stored is a significant challenge to society, and therefore to science and technology, as it seeks to obtain a return on the huge investment that is being made in computerization and data collection. Advances in the development of automated techniques for data analysis and decision making requires interdisciplinary work in areas such as machine learning algorithms and foundations, statistics, complexity theory, optimization, data mining, etc.

The Ph.D. Program in Machine Learning is for students who are interested in research in Machine Learning.  For questions and concerns, please   contact us .

The PhD program is a full-time in-person committment and is not offered on-line or part-time.

PhD Requirements

Requirements for the phd in machine learning.

• Completion of required courses , (6 Core Courses + 1 Elective)
• Mastery of proficiencies in Teaching and Presentation skills.
• Successful defense of a Ph.D. thesis.

Teaching Ph.D. students are required to serve as Teaching Assistants for two semesters in Machine Learning courses (10-xxx), beginning in their second year. This fulfills their Teaching Skills requirement.

Conference Presentation Skills During their second or third year, Ph.D. students must give a talk at least 30 minutes long, and invite members of the Speaking Skills committee to attend and evaluate it.

Research It is expected that all Ph.D. students engage in active research from their first semester. Moreover, advisor selection occurs in the first month of entering the Ph.D. program, with the option to change at a later time. Roughly half of a student's time should be allocated to research and lab work, and half to courses until these are completed.

Master of Science in Machine Learning Research - along the way to your PhD Degree.

Other Requirements In addition, students must follow all university policies and procedures .

Rules for the MLD PhD Thesis Committee (applicable to all ML PhDs): The committee should be assembled by the student and their advisor, and approved by the PhD Program Director(s).  It must include:

• At least one MLD Core Faculty member
• At least one additional MLD Core or Affiliated Faculty member
• At least one External Member, usually meaning external to CMU
• A total of at least four members, including the advisor who is the committee chair

Financial Support

Application Information

For applicants applying in Fall 2024 for a start date of August 2025 in the Machine Learning PhD program, GRE Scores are OPTIONAL. The committee uses GRE scores to gauge quantitative skills, and to a lesser extent, also verbal skills.

Proof of English Language Proficiency If you will be studying on an F-1 or J-1 visa, and English is not a native language for you (native language…meaning spoken at home and from birth), we are required to formally evaluate your English proficiency. We require applicants who will be studying on an F-1 or J-1 visa, and for whom English is not a native language, to demonstrate English proficiency via one of these standardized tests: TOEFL (preferred), IELTS, or Duolingo.  We discourage the use of the "TOEFL ITP Plus for China," since speaking is not scored. We do not issue waivers for non-native speakers of English.   In particular, we do not issue waivers based on previous study at a U.S. high school, college, or university.  We also do not issue waivers based on previous study at an English-language high school, college, or university outside of the United States.  No amount of educational experience in English, regardless of which country it occurred in, will result in a test waiver.

Submit valid, recent scores:   If as described above you are required to submit proof of English proficiency, your TOEFL, IELTS or Duolingo test scores will be considered valid as follows: If you have not received a bachelor’s degree in the U.S., you will need to submit an English proficiency score no older than two years. (scores from exams taken before Sept. 1, 2023, will not be accepted.) If you are currently working on or have received a bachelor's and/or a master's degree in the U.S., you may submit an expired test score up to five years old. (scores from exams taken before Sept. 1, 2019, will not be accepted.)

Graduate Online Application

• Admissions application opens September 4, 2024
• Early Application Deadline – November 20, 2024 (3:00 p.m. EST)
• Final Application Deadline - December 11, 2024 (3:00 p.m. EST)

Artificial Intelligence

• Skip to content

MastersInAI.org

PhD in Artificial Intelligence Programs

On This Page:

Universities offer a variety of Doctor of Philosophy (Ph.D.) programs related to Artificial Intelligence (AI.) Some of these are titled as Ph.D.s in AI, whereas most are Ph.D.s in Computer Science or related engineering disciplines with a specialization or focus in AI. Admissions requirements usually include a related bachelor’s degree and, sometimes, a master’s degree. Moreover, most Ph.D. programs expect academic excellence and strong recommendations. The AI Ph.D. programs take three to five or more years, depending on if you have a master’s and the complexity of your dissertation. People with Ph.D.s in AI usually go on to tenure track professorships, postdoctoral research positions, or high-level software engineering positions.

What Are Artificial Intelligence Ph.D. Programs?

Ph.D. programs in AI focus on mastering advanced theoretical subjects, such as decision theory, algorithms, optimization, and stochastic processes. Artificial intelligence covers anything where a computer behaves, rationalizes, or learns like a human. Ph.D.s are usually the endpoint to a long educational career. By the time scholars earn Ph.D.s, they have probably been in school for well over 20 years.

People with an AI Ph.D. degree are capable of formulating and executing novel research into the subtopics of AI. Some of the subtopics include:

• Environment adaptation in self-driving vehicles
• Natural language processing in robotics
• Cheating detection in higher education
• Diagnosing and treating diseased in healthcare

AI Ph.D. programs require candidates to focus most of their coursework and research on AI topics. Most culminate in a dissertation of published research. Many AI Ph.D. recipients’ dissertations are published in peer-reviewed journals or presented at industry-leading conferences. They go on to lead careers as experts in AI technology.

Types of Artificial Intelligence Ph.D. Programs

Most AI Ph.D. programs are a Ph.D. in Computer Science with a concentration in AI. These degrees involve general, advanced level computer science courses for the first year or two and then specialize in AI courses and research for the remainder of the curriculum.

AI Ph.D.s offered in other colleges like Computer Engineering, Systems Engineering, Mechanical Engineering, or Electrical Engineering are similar to Ph.D.s in Computer Science. They often involve similar coursework and research. For instance, colleges like Indiana University Bloomington’s Computing and Engineering have departments specializing in AI or Intelligent Engineering. Some colleges, however, may focus more on a specific discipline. For example, a Ph.D. in Mechanical Engineering with an AI focus is more likely to involve electric vehicles than targeted online advertising.

Some AI programs fall under a Computational Linguistics specialization, like CUNY . These programs emphasize the natural language processing aspect of AI. Computational Linguistics programs still involve significant computer science and engineering but also require advanced knowledge in language and speech.

Other unique programs offer a joint Ph.D. with non-engineering disciplines, such as Carnegie Mellon’s Joint Ph.D. in Machine Learning and Public Policy, Statistics, or Neural Computation .

How Ph.D. in Artificial Intelligence Programs Work

Ph.D. programs usually take three to six years to complete. For example, Harvard lays out a three+ year track where the last year(s) is spent completing your research and defending your dissertation. Many Ph.D. programs have a residency requirement where you must take classes on-campus for one to three years. Moreover, most universities, such as Brandeis , require Ph.D. students to grade and/or teach for one to four semesters. Despite these requirements, several Ph.D. programs allow for part-time or full-time students, like Drexel .

Admissions Requirements

Ph.D. programs in AI admit the strongest students. Most applications require a resume, transcripts, letters of recommendation, and a statement of interest. Many programs require a minimum undergraduate GPA of 3.0 or higher, although some allow for statements of explanation if you have a lower GPA due to illness or other excusable causes for a low GPA.

Many universities, like Cornell , recently made the GRE either optional or not required because the GRE provides little prediction into the success of research and represents a COVID-19 risk. These programs may require the GRE again in the future. However, many schools still require the IELTS/TOEFL for international applicants.

Curriculum and Coursework

The curriculum for AI Ph.D.s varies based on the applicants’ prior education for many universities. Some programs allow applicants to receive credit for relevant master’s programs completed prior to admission. The programs require about 30 hours of advanced research and classes. Other programs do not give credit for master’s programs completed elsewhere. These require over 60 hours of electives, in addition to the 30-hours of fundamental and core classes in addition to the advanced courses.

For programs with more specific specialties, the courses are usually narrowly focused. For example, Duke’s Robotics track requires ten classes, at least three of which are focused on AI as it relates to robotics. Others allow for non-AI-specific courses such as computer networks.

Many Ph.D. programs have strict GPA requirements to remain in the program. For example, Northeastern requires PhD candidates to maintain at least a 3.5 GPA. Other programs automatically dismiss students with too many Cs in courses.

Common specializations include:

• Computational Linguistics
• Automotive Systems
• Data Science

Artificial Intelligence Dissertations

Most Ph.D. programs require a dissertation. The dissertation takes at least two years to research and write, usually starting in the second or third year of the Ph.D. curriculum. Moreover, many programs require an oral presentation or defense of the dissertation. Some universities give an award for the best dissertation of the year. For example, Boston University gave a best dissertation award to Hao Chen for the dissertation entitled “ Improving Data Center Efficiency Through Smart Grid Integration and Intelligent Analytics .”

A couple of programs require publications, like Capitol Technology , or additional course electives, like LIU . For example, The Ohio State University requires 27 hours of graded coursework and three hours with an advisor for non-thesis path candidates. Thesis-path candidates only have to take 18 hours of graded coursework but must spend 12 hours with their advisors.

Are There Online Ph.D. in Artificial Intelligence Programs?

Officially, the majority of AI Ph.D. programs are in-person. Only one university, Capitol Technology University , allows for a fully online program. This is one of the most expensive Ph.D.s in the field, costing about $60,000. However, it is also one of the most flexible programs. It allows you to complete your coursework on your own schedule, perhaps even while working. Moreover, it allows for either a dissertation path or a publication path. The coursework is fully focused on AI research and writing, thus eliminating requirements for more general courses like algorithms or networks.

One detail you should consider is that the Capitol Technology Ph.D. program is heavily driven by a faculty mentor. This is someone you will need consistent contact with and open communication. The website only lists the director, so there is a significant element of uncertainty on how the program will work for you. But doctoral candidates who are self-driven and have a solid idea of their research path have a higher likelihood of succeeding.

If you need flexibility in your Ph.D. program, you may find some professors at traditional universities will work with you on how you meet and conduct the research, or you may find an alternative degree program that is online. Although a Ph.D. program may not be officially online, you may be able to spend just a semester or two on campus and then perform the rest of the Ph.D. requirements remotely. This is most likely possible if the university has an online master’s program where you can take classes. For example, the Georgia Institute of Technology does not have a residency requirement, has an online master’s of computer science program , and some professors will work flexibly with doctoral candidates with whom they have a close relationship.

What Jobs Can You Get with a Ph.D. in Artificial Intelligence?

Many Ph.D. graduates work as tenure track professors at universities with AI classes. Others work as postdoc research scientists at universities. Both of these roles are expected to conduct research and publish, but professors have more of an expectation to teach, as well. Universities usually have a small number of these positions available. Moreover, postdoc research positions tend to only last for a limited amount of time.

Other engineers with AI-focused-Ph.D.s conduct research and do software development in the private sector at AI-intensive companies. For example, Google uses AI in many departments. Its assistant uses natural language processing to interface with users through voice. Moreover, Google uses AI to generate news feeds for users. Google, and other industry leaders, have a strong preference for engineers with Ph.D.s. This career path is often highly sought by new Ph.D. recipients.

Another private sector industry shifting to AI is vehicle manufacturing. For example, self-driving cars use significant AI to make ethical and legal decisions while operating. Another example is that electric vehicles use AI techniques to optimize performance and power usage.

Some AI Ph.D. recipients become c-suite executives, such as Chief Technology Officers (CTO). For example, Dr. Ted Gaubert has a Ph.D. in engineering and works as a CTO for an AI-intensive company. Another CTO, Dr. David Talby , revolutionized AI with a new natural language processing library, Spark. CTO positions in AI-focused companies often have decades of experience in the AI field.

How Much Do Ph.D. in Artificial Intelligence Programs Cost?

The tuition for many Ph.D. programs is paid through fellowships, graduate research assistantships, and teaching assistantships. For example, Harvard provides full support for Ph.D. candidates. Some programs mandate teaching or research to attend based on the assumption that Ph.D. candidates need financial assistance.

Fellowships are often reserved for applicants with an exceptional academic and research background. These are usually named for eminent alumni, professors, or other scholars associated with the university. Receiving such a fellowship is a highly respected honor.

For programs that do not provide full assistance, the usual cost is about $500 to $1,000 per credit hour, plus university fees. On the low end, Northern Illinois University charges about $557 per credit hour . With 30 to 60 hours required, this means the programs cost about $30,000 to over $60,000 out of pocket. Typically, Ph.D. programs that do not provide funding for any Ph.D. candidates are less reputable or provide other benefits, such as flexibility, online programs, or fewer requirements.

How Much Does a Ph.D. in AI Make?

Engineers with AI Ph.D.s earn well into the six-figure range in the private sector. For example, OpenAI , a non-profit, pays its top researchers over $400,000 per year. Amazon pays its data scientists with Ph.D.s over $200,000 in salary. Directors and executives with Ph.D.s often earn over $1,000,000 in private industry.

When considering working in the private industry, professionals usually compare offers based on total compensation, not just salary. Many companies offer large stock and bonus packages to Ph.D.-level engineers and scientists.

Startups sometimes pay less in salary, but much more in stock options. For example, the salary may be $50,000 to $100,000, but when the startup goes public, you may end up with hundreds of thousands in stock options. This creates a sense of ownership and investment in the success of the startup.

Computer science professors and postdoctoral researchers earn about $90,000 to $160,000 from universities. However, they increase their competition by writing books, speaking at conferences, and advising companies. Startups often employ professors for advice on the feasibility and design of their technology.

Schools with PhD in Artificial Intelligence Programs

Arizona state university.

School of Computing and Augmented Intelligence

Tempe, Arizona

Ph.D. in Computer Science (Artificial Intelligence Research)

Ph.d. in computing and information sciences (artificial intelligence research), university of california-riverside.

Department of Electrical and Computer Engineering

Riverside, California

Ph.D. in Electrical Engineering - Intelligent Systems Research Area

University of california-san diego.

Electrical and Computer Engineering Department

La Jolla, California

Ph.D. in Intelligent Systems, Robotics and Control

Colorado state university-fort collins.

The Graduate School

Fort Collins, Colorado

Ph.D. in Computer Science - Artificial Intelligence Research Area

University of colorado boulder.

Paul M. Rady Mechanical Engineering

Boulder, Colorado

PhD in Robotics and Systems Design

District of columbia, georgetown university.

Department of Linguistics

Washington, District of Columbia

Doctor of Philosophy (Ph.D.) in Linguistics - Computational Linguistics

The university of west florida.

Department of Intelligent Systems and Robotics

Pensacola, Florida

Ph.D. in Intelligent Systems and Robotics

University of central florida.

Department of Electrical & Computer Engineering

Orlando, Florida

Doctorate in Computer Engineering - Intelligent Systems and Machine Learning

Georgia institute of technology.

Colleges of Computing, Engineering, and Sciences

Atlanta, Georgia

Ph.D. in Machine Learning

Northern illinois university.

Dekalb, Illinois

Ph.D. in Computer Science - Artificial Intelligence Area of Emphasis

Ph.d. in computer science - machine learning area of emphasis, northwestern university.

McCormick School of Engineering

Evanston, Illinois

PhD in Computer Science - Artificial Intelligence and Machine Learning Research Group

Indiana university bloomington.

Department of Intelligent Systems Engineering

Bloomington, Indiana

Ph.D. in Intelligent Systems Engineering

Ph.d. in linguistics - computational linguistics concentration, capitol technology university.

Doctoral Programs Department

Laurel, Maryland

Doctor of Philosophy (PhD) in Artificial Intelligence

Offered Online

Johns Hopkins University

Whiting School of Engineering

Baltimore, Maryland

Doctor of Philosophy in Mechanical Engineering - Robotics

Massachusetts, boston university.

College of Engineering

Boston, Massachusetts

PhD in Computer Engineering - Data Science and Intelligent Systems Research Area

Phd in systems engineering - automation, robotics, and control, brandeis university.

Department of Computer Science

Waltham, Massachusetts

Ph.D. in Computer Science - Computational Linguistics

Harvard university.

School of Engineering and Applied Sciences

Cambridge, Massachusetts

Ph.D. in Applied Mathematics

Northeastern university.

Khoury College of Computer Science

Ph.D. in Computer Science - Artificial Intelligence Area

University of michigan-ann arbor.

Electrical Engineering and Computer Science Department

Ann Arbor, Michigan

PhD in Electrical and Computer Engineering - Robotics

University of nebraska at omaha.

College of Information Science & Technology

Omaha, Nebraska

PhD in Information Technology - Artificial Intelligence Concentration

University of nevada-reno.

Computer Science and Engineering Department

Reno, Nevada

Ph.D. in Computer Science & Engineering - Intelligent and Autonomous Systems Research

Rutgers university.

New Brunswick, New Jersey

Ph.D. in Linguistics with Computational Linguistics Certificate

Stevens institute of technology.

Schaefer School Of Engineering & Science

Hoboken, New Jersey

Ph.D. in Computer Engineering

Ph.d. in electrical engineering - applied artificial intelligence, ph.d. in electrical engineering - robotics and smart systems research, cornell university.

Ithaca, New York

Linguistics Ph.D. - Computational Linguistics

Ph.d.in computer science, cuny graduate school and university center.

New York, New York

Ph.D. in Linguistics - Computational Linguistics

Long island university-brooklyn campus.

Graduate Department

Brooklyn, New York

Dual PharmD/M.S. in Artificial Intelligence

Rochester institute of technology.

Golisano College of Computing and Information Sciences

Rochester, New York

North Carolina

Duke university.

Duke Robotics

Durham, North Carolina

Ph.D in ECE - Robotics Track

Ph.d. in mems - robotics track, ohio state university-main campus.

Department of Mechanical and Aerospace Engineering

Columbus, Ohio

PhD in Mechanical Engineering - Automotive Systems and Mobility (Connected and Automated Vehicles)

University of cincinnati.

College of Engineering and Applied Science

Cincinnati, Ohio

PhD in Computer Science and Engineering - Intelligent Systems Group

Oregon state university.

Corvallis, Oregon

Ph.D. in Artificial Intelligence

Pennsylvania, carnegie mellon university.

Machine Learning Department

Pittsburgh, Pennsylvania

PhD in Machine Learning & Public Policy

Phd in neural computation & machine learning, phd in statistics & machine learning, phd program in machine learning, drexel university.

Philadelphia, Pennsylvania

Doctorate in Mechanical Engineering - Robotics and Autonomy

Temple university.

Computer & Information Sciences Department

PhD in Computer and Information Science - Artificial Intelligence

University of pittsburgh-pittsburgh campus.

School of Computing and Information

Ph.D. in Intelligent Systems

The university of texas at austin.

Austin, Texas

Ph.D. with Graduate Portfolio Program in Robotics

The university of texas at dallas.

Erik Jonsson School of Engineering and Computer Science

Richardson, Texas

University of Utah

Mechanical Engineering Department

Salt Lake City, Utah

Doctor of Philosophy - Robotics Track

University of washington-seattle campus.

Seattle, Washington

Ph.D. in Machine Learning and Big Data

• Accreditations
• Facts and Figures
• Faculty Positions
• Maps and Directions
• Student Mentoring Program
• Current Students
• Engineering Honors
• Global Programs
• Admissions and Aid
• Prospective Students
• Entry to a Major Process
• Graduate Admissions
• Scholarships and Financial Aid
• Facilities and Equipment
• Faculty Research Specialty Areas
• Undergraduate Research
• Affiliated Faculty
• Emeritus Faculty

Doctor of Philosophy in Artificial Intelligence and Data Science

The Doctor of Philosophy (Ph.D.) degree is a research-oriented degree requiring a minimum of 64 semester credit hours of approved courses and research beyond the Master of Science (M.S.) degree [96 credit hours beyond the Bachelor of Science (B.S.) degree]. The university places limitations on these credit hours in addition to the requirements of the Department of Civil Engineering.

A complete discussion of all university requirements is found in the current Texas A&M University Graduate Catalog .

NOTE: All documents requiring departmental signatures must be submitted to the Civil Engineering Graduate Office in DLEB 101 at least one day prior to the Office of Graduate Studies deadline.

Artificial Intelligence and Data Science Faculty Members

• Dr. Mark Burris
• Dr. Nasir Gharaibeh
• Dr. Stefan Hurlebaus
• Dr. Dominique Lord
• Dr. Xingmao “Samuel” Ma
• Dr. Ali Mostafavi
• Dr. Arash Noshadravan
• Dr. Stephanie Paal
• Dr. Luca Quadrifoglio
• Dr. Scott Socolofsky
• Dr. Yunlong Zhang

Admission Admission to the AI/DS track is conditional upon meeting the general admission requirements. Also, students may only be admitted to the AI/DS track if a faculty member affiliated with the track is willing to supervise (and provide funding support via GAT or GAR or Fellowship) for the student. If a current student is approved to change from one track to another, they must complete the Track Change Request Form and send it to the CVEN Graduate Advising Office so notification can be sent to their original area coordinator. Please read the CVEN department policy on changing tracks.

Departmental Requirements In addition to fulfilling the University requirements for the Doctor of Philosophy (Ph.D.) degree, a student enrolled in the Civil Engineering graduate program in the area of Artificial Intelligence and Data Science area must satisfy the following department requirements.

• A minimum of 32 credit hours of graduate-level coursework taken through Texas A&M University [a minimum of 24 credit hours if the student already has taken at least another 24 credit hours of graduate course work for the Master of Science (M.S.) or Master of Engineering (MEng) degree].
• Remaining coursework requirement can be met by 32 hours of CVEN 691.
• Qualifying Exam
• Degree Plan
• Written Preliminary Exam
• Research Proposal
• Oral Preliminary Exam
• Completion of Dissertation
• Final Defense

Dissertation Topic Students pursuing the AI/DS track would work on dissertation topics with a great extent of interdisciplinary elements spanning across civil engineering and computer science/AI. Such interdisciplinary research would require a student to develop depth of knowledge and skills across both domains.

Committee The committee of Ph.D. students in the AI/DS track can be composed of faculty from different departments with backgrounds and skills related to the subject matter of the dissertation research.

Students in the AI/DS track are strongly encouraged to form their dissertation committee prior to the qualification exam. If the dissertation committee is formed prior to the qualification exam, the exam questions will be developed by the committee in coordination with the AI/DS Track Coordinator. If the student's dissertation committee is not formed at the time of the qualification examination, the Track Coordinator and the student advisor will handle the development of the qualification examination.

PhD in Artificial Intelligence

To enter the Doctor of Philosophy in Artificial Intelligence, you must apply online through the UGA  Graduate School web page . There is an application fee, which must be paid at the time the application is submitted.

There are several items which must be included in the application:

• Standardized test scores, including the GRE. 
• 3 letters of recommendation, preferably from university faculty and/or professional supervisors. We encourage you to submit the letters to the graduate school online as you complete the application process.
• A sample of scholarly writing, in English. This can be anything you've written but should give an accurate indication of your writing abilities. The writing sample can be a term paper, research report, journal article, published paper, college paper, etc.
• A completed  Application for Graduate Assistantship , if you are interested in receiving funding. 
• A Statement of Purpose.
• A Resume or Curriculum Vitae.

Further information on program admissions is found in the AI Institute Frequently Asked Questions (FAQ) . 

International Students should also review the links on the  Information for International Students  page for additional information relevant to the application process.

Graduate School Policies

University of Georgia Graduate School policies and requirements apply in addition to (and, in cases of conflict, take precedence over) those described here. It is essential that graduate students familiarize themselves with Graduate School policies, including minimum and continuous enrollment  and other policies contained in the Graduate School Bulletin.

Students should also familiarize themselves with Graduate School Dates and Deadlines relevant to the degree.

Degree Requirements

Students of the doctoral program must complete a minimum of 40 hours of graduate coursework and 6 hours of dissertation credit (for a total of 46 credit hours), pass a comprehensive examination, and write and defend a dissertation. In addition, the University requires that all first-year graduate students enroll in a 1-credit-hour GradFirst seminar . Each of these requirements is described in greater detail below.

The degree program is offered using an in-person format, and classes are in general scheduled for full-time students. There are currently no special provisions for part-time, online, or off-campus students. Students are expected to attend all meetings of classes for which they are registered.

Program of Study

The Program of Study must include a minimum of 40 hours of graduate course work and a minimum of 6 hours of dissertation credit. Of the 40 hours of graduate course work, at least 20 hours must be 8000-level or 9000-level hours.

Required Courses

The following courses must be completed unless specifically waived for students entering the program with a master’s degree in Artificial Intelligence or a related field, or for students with substantially related graduate course work. All waived credits may be replaced by an equal number of doctoral research or doctoral dissertation credits (ARTI 9000, Doctoral Research or ARTI 9300, Doctoral Dissertation). Substitutions must be approved for a particular student by that student's Advisory Committee and by the Graduate Coordinator.

• PHIL/LING 6510  Deductive Systems (3 hours)
• CSCI 6380  Data Mining (4 hours) or CSCI 8950  Machine Learning (4 hours)
• CSCI/PHIL 6550  Artificial Intelligence (3 hours)
• ARTI 6950  Faculty Research Seminar (1 hour)
• ARTI/PHIL 6340 Ethics and Artificial Intelligence (3 hours)

Elective Courses

In addition to the required courses above, at least 6 additional courses must be taken from Groups A and Group B below, subject to the following requirements. 

• At least 2 courses must be taken from Group A, from at least 2 areas.
• At least 2 courses must be taken from Group B, from at least 2 areas.
• At least 3 courses must be taken from a single area comprising the student’s chosen area of emphasis .

Since not all courses have the same number of credit hours, Ph.D. students may need to take additional graduate courses to complete the 40 hours.

AREA 1: Artificial Intelligence Methodologies

• CSCI 6560  Evolutionary Computing (4 hours)
• CSCI 8050  Knowledge Based Systems (4 hours)
• CSCI/PHIL 8650  Logic and Logic Programming (4 hours)
• CSCI 8920  Decision Making Under Uncertainty (4 hours)
• CSCI/ENGR 8940  Computational Intelligence (4 hours)
• CSCI/ARTI 8950  Machine Learning (4 hours)

AREA 2: Machine Learning and Data Science

• CSCI 6360  Data Science II (4 hours)
• CSCI 8360  Data Science Practicum (4 hours)
• CSCI 8945  Advanced Representation Learning (4 hours)
• CSCI 8955  Advanced Data Analytics (4 hours)
• CSCI 8960  Privacy-Preserving Data Analysis (4 hours)

AREA 3: Machine Vision and Robotics

• CSCI/ARTI 6530  Introduction to Robotics (4 hours)
• CSCI 6800  Human Computer Interaction (4 hours)
• CSCI 6850  Biomedical Image Analysis (4 hours)
• CSCI 8850  Advanced Biomedical Image Analysis (4 hours)
• CSCI 8820  Computer Vision and Pattern Recognition (4 hours)
• CSCI 8530  Advanced Topics in Robotics (4 hours)
• CSCI 8535  Multi Robot Systems (4 hours)

AREA 4: Cognitive Modeling and Logic

• PHIL/LING 6300  Philosophy of Language (3 hours)
• PHIL 6310  Philosophy of Mind (3 hours)
• PHIL/LING 6520  Model Theory (3 hours)
• PHIL 8310  Seminar in Philosophy of Mind (max of 3 hours)
• PHIL 8500  Seminar in Problems of Logic (max of 3 hours)
• PHIL 8600  Seminar in Metaphysics (max of 3 hours)
• PHIL 8610  Epistemology (max of 3 hours)
• PSYC 6100  Cognitive Psychology (3 hours)
• PSYC 8240  Judgment and Decision Making (3 hours)
• CSCI 6860  Computational Neuroscience (4 hours)

AREA 5: Language and Computation

• ENGL 6885  Introduction to Humanities Computing (3 hours)
• LING 6021  Phonetics and Phonology (3 hours)
• LING 6080  Language and Complex Systems (3 hours)
• LING 6570  Natural Language Processing (3 hours)
• LING 8150  Generative Syntax (3 hours)
• LING 8580  Seminar in Computational Linguistics (3 hours)

AREA 6: Artificial Intelligence Applications

• ELEE 6280  Introduction to Robotics Engineering (3 hours)
• ENGL 6826  Style: Language, Genre, Cognition (3 hours)
• ENGL/LING 6885  Introduction to Humanities Computing (3 hours)
• FORS 8450  Advanced Forest Planning and Harvest Scheduling (3 hours)
• INFO 8000  Foundations of Informatics for Research and Practice
• MIST 7770  Business Intelligence (3 hours)

Students may under special circumstances use up to 6 hours from the following list to apply towards the Electives group requirement. 

• ARTI 8800  Directed Readings in Artificial Intelligence
• ARTI 8000  Topics in Artificial Intelligence

Other courses may be substituted for those on the Electives lists, provided the subject matter of the course is sufficiently related to artificial intelligence and consistent with the educational objectives of the Ph.D. degree program. Substitutions can be made only with the permission of the student's Advisory Committee and the Graduate Coordinator.

In addition to the specific PhD program requirements, all first-year UGA graduate students must enroll in a 1 credit-hour GRSC 7001 (GradFIRST) seminar which provides foundational training in research, scholarship, and professional development. Students may enroll in a section offered by any department, but it is recommended that they enroll in a section offered by AI Faculty Fellows for AI students. More information is available at the  Graduate School website .

Core Competency

Core competency must be exhibited by each student and certified by the student’s advisory committee. This takes the form of achievement in the required courses of the curriculum. Students entering the Ph.D. program with a previous graduate degree sufficient to cover this basic knowledge will need to work with their advisory committee to certify their core competency. Students entering the Ph.D. program without sufficient graduate background to certify core competency must take at least three of the required courses, and then pursue certification with their advisory committee. A grade average of at least 3.56 (e.g., A-, A-, B+) must be achieved for three required courses (excluding ARTI 6950). Students below this average may take the fourth required course and achieve a grade average of at least 3.32 (e.g., A-, B+, B+, B).

Core competency is certified by the unanimous approval of the student's Advisory Committee as well as the approval by the Graduate Coordinator. Students are strongly encouraged to meet the core competency requirement within their first three enrolled academic semesters (excluding summer semester).  Core Competency Certification must be completed before approval of the Final Program of Study.

Comprehensive Examination

Each student of the doctoral program must pass a Ph.D. Comprehensive Examination covering the student's advanced coursework. The examination consists of a written part and an oral part. Students have at most two attempts to pass the written part. The oral part may not be attempted unless the written part has been passed.

Admission to Candidacy

The student is responsible for initiating an application for admission to candidacy once all requirements, except the dissertation prospectus and the dissertation, have been completed.

Dissertation and Dissertation Credit Hours

In addition to the coursework and comprehensive examination, every student must conduct research in artificial intelligence under the direction of an advisory committee and report the results of his or her research in a dissertation acceptable to the Graduate School. The dissertation must represent originality in research, independent thinking, scholarly ability, and technical mastery of a field of study. The dissertation must also demonstrate competent style and organization. While working on his/her dissertation, the student must enroll for a minimum of 6 credit hours of ARTI 9300 Doctoral Dissertation spread over at least 2 semesters.

Advisory Committee

Before the end of the third semester, each student admitted into the program should approach relevant faculty members and form an advisory committee. Until the committee is formed, the student will be advised by the graduate coordinator. The committee consists of a major professor and two other faculty members, as follows:

• The major professor and at least one other member must be full members of the Graduate Program Faculty.
• The major professor and at least one other member must be Institute for Artificial Intelligence Faculty Fellows.

Deviations from the 3-member advisory committee structure, including having more members, are in some cases permitted but must conform to Graduate School policies. 

The major professor and advisory committee shall guide the student in planning the dissertation.  The committee shall agree upon, document, and communicate expectations for the dissertation. These expectations may include publication or submission requirements, but, should not exceed reasonable expectations for the given research domain. During the planning stage, the student will prepare a dissertation prospectus in the form of a detailed written dissertation proposal. It should clearly define the problem to be addressed, critique the current state-of-the-art, and explain the contributions to research expected by the dissertation work. When the major professor certifies that the dissertation prospectus is satisfactory, it must be formally considered by the advisory committee in a meeting with the student. This formal consideration may not take the place of the comprehensive oral examination.

Approval of the dissertation prospectus signifies that members of the advisory committee believe that it proposes a satisfactory research study. Approval of the prospectus requires the agreement of the advisory committee with no more than one dissenting vote as evidenced by their signing an appropriate form to be filed with the graduate coordinator’s office.  

Graduation Requirements - Forms and Timeline

Before the end of the third semester in residence, a student must begin submitting to the Graduate School, through the graduate coordinator, the following forms: (i) a Preliminary Program of Study Form and (ii) an Advisory Committee Form. The Program of Study Form indicates how and when degree requirements will be met and must be formulated in consultation with the student's major professor. An Application for Graduation Form must also be submitted directly to the Graduate School. Forms and Timing must be submitted as follows:

• Advisory Committee Form (G130)—end of third semester
• Core Competency Form (Internal to IAI)—beginning of fourth semester
• Preliminary Doctoral Program of Study Form—Fourth semester
• Final Program of Study Form (G138)—before Comprehensive Examination
• Application for Admission to Candidacy (G162)—after Comprehensive Examination
• Application for Graduation Form (on Athena)—beginning of last semester
• Approval Form for Doctoral Dissertation (G164)—last semester
• ETD Submission Approval Form (G129)—last semester

Students should frequently check the Graduate School Dates and Deadlines webpage to ensure that all necessary forms are completed in a timely manner.

Student Handbook

Additional information on degree requirements and AI Institute policies can be found in the AI Student Handbook .

For information regarding the graduate programs in IAI, please contact: 

Evette Dunbar [email protected] Boyd GSRC, Room 516 706-542-0358

We appreciate your financial support. Your gift is important to us and helps support critical opportunities for students and faculty alike, including lectures, travel support, and any number of educational events that augment the classroom experience.  Click here to learn more about giving .

Every dollar given has a direct impact upon our students and faculty.

PhD Programs in Artificial Intelligence 

The ultimate graduate degree in artificial intelligence is a doctoral degree, also referred to as a Doctor of Philosophy or PhD. Many PhDs in artificial intelligence are in computer science or engineering with a specialization or research focus in artificial intelligence. PhD students will learn advanced subjects in the discipline, like algorithms, decision theory, stochastic processes, and optimization. 

What are PhD Programs in Artificial Intelligence?

As mentioned above, PhD programs in artificial intelligence are usually PhDs in computer science with a concentration in artificial intelligence. There are other PhDs with specializations in artificial intelligence, as you will see in the titles of many of the degrees listed below. These may include Mechanical Engineering, Machine Learning, and Computational Linguistics.

Most classes in a PhD program will be on-campus, but there may be a few offered online. (Only one PhD program in AI is offered completely online, as you will see in the list below). Some colleges and universities require PhD students to attend school full-time, while others admit part-time students.

Cost  According to EducationData.org, the cost of a Doctor of Philosophy averages $106,860 as of 2023. Programs vary widely in their cost, however, depending upon whether the university is public or private and how much financial assistance is available.

Who Should Study for a PhD in Artificial Intelligence?

Those who get PhDs in artificial intelligence are often interested in research positions, teaching, or software engineering. 

Jobs that are commonly held by those with PhDs in artificial intelligence include, but are not limited to:

• Research scientists
• Software engineers
• Executives in business (Chief Technology Officer, etc.)
• Principal scientist
• Lead data scientist
• Machine learning engineer
• Computer scientist

According to Payscale.com, the average salary for a person holding a PhD in Artificial Intelligence as of 2023 is $115,000 annually. The highest salaries are earned by executives such as Chief Technology Officers holding PhDs in AI, who average $198,235 annually. Computer scientists with PhDs in artificial intelligence fall at the lower end of the salary range, making $91,926 per year.

What are the Admissions Requirements for PhD Programs in Artificial Intelligence? 

In order to be admitted to a PhD program in artificial intelligence, you will need at least a bachelor’s degree or a master’s degree in a related field. You must provide a resume, transcripts, a solid academic background, a personal statement of interest, and strong personal and professional recommendations. The Graduate Record Examination (GRE) may or may not be required. Check with the school(s) in which you are interested in applying for their GRE policies for doctoral students.

Be prepared for a long period of study, as the average PhD program takes three to six years to complete.  Keep in mind that, once admitted, many doctoral programs in AI will require you to maintain a 3.5 GPA in order to remain in the program. 

What is Studied in a PhD Program in Artificial Intelligence?

Courses that you will study in a PhD program in artificial intelligence will vary depending upon the specialization you choose, as well as your prior education. Some schools will give you credit for your master’s degree classes, while others will not. Expect courses in algorithms, analytics, statistics, machine learning, and robotics, among others.

Dissertation

The dissertation is the culmination of your doctoral degree studies, and shows the published research you have conducted on artificial intelligence and/or your specialization. It will take about two years to research and write a dissertation, and you will start working on it during the second or third year of your PhD program. 

Specializations

Potential specializations in PhD programs in AI are listed in the degrees below. They include, but are not limited to:

• Data Science
• Automotive Systems
• Computational Linguistics
• Machine Learning
• Autonomous Systems

Doctoral Programs in Artificial Intelligence 

Below you will find a list of PhD programs in various fields of artificial intelligence across the U.S., as of 2023.

PhD. Programs in Computational Linguistics and Natural Language Processing

District of Columbia

• Doctor of Philosophy in Linguistics: Computational Linguistics
•  Doctor of Philosophy in Linguistics: Concentration in Computational Linguistics

Massachusetts

• Doctor of Philosophy in Computer Science with Studies in Computational Linguistics
• Doctor of Philosophy in Linguistics: Computational Methods

New York 

• Doctor of Philosophy in Linguistics : Computational Linguistics
• Doctor of Philosophy in Linguistics : Concentration in Computational Linguistics
• D octor of Philosophy in Linguistics-Computational Linguistics track

Doctoral Programs in Machine Learning

• Doctor of Philosophy in Computer Engineering: Intelligent Systems and Machine Learning
• Doctor of Philosophy in Machine Learning
• Doctor of Philosophy in Computer Science: Artificial Intelligence and Machine Learning
• Doctor of Philosophy in Computer Science: Machine Learning emphasis
• Doctor of Philosophy in Applied Mathematics: Machine Learning and Artificial Intelligence concentration

Pennsylvania

• Doctor of Philosophy in Statistics & Machine Learning
• Doctor of Philosophy in Machine Learning & Public Policy
• Doctor of Philosophy in Neural Computation & Machine Learning
• Doctor of Philosophy in Statistics: Machine Learning and Big Data track

Doctoral Programs in Robotics and Autonomous Systems

• Doctor of Philosophy in Electrical & Computer Engineering: Intelligent Systems, Robotics and Control
• Doctor of Philosophy in Mechanical Engineering: Robotics and Systems Design
• Doctor of Philosophy in Intelligent Systems & Robotics
• D octor of Philosophy in Mechanical Engineering: Robotics
• Doctor of Philosophy in Systems Engineering: Automation, Robotics and Control
• Doctor of Philosophy in Human-Robot Interaction
• Doctor of Philosophy in Electrical and Computer Engineering: Robotics
• Doctor of Philosophy in Computer Science & Engineering- Intelligent & Autonomous Systems Research

New Jersey 

• Doctor of Philosophy in Electrical Engineering: Robotics and Smart Systems Research

North Carolina

• Doctor of Philosophy in Electrical and Computer Engineering: Signal and Information Processing Track: Robotics Certificate
• Doctor of Philosophy in Mechanical Engineering and Materials Science: Robotics Certificate
• Doctor of Philosophy in Mechanical Engineering: Automotive Systems and Mobility
• Offers BS-PhD and MS-PhD paths
• Doctor of Philosophy in Mechanical Engineering: Robotics and Autonomy

Texas 

• Doctor of Philosophy in Robotics
• Doctor of Philosophy in Mechanical Engineering: Robotics Track
• Doctor of Philosophy in Computer Science: Artificial Intelligence Research
• Doctor of Philosophy in Electrical Engineering: Intelligent Systems Research
• Doctor of Philosophy in Computer Science: Artificial Intelligence Emphasis
• Doctor of Philosophy in Intelligent Systems Engineering
• Doctor of Philosophy in Artificial Intelligence
• Online program
• Doctor of Philosophy in Computer Engineering: Data Science and Intelligent Systems Research Area
• Doctor of Philosophy in Computer Science: Artificial Intelligence
• Doctor of Philosophy in Information Technology: Artificial Intelligence concentration
• Doctor of Philosophy in Computer Engineering
• Doctor of Philosophy in Electrical Engineering: Applied Artificial Intelligence
• Pharm.D. and Master of Science in Artificial Intelligence
• Doctor of Philosophy in Computing and Information Sciences: Artificial Intelligence Research

Oregon 

• Doctor of Philosophy in Computer and Information Science: Artificial Intelligence
• Doctor of Philosophy in Intelligent Systems
• Doctor of Philosophy in Computing and Information Sciences – Artificial Intelligence Research

Master of Philosophy in Artificial Intelligence Doctor of Philosophy in Artificial Intelligence

MPhil(AI) PhD(AI)

MPhil 2 years PhD 3 years (with a relevant research master’s degree), 4 years (without a relevant research master’s degree)

Artificial Intelligence Thrust Area Information Hub

Program Co-Directors: Prof Nevin ZHANG, Professor of Computer Science and Engineering Prof Fangzhen LIN, Professor of Computer Science and Engineering

https://gz.ust.hk/academics/four-hubs/information-hub

The Artificial Intelligence Thrust Area under the Information Hub at HKUST(GZ) is established to project and grow HKUST’s strength in AI and contribute to the nation’s drive to become a world leader in AI. The Master of Philosophy (MPhil) and Doctor of Philosophy (PhD) Programs in Artificial Intelligence are offered as an integral part of the endeavor to establish a world-renowned AI research center with the mission to advance applied research in AI as well as fundamental research relevant to the application areas. The MPhil Program aims to train students in independent and interdisciplinary research in AI. An MPhil graduate is expected to demonstrate sound knowledge in the discipline and is able to synthesize and create new knowledge, making a contribution to the field.

The PhD Program aims to train students in original research in AI, and to cultivate independent, interdisciplinary and innovative thinking that is essential for a successful career in AI-related industry or academia. A PhD graduate is expected to demonstrate mastery of knowledge in the chosen discipline and is able to synthesize and create new knowledge, making original and substantial scientific contributions to the discipline.

On successful completion of the MPhil program, graduates will be able to:

• Demonstrate thorough knowledge of the literature and a comprehensive understanding of scientific methods and techniques relevant to AI;
• Demonstrate practical skills in building AI systems;
• Independently pursue research or innovation of significance in AI applications; and
• Demonstrate skills in oral and written communication sufficient for a professional career.

On successful completion of the PhD program, graduates will be able to:

• Critically apply theories, methodologies, and knowledge to address fundamental questions in AI;

Minimum Credit Requirement

MPhil:  15 credits  PhD:  21 credits   

Credit Transfer 

Students who have taken equivalent courses at HKUST or other recognized universities may be granted credit transfer on a case-by-case basis, up to a maximum of 3 credits for MPhil students, and 6 credits for PhD students.   

Cross-disciplinary Core Courses

2 credits  

All students are required to complete either IIMP 6010 or IIMP 6030. Students may complete the remaining courses as part of the credit requirements, as requested by the Program Planning cum Thesis Supervision Committee.

• Hub Core Courses

Students are required to complete at least one Hub core course (2 credits) from the Information Hub and at least one core course (2 credits) from other Hubs.

  Information Hub Core Course

  Other Hub Core Courses

Courses on Domain Knowledge

MPhil:  minimum 9 credits of coursework  PhD:  minimum 15 credits of coursework   

Under this requirement, each student is required to take one of the required courses, and other electives to form an individualized curriculum relevant to the cross-disciplinary thesis research.  

For PhD, students must complete the AI required course in the first year of their study and obtain a B+ or above. Students who cannot meet the B+ requirement have to retake the course or take another AI required course in the second year to make it up.

Only one Independent Study course may be used to satisfy the course requirements.  

To ensure that students will take appropriate courses to equip them with needed domain knowledge, each student has a Program Planning cum Thesis Supervision Committee to approve the courses to be taken soonest after program commencement and no later than the end of the first year. Depending on the approved curriculum, individual students may be required to complete additional credits beyond the minimal credit requirements.

  Required Course List

Students are required to take one of the required courses listed below:  

  Sample Elective Course List

To meet individual needs, students will be taking courses in different areas, which may include but not limited to courses and areas listed below.  

#: Full-time PhD students are encouraged to take at least 6 months of AI related research internship.  

• Additional Foundation Courses

Individual students may be required to take foundation courses to strengthen their academic background and research capacity in related areas, which will be specified by the Program Planning cum Thesis Supervision Committee. The credits earned cannot be counted toward the credit requirements.

• Graduate Teaching Assistant Training

All full-time RPg students are required to complete PDEV 6800. The course is composed of a 10-hour training offered by the Center for Education Innovation (CEI), and session(s) of instructional delivery to be assigned by the respective departments. Upon satisfactory completion of the training conducted by CEI, MPhil students are required to give at least one 30-minute session of instructional delivery in front of a group of students for one term. PhD students are required to give at least one such session each in two different terms. The instructional delivery will be formally assessed.

• Professional Development Course Requirement

Students are required to complete PDEV 6770. The 1 credit earned from PDEV 6770 cannot be counted toward the credit requirements.

PhD students who are HKUST MPhil graduates and have completed PDEV 6770 or other professional development courses offered by the University before may be exempted from taking PDEV 6770, subject to prior approval of the Program Planning cum Thesis Supervision Committee.

Students are required to complete INFH 6780. The 1 credit earned from INFH 6780 cannot be counted toward the credit requirements.

PhD students who are HKUST MPhil graduates and have completed INFH 6780 or other professional development courses offered by the University before may be exempted from taking INFH 6780, subject to prior approval of the Program Planning cum Thesis Supervision Committee.

• English Language Requirement

Full-time RPg students are required to take an English Language Proficiency Assessment (ELPA) Speaking Test administered by the Center for Language Education before the start of their first term of study. Students whose ELPA Speaking Test score is below Level 4, or who failed to take the test in their first term of study, are required to take LANG 5000 until they pass the course by attaining at least Level 4 in the ELPA Speaking Test before graduation. The 1 credit earned from LANG 5000 cannot be counted toward the credit requirements.  

Students are required to take one of the above three courses. The credit earned cannot be counted toward the credit requirements. Students can be exempted from taking this course with the approval of the Program Planning cum Thesis Supervision Committee.

• Postgraduate Seminar

Students are required to complete AIAA 6101 and AIAA 6102 in two terms. The credit earned cannot be counted toward the credit requirements.

• PhD Qualifying Examination

PhD students are required to pass a qualifying examination to obtain PhD candidacy following established policy.

• Thesis Research

  MPhil:

• Registration in AIAA 6990; and
• Presentation and oral defense of the MPhil thesis.
• Registration in AIAA 7990; and
• Presentation and oral defense of the PhD thesis.

Last Update: 24 July 2020

To qualify for admission, applicants must meet all of the following requirements. Admission is selective and meeting these minimum requirements does not guarantee admission.

Applicants seeking admission to a master's degree program should have obtained a bachelor’s degree from a recognized institution, or an approved equivalent qualification;

Applicants seeking admission to a doctoral degree program should have obtained a bachelor’s degree with a proven record of outstanding performance from a recognized institution; or presented evidence of satisfactory work at the postgraduate level on a full-time basis for at least one year, or on a part-time basis for at least two years.

Applicants have to fulfill English Language requirements with one of the following proficiency attainments:

TOEFL-iBT: 80*

TOEFL-pBT: 550

TOEFL-Revised paper-delivered test: 60 (total scores for Reading, Listening and Writing sections)

IELTS (Academic Module): Overall score: 6.5  and All sub-score: 5.5

* refers to the total score in one single attempt

Applicants are not required to present TOEFL or IELTS score if

their first language is English, or

they obtained the bachelor's degree (or equivalent) from an institution where the medium of instruction was English.

© 2017 HKUST    ALL RIGHTS RESERVED    DESIGNED BY PTC

Department of Linguistics and Philosophy

Ethics & ai @ mit.

Computing and AI: Humanistic Perspectives from MIT

The advent of artificial intelligence presents our species with an historic opportunity — disguised as an existential challenge: Can we stay human in the age of AI?

The Tools of Moral Philosophy

"Framing a discussion of the risks of advanced technology entirely in terms of ethics suggests that the problems raised are ones that can and should be solved by individual action."

Computing and AI Series: Humanistic Perspective from Philosophy at MIT

“The MIT Schwartzman College of Computing presents an opportunity for MIT to be an intellectual leader in the ethics of technology. The Ethics Lab we propose could turn this opportunity into reality."

Ethical AI by Design | Q&A with Abby Everett Jaques PhD ’18

MIT postdoctoral associate is bringing philosophical tools to ethical questions related to information technologies.

Computing and AI Series: Humanistic Perspective from Women’s and Gender Studies at MIT

"The Schwarzman College of Computing presents MIT with a unique opportunity to take a leadership role in addressing some of most pressing challenges that have emerged from the ubiquitous role computing technologies now play in our society — including [...]

SHASS Computing and Society Concentration

The Computing and Society Concentration introduces students to critical thinking about the social and ethical dimensions of computing.

Drawing on selected classes from nine MIT-SHASS units, this concentration helps students understand that computing is a social practice with profound human implications, and that the humanities and social sciences offer insights to improve the cultural, ethical and political impact of contemporary and future technology.

Concentration requirement: four subjects taken from nine MIT-SHASS units:

The Schwarzman College of Computing

A new era of computing and AI education, research, and innovation.

Ethics & Computing Lecture Series

A lecture series focused on the interface of ethics, politics, and computing. 

Philosophy & Computing Email List

Subscribe for anything to do with philosophy and computing

Department of Philosophy

Dietrich college of humanities and social sciences, ethics & artificial intelligence.

Recent advances in computer science and robotics are rapidly producing computational systems with the ability to perform tasks that were once reserved solely for humans. In a variety of sectors of life, from driverless cars and automated factories, to medical diagnostics and personal care robots, to military drones and cyber defense systems, the deployment of computational decision makers raises a complex array of ethical issues.

Work on these issues in the Philosophy Department is distinctive for its interdisciplinary character , its deep connections with the technical disciplines driving these advances, and its focus on concrete ethical problems that are pressing issues now, or (likely) in the next five to ten years. It is also distinctive for its focus on the likely broader social impacts of the integration of artificial decision makers into various social systems, including the assessment of ways that human social relations are likely to change in light of the new forms of interaction that such systems are likely to enable. Moreover, we work to proactively shape the development of these technologies towards more ethical processes, products, and outcomes.

Danks has explored the importance of trust in systems with autonomous capabilities, including the ways this important ethical value can impact human understanding of those systems, and our incorporation of them into functional and effective teams. He has examined these issues in a range of domains, including security & defense, cyber-systems, healthcare, and autonomous vehicles. This work raises and addresses basic questions about responsibility, liability, and the permissible division of labor between human and computational systems.

London and Danks have explored challenges in the development and regulation of autonomous systems, and the respects in which oversight mechanisms from other areas might usefully be emulated in this context. Their work draws heavily on a model of technological development articulated by London and colleagues in the context of drug development. They have also explored issues of bias in algorithms and the way that normative standards for ethically appropriate processes and outcomes are essential to evaluating the potential for different kinds of bias in algorithms.

As ambitions increase for developing computational systems that are capable of making a wider range of ethical decisions it becomes increasingly important to understand the structure of ethical decision making and the prospects for implementing such structures in various kinds of formal systems. Here, foundational issues about methodology in theoretical and applied ethics intersect with formal methods from game and decision theory, causal modeling, and machine learning.

Finally, the work of several faculty, such as Wenner and London, is relevant to understanding the conditions under which these new technologies might be used in ways that enhance human capabilities or exacerbate domination or exploitation.

• Graduate Application
• Support Philosophy @ CMU

Doctor of Philosophy with a major in Machine Learning

The Doctor of Philosophy with a major in Machine Learning program has the following principal objectives, each of which supports an aspect of the Institute’s mission:

• Create students that are able to advance the state of knowledge and practice in machine learning through innovative research contributions.
• Create students who are able to integrate and apply principles from computing, statistics, optimization, engineering, mathematics and science to innovate, and create machine learning models and apply them to solve important real-world data intensive problems.
• Create students who are able to participate in multidisciplinary teams that include individuals whose primary background is in statistics, optimization, engineering, mathematics and science.
• Provide a high quality education that prepares individuals for careers in industry, government (e.g., national laboratories), and academia, both in terms of knowledge, computational (e.g., software development) skills, and mathematical modeling skills.
• Foster multidisciplinary collaboration among researchers and educators in areas such as computer science, statistics, optimization, engineering, social science, and computational biology.
• Foster economic development in the state of Georgia.
• Advance Georgia Tech’s position of academic leadership by attracting high quality students who would not otherwise apply to Tech for graduate study.

All PhD programs must incorporate a standard set of Requirements for the Doctoral Degree .

The central goal of the PhD program is to train students to perform original, independent research.  The most important part of the curriculum is the successful defense of a PhD Dissertation, which demonstrates this research ability.  The academic requirements are designed in service of this goal.

The curriculum for the PhD in Machine Learning is truly multidisciplinary, containing courses taught in nine schools across three colleges at Georgia Tech: the Schools of Computational Science and Engineering, Computer Science, and Interactive Computing in the College of Computing; the Schools of Aerospace Engineering, Chemical and Biomolecular Engineering, Industrial and Systems Engineering, Electrical and Computer Engineering, and Biomedical Engineering in the College of Engineering; and the School of Mathematics in the College of Science.

Summary of General Requirements for a PhD in Machine Learning

• Core curriculum (4 courses, 12 hours). Machine Learning PhD students will be required to complete courses in four different areas: Mathematical Foundations, Probabilistic and Statistical Methods in Machine Learning, ML Theory and Methods, and Optimization.   
• Area electives (5 courses, 15 hours).
• Responsible Conduct of Research (RCR) (1 course, 1 hour, pass/fail).  Georgia Tech requires that all PhD students complete an RCR requirement that consists of an online component and in-person training. The online component is completed during the student’s first semester enrolled at Georgia Tech.  The in-person training is satisfied by taking PHIL 6000 or their associated academic program’s in-house RCR course.
• Qualifying examination (1 course, 3 hours). This consists of a one-semester independent literature review followed by an oral examination.
• Doctoral minor (2 courses, 6 hours).
• Research Proposal.  The purpose of the proposal is to give the faculty an opportunity to give feedback on the student’s research direction, and to make sure they are developing into able communicators.
• PhD Dissertation.

Almost all of the courses in both the core and elective categories are already taught regularly at Georgia Tech.  However, two core courses (designated in the next section) are being developed specifically for this program.  The proposed outlines for these courses can be found in the Appendix. Students who complete these required courses as part of a master’s program will not need to repeat the courses if they are admitted to the ML PhD program.

Core Courses

Machine Learning PhD students will be required to complete courses in four different areas. With the exception of the Foundations course, each of these area requirements can be satisfied using existing courses from the College of Computing or Schools of ECE, ISyE, and Mathematics.

Machine Learning core:

Mathematical Foundations of Machine Learning. This required course is the gateway into the program, and covers the key subjects from applied mathematics needed for a rigorous graduate program in ML. Particular emphasis will be put on advanced concepts in linear algebra and probabilistic modeling. This course is cross-listed between CS, CSE, ECE, and ISyE.

ECE 7750 / ISYE 7750 / CS 7750 / CSE 7750 Mathematical Foundations of Machine Learning

Probabilistic and Statistical Methods in Machine Learning

• ISYE 6412 , Theoretical Statistics
• ECE 7751 / ISYE 7751 / CS 7751 / CSE 7751 Probabilistic Graphical Models
• MATH 7251 High Dimension Probability
• MATH 7252 High Dimension Statistics

Machine Learning: Theory and Methods.   This course serves as an introduction to the foundational problems, algorithms, and modeling techniques in machine learning.  Each of the courses listed below treats roughly the same material using a mix of applied mathematics and computer science, and each has a different balance between the two. 

• CS 7545 Machine Learning Theory and Methods
• CS 7616 , Pattern Recognition
• CSE 6740 / ISYE 6740 , Computational Data Analysis
• ECE 6254 , Statistical Machine Learning
• ECE 6273 , Methods of Pattern Recognition with Applications to Voice

Optimization.   Optimization plays a crucial role in both developing new machine learning algorithms and analyzing their performance.  The three courses below all provide a rigorous introduction to this topic; each emphasizes different material and provides a unique balance of mathematics and algorithms.

• ECE 8823 , Convex Optimization: Theory, Algorithms, and Applications
• ISYE 6661 , Linear Optimization
• ISYE 6663 , Nonlinear Optimization
• ISYE 7683 , Advanced Nonlinear Programming

After core requirements are satisfied, all courses listed in the core not already taken can be used as (appropriately classified) electives.

In addition to meeting the core area requirements, each student is required to complete five elective courses. These courses are required for getting a complete breadth in ML. These courses must be chosen from at least two of the five subject areas listed below. In addition, students can use up to six special problems research hours to satisfy this requirement. 

i. Statistics and Applied Probability : To build breadth and depth in the areas of statistics and probability as applied to ML.

• AE 6505 , Kalman Filtering
• AE 8803 Gaussian Processes
• BMED 6700 , Biostatistics
• ECE 6558 , Stochastic Systems
• ECE 6601 , Random Processes
• ECE 6605 , Information Theory
• ISYE 6402 , Time Series Analysis
• ISYE 6404 , Nonparametric Data Analysis
• ISYE 6413 , Design and Analysis of Experiments
• ISYE 6414 , Regression Analysis
• ISYE 6416 , Computational Statistics
• ISYE 6420 , Bayesian Statistics
• ISYE 6761 , Stochastic Processes I
• ISYE 6762 , Stochastic Processes II
• ISYE 7400 , Adv Design-Experiments
• ISYE 7401 , Adv Statistical Modeling
• ISYE 7405 , Multivariate Data Analysis
• ISYE 8803 , Statistical and Probabilistic Methods for Data Science
• ISYE 8813 , Special Topics in Data Science
• MATH 6221 , Probability Theory for Scientists and Engineers
• MATH 6266 , Statistical Linear Modeling
• MATH 6267 , Multivariate Statistical Analysis
• MATH 7244 , Stochastic Processes and Stochastic Calculus I
• MATH 7245 , Stochastic Processes and Stochastic Calculus II

ii. Advanced Theory: To build a deeper understanding of foundations of ML.

• AE 8803 , Optimal Transport Theory and Applications
• CS 7280 , Network Science
• CS 7510 , Graph Algorithms
• CS 7520 , Approximation Algorithms
• CS 7530 , Randomized Algorithms
• CS 7535 , Markov Chain Monte Carlo Algorithms
• CS 7540 , Spectral Algorithms
• CS 8803 , Continuous Algorithms
• ECE 6283 , Harmonic Analysis and Signal Processing
• ECE 6555 , Linear Estimation
• ISYE 7682 , Convexity
• MATH 6112 , Advanced Linear Algebra
• MATH 6241 , Probability I
• MATH 6262 , Advanced Statistical Inference
• MATH 6263 , Testing Statistical Hypotheses
• MATH 6580 , Introduction to Hilbert Space
• MATH 7338 , Functional Analysis
• MATH 7586 , Tensor Analysis
• MATH 88XX, Special Topics: High Dimensional Probability and Statistics

iii. Applications: To develop a breadth and depth in variety of applications domains impacted by/with ML.

• AE 6373 , Advanced Design Methods
• AE 8803 , Machine Learning for Control Systems
• AE 8803 , Nonlinear Stochastic Optimal Control
• BMED 6780 , Medical Image Processing
• BMED 6790 / ECE 6790 , Information Processing Models in Neural Systems
• BMED 7610 , Quantitative Neuroscience
• BMED 8813 BHI, Biomedical and Health Informatics
• BMED 8813 MHI, mHealth Informatics
• BMED 8813 MLB, Machine Learning in Biomedicine
• BMED 8823 ALG, OMICS Data and Bioinformatics Algorithms
• CHBE 6745 , Data Analytics for Chemical Engineers
• CHBE 6746 , Data-Driven Process Engineering
• CS 6440 , Introduction to Health Informatics
• CS 6465 , Computational Journalism
• CS 6471 , Computational Social Science
• CS 6474 , Social Computing
• CS 6475 , Computational Photography
• CS 6476 , Computer Vision
• CS 6601 , Artificial Intelligence
• CS 7450 , Information Visualization
• CS 7476 , Advanced Computer Vision
• CS 7630 , Autonomous Robots
• CS 7632 , Game AI
• CS 7636 , Computational Perception
• CS 7643 , Deep Learning
• CS 7646 , Machine Learning for Trading
• CS 7647 , Machine Learning with Limited Supervision
• CS 7650 , Natural Language Processing
• CSE 6141 , Massive Graph Analysis
• CSE 6240 , Web Search and Text Mining
• CSE 6242 , Data and Visual Analytics
• CSE 6301 , Algorithms in Bioinformatics and Computational Biology
• ECE 4580 , Computational Computer Vision
• ECE 6255 , Digital Processing of Speech Signals
• ECE 6258 , Digital Image Processing
• ECE 6260 , Data Compression and Modeling
• ECE 6273 , Methods of Pattern Recognition with Application to Voice
• ECE 6550 , Linear Systems and Controls
• ECE 8813 , Network Security
• ISYE 6421 , Biostatistics
• ISYE 6810 , Systems Monitoring and Prognosis
• ISYE 7201 , Production Systems
• ISYE 7204 , Info Prod & Ser Sys
• ISYE 7203 , Logistics Systems
• ISYE 8813 , Supply Chain Inventory Theory
• HS 6000 , Healthcare Delivery
• MATH 6759 , Stochastic Processes in Finance
• MATH 6783 , Financial Data Analysis

iv. Computing and Optimization: To provide more breadth and foundation in areas of math, optimization and computation for ML.

• AE 6513 , Mathematical Planning and Decision-Making for Autonomy
• AE 8803 , Optimization-Based Learning Control and Games
• CS 6515 , Introduction to Graduate Algorithms
• CS 6550 , Design and Analysis of Algorithms
• CSE 6140 , Computational Science and Engineering Algorithms
• CSE 6643 , Numerical Linear Algebra
• CSE 6644 , Iterative Methods for Systems of Equations
• CSE 6710 , Numerical Methods I
• CSE 6711 , Numerical Methods II
• ECE 6553 , Optimal Control and Optimization
• ISYE 6644 , Simulation
• ISYE 6645 , Monte Carlo Methods
• ISYE 6662 , Discrete Optimization
• ISYE 6664 , Stochastic Optimization
• ISYE 6679 , Computational methods for optimization
• ISYE 7686 , Advanced Combinatorial Optimization
• ISYE 7687 , Advanced Integer Programming

v. Platforms : To provide breadth and depth in computing platforms that support ML and Computation.

• CS 6421 , Temporal, Spatial, and Active Databases
• CS 6430 , Parallel and Distributed Databases
• CS 6290 , High-Performance Computer Architecture
• CSE 6220 , High Performance Computing
• CSE 6230 , High Performance Parallel Computing

Qualifying Examination

The purpose of the Qualifying Examination is to judge the candidate’s potential as an independent researcher.

The Ph.D. qualifying exam consists of a focused literature review that will take place over the course of one semester.  At the beginning of the second semester of their second year, a qualifying committee consisting of three members of the ML faculty will assign, in consultation with the student and the student’s advisor, a course of study consisting of influential papers, books, or other intellectual artifacts relevant to the student’s research interests.  The student’s focus area and current research efforts (and related portfolio) will be considered in defining the course of study.

At the end of the semester, the student will submit a written summary of each artifact which highlights their understanding of the importance (and weaknesses) of the work in question and the relationship of this work to their current research.  Subsequently, the student will have a closed oral exam with the three members of the committee.  The exam will be interactive, with the student and the committee discussing and criticizing each work and posing questions related the students current research to determine the breadth of student’s knowledge in that specific area.  

The success of the examination will be determined by the committee’s qualitative assessment of the student’s understanding of the theory, methods, and ultimate impact of the assigned syllabus.

The student will be given a passing grade for meeting the requirements of the committee in both the written and the oral part. Unsatisfactory performance on either part will require the student to redo the entire qualifying exam in the following semester year. Each student will be allowed only two attempts at the exam.

Students are expected to perform the review by the end of their second year in the program.

Doctoral Dissertation

The primary requirement of the PhD student is to do original and substantial research.  This research is reported for review in the PhD dissertation, and presented at the final defense.  As the first step towards completing a dissertation, the student must prepare and defend a Research Proposal.  The proposal is a document of no more than 20 pages in length that carefully describes the topic of the dissertation, including references to prior work, and any preliminary results to date.  The written proposal is submitted to a committee of three faculty members from the ML PhD program, and is presented in a public seminar shortly thereafter.  The committee members provide feedback on the proposed research directions, comments on the strength of writing and oral presentation skills, and might suggest further courses to solidify the student’s background.  Approval of the Research Proposal by the committee is required at least six months prior to the scheduling of the PhD defense. It is expected that the student complete this proposal requirement no later than their fourth year in the program. The PhD thesis committee consists of five faculty members: the student’s advisor, three additional members from the ML PhD program, and one faculty member external to the ML program.  The committee is charged with approving the written dissertation and administering the final defense.  The defense consists of a public seminar followed by oral examination from the thesis committee.

Doctoral minor (2 courses, 6 hours): 

The minor follows the standard Georgia Tech requirement: 6 hours, preferably outside the student’s home unit, with a GPA in those graduate-level courses of at least 3.0.  The courses for the minor should form a cohesive program of study outside the area of Machine Learning; no ML core or elective courses may be used to fulfill this requirement and must be approved by your thesis advisor and ML Academic Advisor.  Typical programs will consist of three courses two courses from the same school (any school at the Institute) or two courses from the same area of study. 

This site uses cookies. Review the Privacy & Legal Notice . Email questions to [email protected]

Print Options

Send Page to Printer

Print this page.

Download Page (PDF)

The PDF will include all information unique to this page.

• Table of Contents
• Random Entry
• Chronological
• Editorial Information
• About the SEP
• Editorial Board
• How to Cite the SEP
• Special Characters
• Advanced Tools
• Support the SEP
• PDFs for SEP Friends
• Make a Donation
• SEPIA for Libraries
• Entry Contents

Bibliography

Academic tools.

• Friends PDF Preview
• Author and Citation Info
• Back to Top

Artificial Intelligence

Artificial intelligence (AI) is the field devoted to building artificial animals (or at least artificial creatures that – in suitable contexts – appear to be animals) and, for many, artificial persons (or at least artificial creatures that – in suitable contexts – appear to be persons). [ 1 ] Such goals immediately ensure that AI is a discipline of considerable interest to many philosophers, and this has been confirmed (e.g.) by the energetic attempt, on the part of numerous philosophers, to show that these goals are in fact un/attainable. On the constructive side, many of the core formalisms and techniques used in AI come out of, and are indeed still much used and refined in, philosophy: first-order logic and its extensions; intensional logics suitable for the modeling of doxastic attitudes and deontic reasoning; inductive logic, probability theory, and probabilistic reasoning; practical reasoning and planning, and so on. In light of this, some philosophers conduct AI research and development as philosophy.

In the present entry, the history of AI is briefly recounted, proposed definitions of the field are discussed, and an overview of the field is provided. In addition, both philosophical AI (AI pursued as and out of philosophy) and philosophy of AI are discussed, via examples of both. The entry ends with some de rigueur speculative commentary regarding the future of AI.

1. The History of AI

2. what exactly is ai, 3.1 the intelligent agent continuum, 3.2 logic-based ai: some surgical points, 3.3 non-logicist ai: a summary, 3.4 ai beyond the clash of paradigms, 4.1 bloom in machine learning, 4.2 the resurgence of neurocomputational techniques, 4.3 the resurgence of probabilistic techniques, 5. ai in the wild, 6. moral ai, 7. philosophical ai, 8.1 “strong” versus “weak” ai, 8.2 the chinese room argument against “strong ai”, 8.3 the gödelian argument against “strong ai”, 8.4 additional topics and readings in philosophy of ai, 9. the future, online courses on ai, related entries.

The field of artificial intelligence (AI) officially started in 1956, launched by a small but now-famous DARPA -sponsored summer conference at Dartmouth College, in Hanover, New Hampshire. (The 50-year celebration of this conference, AI@50 , was held in July 2006 at Dartmouth, with five of the original participants making it back. [ 2 ] What happened at this historic conference figures in the final section of this entry.) Ten thinkers attended, including John McCarthy (who was working at Dartmouth in 1956), Claude Shannon, Marvin Minsky, Arthur Samuel, Trenchard Moore (apparently the lone note-taker at the original conference), Ray Solomonoff, Oliver Selfridge, Allen Newell, and Herbert Simon. From where we stand now, into the start of the new millennium, the Dartmouth conference is memorable for many reasons, including this pair: one, the term ‘artificial intelligence’ was coined there (and has long been firmly entrenched, despite being disliked by some of the attendees, e.g., Moore); two, Newell and Simon revealed a program – Logic Theorist (LT) – agreed by the attendees (and, indeed, by nearly all those who learned of and about it soon after the conference) to be a remarkable achievement. LT was capable of proving elementary theorems in the propositional calculus. [ 3 ] [ 4 ]

Though the term ‘artificial intelligence’ made its advent at the 1956 conference, certainly the field of AI, operationally defined (defined, i.e., as a field constituted by practitioners who think and act in certain ways), was in operation before 1956. For example, in a famous Mind paper of 1950, Alan Turing argues that the question “Can a machine think?” (and here Turing is talking about standard computing machines: machines capable of computing functions from the natural numbers (or pairs, triples, … thereof) to the natural numbers that a Turing machine or equivalent can handle) should be replaced with the question “Can a machine be linguistically indistinguishable from a human?.” Specifically, he proposes a test, the “ Turing Test ” (TT) as it’s now known. In the TT, a woman and a computer are sequestered in sealed rooms, and a human judge, in the dark as to which of the two rooms contains which contestant, asks questions by email (actually, by teletype , to use the original term) of the two. If, on the strength of returned answers, the judge can do no better than 50/50 when delivering a verdict as to which room houses which player, we say that the computer in question has passed the TT. Passing in this sense operationalizes linguistic indistinguishability. Later, we shall discuss the role that TT has played, and indeed continues to play, in attempts to define AI. At the moment, though, the point is that in his paper, Turing explicitly lays down the call for building machines that would provide an existence proof of an affirmative answer to his question. The call even includes a suggestion for how such construction should proceed. (He suggests that “child machines” be built, and that these machines could then gradually grow up on their own to learn to communicate in natural language at the level of adult humans. This suggestion has arguably been followed by Rodney Brooks and the philosopher Daniel Dennett (1994) in the Cog Project. In addition, the Spielberg/Kubrick movie A.I. is at least in part a cinematic exploration of Turing’s suggestion. [ 5 ] ) The TT continues to be at the heart of AI and discussions of its foundations, as confirmed by the appearance of (Moor 2003). In fact, the TT continues to be used to define the field, as in Nilsson’s (1998) position, expressed in his textbook for the field, that AI simply is the field devoted to building an artifact able to negotiate this test. Energy supplied by the dream of engineering a computer that can pass TT, or by controversy surrounding claims that it has already been passed, is if anything stronger than ever, and the reader has only to do an internet search via the string

turing test passed

to find up-to-the-minute attempts at reaching this dream, and attempts (sometimes made by philosophers) to debunk claims that some such attempt has succeeded.

Returning to the issue of the historical record, even if one bolsters the claim that AI started at the 1956 conference by adding the proviso that ‘artificial intelligence’ refers to a nuts-and-bolts engineering pursuit (in which case Turing’s philosophical discussion, despite calls for a child machine, wouldn’t exactly count as AI per se), one must confront the fact that Turing, and indeed many predecessors, did attempt to build intelligent artifacts. In Turing’s case, such building was surprisingly well-understood before the advent of programmable computers: Turing wrote a program for playing chess before there were computers to run such programs on, by slavishly following the code himself. He did this well before 1950, and long before Newell (1973) gave thought in print to the possibility of a sustained, serious attempt at building a good chess-playing computer. [ 6 ]

From the perspective of philosophy, which views the systematic investigation of mechanical intelligence as meaningful and productive separate from the specific logicist formalisms (e.g., first-order logic) and problems (e.g., the Entscheidungsproblem ) that gave birth to computer science, neither the 1956 conference, nor Turing’s Mind paper, come close to marking the start of AI. This is easy enough to see. For example, Descartes proposed TT (not the TT by name, of course) long before Turing was born. [ 7 ] Here’s the relevant passage:

If there were machines which bore a resemblance to our body and imitated our actions as far as it was morally possible to do so, we should always have two very certain tests by which to recognise that, for all that, they were not real men. The first is, that they could never use speech or other signs as we do when placing our thoughts on record for the benefit of others. For we can easily understand a machine’s being constituted so that it can utter words, and even emit some responses to action on it of a corporeal kind, which brings about a change in its organs; for instance, if it is touched in a particular part it may ask what we wish to say to it; if in another part it may exclaim that it is being hurt, and so on. But it never happens that it arranges its speech in various ways, in order to reply appropriately to everything that may be said in its presence, as even the lowest type of man can do. And the second difference is, that although machines can perform certain things as well as or perhaps better than any of us can do, they infallibly fall short in others, by which means we may discover that they did not act from knowledge, but only for the disposition of their organs. For while reason is a universal instrument which can serve for all contingencies, these organs have need of some special adaptation for every particular action. From this it follows that it is morally impossible that there should be sufficient diversity in any machine to allow it to act in all the events of life in the same way as our reason causes us to act. (Descartes 1637, p. 116)

At the moment, Descartes is certainly carrying the day. [ 8 ] Turing predicted that his test would be passed by 2000, but the fireworks across the globe at the start of the new millennium have long since died down, and the most articulate of computers still can’t meaningfully debate a sharp toddler. Moreover, while in certain focussed areas machines out-perform minds (IBM’s famous Deep Blue prevailed in chess over Gary Kasparov, e.g.; and more recently, AI systems have prevailed in other games, e.g. Jeopardy! and Go, about which more will momentarily be said), minds have a (Cartesian) capacity for cultivating their expertise in virtually any sphere. (If it were announced to Deep Blue, or any current successor, that chess was no longer to be the game of choice, but rather a heretofore unplayed variant of chess, the machine would be trounced by human children of average intelligence having no chess expertise.) AI simply hasn’t managed to create general intelligence; it hasn’t even managed to produce an artifact indicating that eventually it will create such a thing.

But what about IBM Watson’s famous nail-biting victory in the Jeopardy! game-show contest? [ 9 ] That certainly seems to be a machine triumph over humans on their “home field,” since Jeopardy! delivers a human-level linguistic challenge ranging across many domains. Indeed, among many AI cognoscenti, Watson’s success is considered to be much more impressive than Deep Blue’s, for numerous reasons. One reason is that while chess is generally considered to be well-understood from the formal-computational perspective (after all, it’s well-known that there exists a perfect strategy for playing chess), in open-domain question-answering (QA), as in any significant natural-language processing task, there is no consensus as to what problem, formally speaking, one is trying to solve. Briefly, question-answering (QA) is what the reader would think it is: one asks a question of a machine, and gets an answer, where the answer has to be produced via some “significant” computational process. (See Strzalkowski & Harabagiu (2006) for an overview of what QA, historically, has been as a field.) A bit more precisely, there is no agreement as to what underlying function, formally speaking, question-answering capability computes. This lack of agreement stems quite naturally from the fact that there is of course no consensus as to what natural languages are , formally speaking. [ 10 ] Despite this murkiness, and in the face of an almost universal belief that open-domain question-answering would remain unsolved for a decade or more, Watson decisively beat the two top human Jeopardy! champions on the planet. During the contest, Watson had to answer questions that required not only command of simple factoids ( Question 1 ), but also of some amount of rudimentary reasoning (in the form of temporal reasoning) and commonsense ( Question 2 ):

Question 1 : The only two consecutive U.S. presidents with the same first name.

Question 2 : In May 1898, Portugal celebrated the 400th anniversary of this explorer’s arrival in India.

While Watson is demonstrably better than humans in Jeopardy! -style quizzing (a new human Jeopardy! master could arrive on the scene, but as for chess, AI now assumes that a second round of IBM-level investment would vanquish the new human opponent), this approach does not work for the kind of NLP challenge that Descartes described; that is, Watson can’t converse on the fly. After all, some questions don’t hinge on sophisticated information retrieval and machine learning over pre-existing data, but rather on intricate reasoning right on the spot. Such questions may for instance involve anaphora resolution, which require even deeper degrees of commonsensical understanding of time, space, history, folk psychology, and so on. Levesque (2013) has catalogued some alarmingly simple questions which fall in this category. (Marcus, 2013, gives an account of Levesque’s challenges that is accessible to a wider audience.) The other class of question-answering tasks on which Watson fails can be characterized as dynamic question-answering. These are questions for which answers may not be recorded in textual form anywhere at the time of questioning, or for which answers are dependent on factors that change with time. Two questions that fall in this category are given below (Govindarajulu et al. 2013):

Question 3 : If I have 4 foos and 5 bars, and if foos are not the same as bars, how many foos will I have if I get 3 bazes which just happen to be foos?

Question 4 : What was IBM’s Sharpe ratio in the last 60 days of trading?

Closely following Watson’s victory, in March 2016, Google DeepMind’s AlphaGo defeated one of Go’s top-ranked players, Lee Seedol, in four out of five matches. This was considered a landmark achievement within AI, as it was widely believed in the AI community that computer victory in Go was at least a few decades away, partly due to the enormous number of valid sequences of moves in Go compared to that in Chess. [ 11 ] While this is a remarkable achievement, it should be noted that, despite breathless coverage in the popular press, [ 12 ] AlphaGo, while indisputably a great Go player, is just that. For example, neither AlphaGo nor Watson can understand the rules of Go written in plain-and-simple English and produce a computer program that can play the game. It’s interesting that there is one endeavor in AI that tackles a narrow version of this very problem: In general game playing , a machine is given a description of a brand new game just before it has to play the game (Genesereth et al. 2005). However, the description in question is expressed in a formal language, and the machine has to manage to play the game from this description. Note that this is still far from understanding even a simple description of a game in English well enough to play it.

But what if we consider the history of AI not from the perspective of philosophy, but rather from the perspective of the field with which, today, it is most closely connected? The reference here is to computer science. From this perspective, does AI run back to well before Turing? Interestingly enough, the results are the same: we find that AI runs deep into the past, and has always had philosophy in its veins. This is true for the simple reason that computer science grew out of logic and probability theory, [ 13 ] which in turn grew out of (and is still intertwined with) philosophy. Computer science, today, is shot through and through with logic; the two fields cannot be separated. This phenomenon has become an object of study unto itself (Halpern et al. 2001). The situation is no different when we are talking not about traditional logic, but rather about probabilistic formalisms, also a significant component of modern-day AI: These formalisms also grew out of philosophy, as nicely chronicled, in part, by Glymour (1992). For example, in the one mind of Pascal was born a method of rigorously calculating probabilities, conditional probability (which plays a particularly large role in AI, currently), and such fertile philosophico-probabilistic arguments as Pascal’s wager , according to which it is irrational not to become a Christian.

That modern-day AI has its roots in philosophy, and in fact that these historical roots are temporally deeper than even Descartes’ distant day, can be seen by looking to the clever, revealing cover of the second edition (the third edition is the current one) of the comprehensive textbook Artificial Intelligence: A Modern Approach (known in the AI community as simply AIMA2e for Russell & Norvig, 2002).

Cover of AIMA2e (Russell & Norvig 2002)

What you see there is an eclectic collection of memorabilia that might be on and around the desk of some imaginary AI researcher. For example, if you look carefully, you will specifically see: a picture of Turing, a view of Big Ben through a window (perhaps R&N are aware of the fact that Turing famously held at one point that a physical machine with the power of a universal Turing machine is physically impossible: he quipped that it would have to be the size of Big Ben), a planning algorithm described in Aristotle’s De Motu Animalium , Frege’s fascinating notation for first-order logic , a glimpse of Lewis Carroll’s (1958) pictorial representation of syllogistic reasoning, Ramon Lull’s concept-generating wheel from his 13 th -century Ars Magna , and a number of other pregnant items (including, in a clever, recursive, and bordering-on-self-congratulatory touch, a copy of AIMA itself). Though there is insufficient space here to make all the historical connections, we can safely infer from the appearance of these items (and here we of course refer to the ancient ones: Aristotle conceived of planning as information-processing over two-and-a-half millennia back; and in addition, as Glymour (1992) notes, Artistotle can also be credited with devising the first knowledge-bases and ontologies, two types of representation schemes that have long been central to AI) that AI is indeed very, very old. Even those who insist that AI is at least in part an artifact-building enterprise must concede that, in light of these objects, AI is ancient, for it isn’t just theorizing from the perspective that intelligence is at bottom computational that runs back into the remote past of human history: Lull’s wheel, for example, marks an attempt to capture intelligence not only in computation, but in a physical artifact that embodies that computation. [ 14 ]

AIMA has now reached its the third edition, and those interested in the history of AI, and for that matter the history of philosophy of mind, will not be disappointed by examination of the cover of the third installment (the cover of the second edition is almost exactly like the first edition). (All the elements of the cover, separately listed and annotated, can be found online .) One significant addition to the cover of the third edition is a drawing of Thomas Bayes; his appearance reflects the recent rise in the popularity of probabilistic techniques in AI, which we discuss later.

One final point about the history of AI seems worth making.

It is generally assumed that the birth of modern-day AI in the 1950s came in large part because of and through the advent of the modern high-speed digital computer. This assumption accords with common-sense. After all, AI (and, for that matter, to some degree its cousin, cognitive science, particularly computational cognitive modeling, the sub-field of cognitive science devoted to producing computational simulations of human cognition) is aimed at implementing intelligence in a computer, and it stands to reason that such a goal would be inseparably linked with the advent of such devices. However, this is only part of the story: the part that reaches back but to Turing and others (e.g., von Neuman) responsible for the first electronic computers. The other part is that, as already mentioned, AI has a particularly strong tie, historically speaking, to reasoning (logic-based and, in the need to deal with uncertainty, inductive/probabilistic reasoning). In this story, nicely told by Glymour (1992), a search for an answer to the question “What is a proof?” eventually led to an answer based on Frege’s version of first-order logic (FOL): a (finitary) mathematical proof consists in a series of step-by-step inferences from one formula of first-order logic to the next. The obvious extension of this answer (and it isn’t a complete answer, given that lots of classical mathematics, despite conventional wisdom, clearly can’t be expressed in FOL; even the Peano Axioms, to be expressed as a finite set of formulae, require S OL) is to say that not only mathematical thinking, but thinking, period, can be expressed in FOL. (This extension was entertained by many logicians long before the start of information-processing psychology and cognitive science – a fact some cognitive psychologists and cognitive scientists often seem to forget.) Today, logic-based AI is only part of AI, but the point is that this part still lives (with help from logics much more powerful, but much more complicated, than FOL), and it can be traced all the way back to Aristotle’s theory of the syllogism. [ 15 ] In the case of uncertain reasoning, the question isn’t “What is a proof?”, but rather questions such as “What is it rational to believe, in light of certain observations and probabilities?” This is a question posed and tackled long before the arrival of digital computers.

So far we have been proceeding as if we have a firm and precise grasp of the nature of AI. But what exactly is AI? Philosophers arguably know better than anyone that precisely defining a particular discipline to the satisfaction of all relevant parties (including those working in the discipline itself) can be acutely challenging. Philosophers of science certainly have proposed credible accounts of what constitutes at least the general shape and texture of a given field of science and/or engineering, but what exactly is the agreed-upon definition of physics? What about biology? What, for that matter, is philosophy, exactly? These are remarkably difficult, maybe even eternally unanswerable, questions, especially if the target is a consensus definition. Perhaps the most prudent course we can manage here under obvious space constraints is to present in encapsulated form some proposed definitions of AI. We do include a glimpse of recent attempts to define AI in detailed, rigorous fashion (and we suspect that such attempts will be of interest to philosophers of science, and those interested in this sub-area of philosophy).

Russell and Norvig (1995, 2002, 2009), in their aforementioned AIMA text, provide a set of possible answers to the “What is AI?” question that has considerable currency in the field itself. These answers all assume that AI should be defined in terms of its goals: a candidate definition thus has the form “AI is the field that aims at building …” The answers all fall under a quartet of types placed along two dimensions. One dimension is whether the goal is to match human performance, or, instead, ideal rationality. The other dimension is whether the goal is to build systems that reason/think, or rather systems that act. The situation is summed up in this table:

Four Possible Goals for AI According to AIMA

Please note that this quartet of possibilities does reflect (at least a significant portion of) the relevant literature. For example, philosopher John Haugeland (1985) falls into the Human/Reasoning quadrant when he says that AI is “The exciting new effort to make computers think … machines with minds , in the full and literal sense.” (By far, this is the quadrant that most popular narratives affirm and explore. The recent Westworld TV series is a powerful case in point.) Luger and Stubblefield (1993) seem to fall into the Ideal/Act quadrant when they write: “The branch of computer science that is concerned with the automation of intelligent behavior.” The Human/Act position is occupied most prominently by Turing, whose test is passed only by those systems able to act sufficiently like a human. The “thinking rationally” position is defended (e.g.) by Winston (1992). While it might not be entirely uncontroversial to assert that the four bins given here are exhaustive, such an assertion appears to be quite plausible, even when the literature up to the present moment is canvassed.

It’s important to know that the contrast between the focus on systems that think/reason versus systems that act, while found, as we have seen, at the heart of the AIMA texts, and at the heart of AI itself, should not be interpreted as implying that AI researchers view their work as falling all and only within one of these two compartments. Researchers who focus more or less exclusively on knowledge representation and reasoning, are also quite prepared to acknowledge that they are working on (what they take to be) a central component or capability within any one of a family of larger systems spanning the reason/act distinction. The clearest case may come from the work on planning – an AI area traditionally making central use of representation and reasoning. For good or ill, much of this research is done in abstraction (in vitro, as opposed to in vivo), but the researchers involved certainly intend or at least hope that the results of their work can be embedded into systems that actually do things, such as, for example, execute the plans.

What about Russell and Norvig themselves? What is their answer to the What is AI? question? They are firmly in the “acting rationally” camp. In fact, it’s safe to say both that they are the chief proponents of this answer, and that they have been remarkably successful evangelists. Their extremely influential AIMA series can be viewed as a book-length defense and specification of the Ideal/Act category. We will look a bit later at how Russell and Norvig lay out all of AI in terms of intelligent agents , which are systems that act in accordance with various ideal standards for rationality. But first let’s look a bit closer at the view of intelligence underlying the AIMA text. We can do so by turning to Russell (1997). Here Russell recasts the “What is AI?” question as the question “What is intelligence?” (presumably under the assumption that we have a good grasp of what an artifact is), and then he identifies intelligence with rationality . More specifically, Russell sees AI as the field devoted to building intelligent agents , which are functions taking as input tuples of percepts from the external environment, and producing behavior (actions) on the basis of these percepts. Russell’s overall picture is this one:

The Basic Picture Underlying Russell’s Account of Intelligence/Rationality

Let’s unpack this diagram a bit, and take a look, first, at the account of perfect rationality that can be derived from it. The behavior of the agent in the environment \(E\) (from a class \(\bE\) of environments) produces a sequence of states or snapshots of that environment. A performance measure \(U\) evaluates this sequence; notice the box labeled “Performance Measure” in the above figure. We let \(V(f,\bE,U)\) denote the expected utility according to \(U\) of the agent function \(f\) operating on \(\bE\). [ 16 ] Now we identify a perfectly rational agent with the agent function:

According to the above equation, a perfectly rational agent can be taken to be the function \(f_{opt}\) which produces the maximum expected utility in the environment under consideration. Of course, as Russell points out, it’s usually not possible to actually build perfectly rational agents. For example, though it’s easy enough to specify an algorithm for playing invincible chess, it’s not feasible to implement this algorithm. What traditionally happens in AI is that programs that are – to use Russell’s apt terminology – calculatively rational are constructed instead: these are programs that, if executed infinitely fast , would result in perfectly rational behavior. In the case of chess, this would mean that we strive to write a program that runs an algorithm capable, in principle, of finding a flawless move, but we add features that truncate the search for this move in order to play within intervals of digestible duration.

Russell himself champions a new brand of intelligence/rationality for AI; he calls this brand bounded optimality . To understand Russell’s view, first we follow him in introducing a distinction: We say that agents have two components: a program, and a machine upon which the program runs. We write \(Agent(P, M)\) to denote the agent function implemented by program \(P\) running on machine \(M\). Now, let \(\mathcal{P}(M)\) denote the set of all programs \(P\) that can run on machine \(M\). The bounded optimal program \(P_{\opt,M}\) then is:

You can understand this equation in terms of any of the mathematical idealizations for standard computation. For example, machines can be identified with Turing machines minus instructions (i.e., TMs are here viewed architecturally only: as having tapes divided into squares upon which symbols can be written, read/write heads capable of moving up and down the tape to write and erase, and control units which are in one of a finite number of states at any time), and programs can be identified with instructions in the Turing-machine model (telling the machine to write and erase symbols, depending upon what state the machine is in). So, if you are told that you must “program” within the constraints of a 22-state Turing machine, you could search for the “best” program given those constraints. In other words, you could strive to find the optimal program within the bounds of the 22-state architecture. Russell’s (1997) view is thus that AI is the field devoted to creating optimal programs for intelligent agents, under time and space constraints on the machines implementing these programs. [ 17 ]

The reader must have noticed that in the equation for \(P_{\opt,M}\) we have not elaborated on \(\bE\) and \(U\) and how equation \eqref{eq1} might be used to construct an agent if the class of environments \(\bE\) is quite general, or if the true environment \(E\) is simply unknown. Depending on the task for which one is constructing an artificial agent, \(E\) and \(U\) would vary. The mathematical form of the environment \(E\) and the utility function \(U\) would vary wildly from, say, chess to Jeopardy! . Of course, if we were to design a globally intelligent agent, and not just a chess-playing agent, we could get away with having just one pair of \(E\) and \(U\). What would \(E\) look like if we were building a generally intelligent agent and not just an agent that is good at a single task? \(E\) would be a model of not just a single game or a task, but the entire physical-social-virtual universe consisting of many games, tasks, situations, problems, etc. This project is (at least currently) hopelessly difficult as, obviously, we are nowhere near to having such a comprehensive theory-of-everything model. For further discussion of a theoretical architecture put forward for this problem, see the Supplement on the AIXI architecture .

It should be mentioned that there is a different, much more straightforward answer to the “What is AI?” question. This answer, which goes back to the days of the original Dartmouth conference, was expressed by, among others, Newell (1973), one of the grandfathers of modern-day AI (recall that he attended the 1956 conference); it is:

AI is the field devoted to building artifacts that are intelligent, where ‘intelligent’ is operationalized through intelligence tests (such as the Wechsler Adult Intelligence Scale), and other tests of mental ability (including, e.g., tests of mechanical ability, creativity, and so on).

The above definition can be seen as fully specifying a concrete version of Russell and Norvig’s four possible goals. Though few are aware of this now, this answer was taken quite seriously for a while, and in fact underlied one of the most famous programs in the history of AI: the ANALOGY program of Evans (1968), which solved geometric analogy problems of a type seen in many intelligence tests. An attempt to rigorously define this forgotten form of AI (as what they dub Psychometric AI ), and to resurrect it from the days of Newell and Evans, is provided by Bringsjord and Schimanski (2003) [see also e.g. (Bringsjord 2011)]. A sizable private investment has been made in the ongoing attempt, now known as Project Aristo , to build a “digital Aristotle”, in the form of a machine able to excel on standardized tests such at the AP exams tackled by US high school students (Friedland et al. 2004). (Vibrant work in this direction continues today at the Allen Institute for Artificial Intelligence .) [ 18 ] In addition, researchers at Northwestern have forged a connection between AI and tests of mechanical ability (Klenk et al. 2005).

In the end, as is the case with any discipline, to really know precisely what that discipline is requires you to, at least to some degree, dive in and do, or at least dive in and read. Two decades ago such a dive was quite manageable. Today, because the content that has come to constitute AI has mushroomed, the dive (or at least the swim after it) is a bit more demanding.

3. Approaches to AI

There are a number of ways of “carving up” AI. By far the most prudent and productive way to summarize the field is to turn yet again to the AIMA text given its comprehensive overview of the field.

As Russell and Norvig (2009) tell us in the Preface of AIMA :

The main unifying theme is the idea of an intelligent agent. We define AI as the study of agents that receive percepts from the environment and perform actions. Each such agent implements a function that maps percept sequences to actions, and we cover different ways to represent these functions… (Russell & Norvig 2009, vii)

The basic picture is thus summed up in this figure:

Impressionistic Overview of an Intelligent Agent

The content of AIMA derives, essentially, from fleshing out this picture; that is, the above figure corresponds to the different ways of representing the overall function that intelligent agents implement. And there is a progression from the least powerful agents up to the more powerful ones. The following figure gives a high-level view of a simple kind of agent discussed early in the book. (Though simple, this sort of agent corresponds to the architecture of representation-free agents designed and implemented by Rodney Brooks, 1991.)

A Simple Reflex Agent

As the book progresses, agents get increasingly sophisticated, and the implementation of the function they represent thus draws from more and more of what AI can currently muster. The following figure gives an overview of an agent that is a bit smarter than the simple reflex agent. This smarter agent has the ability to internally model the outside world, and is therefore not simply at the mercy of what can at the moment be directly sensed.

A More Sophisticated Reflex Agent

There are seven parts to AIMA . As the reader passes through these parts, she is introduced to agents that take on the powers discussed in each part. Part I is an introduction to the agent-based view. Part II is concerned with giving an intelligent agent the capacity to think ahead a few steps in clearly defined environments. Examples here include agents able to successfully play games of perfect information, such as chess. Part III deals with agents that have declarative knowledge and can reason in ways that will be quite familiar to most philosophers and logicians (e.g., knowledge-based agents deduce what actions should be taken to secure their goals). Part IV of the book outfits agents with the power to handle uncertainty by reasoning in probabilistic fashion. [ 19 ] In Part V, agents are given a capacity to learn. The following figure shows the overall structure of a learning agent.

A Learning Agent

The final set of powers agents are given allow them to communicate. These powers are covered in Part VI.

Philosophers who patiently travel the entire progression of increasingly smart agents will no doubt ask, when reaching the end of Part VII, if anything is missing. Are we given enough, in general, to build an artificial person, or is there enough only to build a mere animal? This question is implicit in the following from Charniak and McDermott (1985):

The ultimate goal of AI (which we are very far from achieving) is to build a person, or, more humbly, an animal. (Charniak & McDermott 1985, 7)

To their credit, Russell & Norvig, in AIMA ’s Chapter 27, “AI: Present and Future,” consider this question, at least to some degree. [ ] They do so by considering some challenges to AI that have hitherto not been met. One of these challenges is described by R&N as follows:

[M]achine learning has made very little progress on the important problem of constructing new representations at levels of abstraction higher than the input vocabulary. In computer vision, for example, learning complex concepts such as Classroom and Cafeteria would be made unnecessarily difficult if the agent were forced to work from pixels as the input representation; instead, the agent needs to be able to form intermediate concepts first, such as Desk and Tray, without explicit human supervision. Similar concepts apply to learning behavior: HavingACupOfTea is a very important high-level step in many plans, but how does it get into an action library that initially contains much simpler actions such as RaiseArm and Swallow? Perhaps this will incorporate deep belief networks – Bayesian networks that have multiple layers of hidden variables, as in the work of Hinton et al. (2006), Hawkins and Blakeslee (2004), and Bengio and LeCun (2007). … Unless we understand such issues, we are faced with the daunting task of constructing large commonsense knowledge bases by hand, and approach that has not fared well to date. (Russell & Norvig 2009, Ch. 27.1)

While there has seen some advances in addressing this challenge (in the form of deep learning or representation learning ), this specific challenge is actually merely a foothill before a range of dizzyingly high mountains that AI must eventually somehow manage to climb. One of those mountains, put simply, is reading . [ 21 ] Despite the fact that, as noted, Part V of AIMA is devoted to machine learning, AI, as it stands, offers next to nothing in the way of a mechanization of learning by reading. Yet when you think about it, reading is probably the dominant way you learn at this stage in your life. Consider what you’re doing at this very moment. It’s a good bet that you are reading this sentence because, earlier, you set yourself the goal of learning about the field of AI. Yet the formal models of learning provided in AIMA ’s Part IV (which are all and only the models at play in AI) cannot be applied to learning by reading. [ 22 ] These models all start with a function-based view of learning. According to this view, to learn is almost invariably to produce an underlying function \(\ff\) on the basis of a restricted set of pairs

For example, consider receiving inputs consisting of 1, 2, 3, 4, and 5, and corresponding range values of 1, 4, 9, 16, and 25; the goal is to “learn” the underlying mapping from natural numbers to natural numbers. In this case, assume that the underlying function is \(n^2\), and that you do “learn” it. While this narrow model of learning can be productively applied to a number of processes, the process of reading isn’t one of them. Learning by reading cannot (at least for the foreseeable future) be modeled as divining a function that produces argument-value pairs. Instead, your reading about AI can pay dividends only if your knowledge has increased in the right way, and if that knowledge leaves you poised to be able to produce behavior taken to confirm sufficient mastery of the subject area in question. This behavior can range from correctly answering and justifying test questions regarding AI, to producing a robust, compelling presentation or paper that signals your achievement.

Two points deserve to be made about machine reading. First, it may not be clear to all readers that reading is an ability that is central to intelligence. The centrality derives from the fact that intelligence requires vast knowledge. We have no other means of getting systematic knowledge into a system than to get it in from text, whether text on the web, text in libraries, newspapers, and so on. You might even say that the big problem with AI has been that machines really don’t know much compared to humans. That can only be because of the fact that humans read (or hear: illiterate people can listen to text being uttered and learn that way). Either machines gain knowledge by humans manually encoding and inserting knowledge, or by reading and listening. These are brute facts. (We leave aside supernatural techniques, of course. Oddly enough, Turing didn’t: he seemed to think ESP should be discussed in connection with the powers of minds and machines. See Turing, 1950.) [ 23 ]

Now for the second point. Humans able to read have invariably also learned a language, and learning languages has been modeled in conformity to the function-based approach adumbrated just above (Osherson et al. 1986). However, this doesn’t entail that an artificial agent able to read, at least to a significant degree, must have really and truly learned a natural language. AI is first and foremost concerned with engineering computational artifacts that measure up to some test (where, yes, sometimes that test is from the human sphere), not with whether these artifacts process information in ways that match those present in the human case. It may or may not be necessary, when engineering a machine that can read, to imbue that machine with human-level linguistic competence. The issue is empirical, and as time unfolds, and the engineering is pursued, we shall no doubt see the issue settled.

Two additional high mountains facing AI are subjective consciousness and creativity, yet it would seem that these great challenges are ones the field apparently hasn’t even come to grips with. Mental phenomena of paramount importance to many philosophers of mind and neuroscience are simply missing from AIMA . For example, consciousness is only mentioned in passing in AIMA , but subjective consciousness is the most important thing in our lives – indeed we only desire to go on living because we wish to go on enjoying subjective states of certain types. Moreover, if human minds are the product of evolution, then presumably phenomenal consciousness has great survival value, and would be of tremendous help to a robot intended to have at least the behavioral repertoire of the first creatures with brains that match our own (hunter-gatherers; see Pinker 1997). Of course, subjective consciousness is largely missing from the sister fields of cognitive psychology and computational cognitive modeling as well. We discuss some of these challenges in the Philosophy of Artificial Intelligence section below. For a list of similar challenges to cognitive science, see the relevant section of the entry on cognitive science . [ 24 ]

To some readers, it might seem in the very least tendentious to point to subjective consciousness as a major challenge to AI that it has yet to address. These readers might be of the view that pointing to this problem is to look at AI through a distinctively philosophical prism, and indeed a controversial philosophical standpoint.

But as its literature makes clear, AI measures itself by looking to animals and humans and picking out in them remarkable mental powers, and by then seeing if these powers can be mechanized. Arguably the power most important to humans (the capacity to experience) is nowhere to be found on the target list of most AI researchers. There may be a good reason for this (no formalism is at hand, perhaps), but there is no denying the state of affairs in question obtains, and that, in light of how AI measures itself, that it’s worrisome.

As to creativity, it’s quite remarkable that the power we most praise in human minds is nowhere to be found in AIMA . Just as in (Charniak & McDermott 1985) one cannot find ‘neural’ in the index, ‘creativity’ can’t be found in the index of AIMA . This is particularly odd because many AI researchers have in fact worked on creativity (especially those coming out of philosophy; e.g., Boden 1994, Bringsjord & Ferrucci 2000).

Although the focus has been on AIMA , any of its counterparts could have been used. As an example, consider Artificial Intelligence: A New Synthesis , by Nils Nilsson. As in the case of AIMA , everything here revolves around a gradual progression from the simplest of agents (in Nilsson’s case, reactive agents ), to ones having more and more of those powers that distinguish persons. Energetic readers can verify that there is a striking parallel between the main sections of Nilsson’s book and AIMA . In addition, Nilsson, like Russell and Norvig, ignores phenomenal consciousness, reading, and creativity. None of the three are even mentioned. Likewise, a recent comprehensive AI textbook by Luger (2008) follows the same pattern.

A final point to wrap up this section. It seems quite plausible to hold that there is a certain inevitability to the structure of an AI textbook, and the apparent reason is perhaps rather interesting. In personal conversation, Jim Hendler, a well-known AI researcher who is one of the main innovators behind Semantic Web (Berners-Lee, Hendler, Lassila 2001), an under-development “AI-ready” version of the World Wide Web, has said that this inevitability can be rather easily displayed when teaching Introduction to AI; here’s how. Begin by asking students what they think AI is. Invariably, many students will volunteer that AI is the field devoted to building artificial creatures that are intelligent. Next, ask for examples of intelligent creatures. Students always respond by giving examples across a continuum: simple multi-cellular organisms, insects, rodents, lower mammals, higher mammals (culminating in the great apes), and finally human persons. When students are asked to describe the differences between the creatures they have cited, they end up essentially describing the progression from simple agents to ones having our (e.g.) communicative powers. This progression gives the skeleton of every comprehensive AI textbook. Why does this happen? The answer seems clear: it happens because we can’t resist conceiving of AI in terms of the powers of extant creatures with which we are familiar. At least at present, persons, and the creatures who enjoy only bits and pieces of personhood, are – to repeat – the measure of AI. [ 25 ]

Reasoning based on classical deductive logic is monotonic; that is, if \(\Phi\vdash\phi\), then for all \(\psi\), \(\Phi\cup \{\psi\}\vdash\phi\). Commonsense reasoning is not monotonic. While you may currently believe on the basis of reasoning that your house is still standing, if while at work you see on your computer screen that a vast tornado is moving through the location of your house, you will drop this belief. The addition of new information causes previous inferences to fail. In the simpler example that has become an AI staple, if I tell you that Tweety is a bird, you will infer that Tweety can fly, but if I then inform you that Tweety is a penguin, the inference evaporates, as well it should. Nonmonotonic (or defeasible) logic includes formalisms designed to capture the mechanisms underlying these kinds of examples. See the separate entry on logic and artificial intelligence , which is focused on nonmonotonic reasoning, and reasoning about time and change. It also provides a history of the early days of logic-based AI, making clear the contributions of those who founded the tradition (e.g., John McCarthy and Pat Hayes; see their seminal 1969 paper).

The formalisms and techniques of logic-based AI have reached a level of impressive maturity – so much so that in various academic and corporate laboratories, implementations of these formalisms and techniques can be used to engineer robust, real-world software. It is strongly recommend that readers who have an interest to learn where AI stands in these areas consult (Mueller 2006), which provides, in one volume, integrated coverage of nonmonotonic reasoning (in the form, specifically, of circumscription), and reasoning about time and change in the situation and event calculi. (The former calculus is also introduced by Thomason. In the second, timepoints are included, among other things.) The other nice thing about (Mueller 2006) is that the logic used is multi-sorted first-order logic (MSL), which has unificatory power that will be known to and appreciated by many technical philosophers and logicians (Manzano 1996).

We now turn to three further topics of importance in AI. They are:

• The overarching scheme of logicist AI, in the context of the attempt to build intelligent artificial agents.
• Common Logic and the intensifying quest for interoperability.
• A technique that can be called encoding down , which can allow machines to reason efficiently over knowledge that, were it not encoded down, would, when reasoned over, lead to paralyzing inefficiency.

This trio is covered in order, beginning with the first.

Detailed accounts of logicist AI that fall under the agent-based scheme can be found in (Lenat 1983, Lenat & Guha 1990, Nilsson 1991, Bringsjord & Ferrucci 1998). [ 26 ] . The core idea is that an intelligent agent receives percepts from the external world in the form of formulae in some logical system (e.g., first-order logic), and infers, on the basis of these percepts and its knowledge base, what actions should be performed to secure the agent’s goals. (This is of course a barbaric simplification. Information from the external world is encoded in formulae, and transducers to accomplish this feat may be components of the agent.)

To clarify things a bit, we consider, briefly, the logicist view in connection with arbitrary logical systems \(\mathcal{L}_{X}\). [ 27 ] We obtain a particular logical system by setting \(X\) in the appropriate way. Some examples: If \(X=I\), then we have a system at the level of FOL [following the standard notation from model theory; see e.g. (Ebbinghaus et al. 1984)]. \(\mathcal{L}_{II}\) is second-order logic, and \(\mathcal{L}_{\omega_I\omega}\) is a “small system” of infinitary logic (countably infinite conjunctions and disjunctions are permitted). These logical systems are all extensional , but there are intensional ones as well. For example, we can have logical systems corresponding to those seen in standard propositional modal logic (Chellas 1980). One possibility, familiar to many philosophers, would be propositional KT45, or \(\mathcal{L}_{KT45}\). [ 28 ] In each case, the system in question includes a relevant alphabet from which well-formed formulae are constructed by way of a formal grammar, a reasoning (or proof) theory, a formal semantics, and at least some meta-theoretical results (soundness, completeness, etc.). Taking off from standard notation, we can thus say that a set of formulas in some particular logical system \(\mathcal{L}_X\), \(\Phi_{\mathcal{L}_X}\), can be used, in conjunction with some reasoning theory, to infer some particular formula \(\phi_{\mathcal{L}_X}\). (The reasoning may be deductive, inductive, abductive, and so on. Logicist AI isn’t in the least restricted to any particular mode of reasoning.) To say that such a situation holds, we write \[ \Phi_{\mathcal{L}_X} \vdash_{\mathcal{L}_X} \phi_{\mathcal{L}_X} \]

When the logical system referred to is clear from context, or when we don’t care about which logical system is involved, we can simply write \[ \Phi \vdash \phi \]

Each logical system, in its formal semantics, will include objects designed to represent ways the world pointed to by formulae in this system can be. Let these ways be denoted by \(W^i_{{\mathcal{L}_X}}\). When we aren’t concerned with which logical system is involved, we can simply write \(W^i\). To say that such a way models a formula \(\phi\) we write \[ W_i \models \phi \]

We extend this to a set of formulas in the natural way: \(W^i\models\Phi\) means that all the elements of \(\Phi\) are true on \(W^i\). Now, using the simple machinery we’ve established, we can describe, in broad strokes, the life of an intelligent agent that conforms to the logicist point of view. This life conforms to the basic cycle that undergirds intelligent agents in the AIMA sense.

To begin, we assume that the human designer, after studying the world, uses the language of a particular logical system to give to our agent an initial set of beliefs \(\Delta_0\) about what this world is like. In doing so, the designer works with a formal model of this world, \(W\), and ensures that \(W\models\Delta_0\). Following tradition, we refer to \(\Delta_0\) as the agent’s (starting) knowledge base . (This terminology, given that we are talking about the agent’s beliefs , is known to be peculiar, but it persists.) Next, the agent ADJUSTS its knowlege base to produce a new one, \(\Delta_1\). We say that adjustment is carried out by way of an operation \(\mathcal{A}\); so \(\mathcal{A}[\Delta_0]=\Delta_1\). How does the adjustment process, \(\mathcal{A}\), work? There are many possibilities. Unfortunately, many believe that the simplest possibility (viz., \(\mathcal{A}[\Delta_i]\) equals the set of all formulas that can be deduced in some elementary manner from \(\Delta_i\)) exhausts all the possibilities. The reality is that adjustment, as indicated above, can come by way of any mode of reasoning – induction, abduction, and yes, various forms of deduction corresponding to the logical system in play. For present purposes, it’s not important that we carefully enumerate all the options.

The cycle continues when the agent ACTS on the environment, in an attempt to secure its goals. Acting, of course, can cause changes to the environment. At this point, the agent SENSES the environment, and this new information \(\Gamma_1\) factors into the process of adjustment, so that \(\mathcal{A}[\Delta_1\cup\Gamma_1]=\Delta_2\). The cycle of SENSES \(\Rightarrow\) ADJUSTS \(\Rightarrow\) ACTS continues to produce the life \(\Delta_0,\Delta_1,\Delta_2,\Delta_3,\ldots,\) … of our agent.

It may strike you as preposterous that logicist AI be touted as an approach taken to replicate all of cognition. Reasoning over formulae in some logical system might be appropriate for computationally capturing high-level tasks like trying to solve a math problem (or devising an outline for an entry in the Stanford Encyclopedia of Philosophy), but how could such reasoning apply to tasks like those a hawk tackles when swooping down to capture scurrying prey? In the human sphere, the task successfully negotiated by athletes would seem to be in the same category. Surely, some will declare, an outfielder chasing down a fly ball doesn’t prove theorems to figure out how to pull off a diving catch to save the game! Two brutally reductionistic arguments can be given in support of this “logicist theory of everything” approach towards cognition. The first stems from the fact that a complete proof calculus for just first-order logic can simulate all of Turing-level computation (Chapter 11, Boolos et al. 2007). The second justification comes from the role logic plays in foundational theories of mathematics and mathematical reasoning. Not only are foundational theories of mathematics cast in logic (Potter 2004), but there have been successful projects resulting in machine verification of ordinary non-trivial theorems, e.g., in the Mizar project alone around 50,000 theorems have been verified (Naumowicz and Kornilowicz 2009). The argument goes that if any approach to AI can be cast mathematically, then it can be cast in a logicist form.

Needless to say, such a declaration has been carefully considered by logicists beyond the reductionistic argument given above. For example, Rosenschein and Kaelbling (1986) describe a method in which logic is used to specify finite state machines. These machines are used at “run time” for rapid, reactive processing. In this approach, though the finite state machines contain no logic in the traditional sense, they are produced by logic and inference. Real robot control via first-order theorem proving has been demonstrated by Amir and Maynard-Reid (1999, 2000, 2001). In fact, you can download version 2.0 of the software that makes this approach real for a Nomad 200 mobile robot in an office environment. Of course, negotiating an office environment is a far cry from the rapid adjustments an outfielder for the Yankees routinely puts on display, but certainly it’s an open question as to whether future machines will be able to mimic such feats through rapid reasoning. The question is open if for no other reason than that all must concede that the constant increase in reasoning speed of first-order theorem provers is breathtaking. (For up-to-date news on this increase, visit and monitor the TPTP site .) There is no known reason why the software engineering in question cannot continue to produce speed gains that would eventually allow an artificial creature to catch a fly ball by processing information in purely logicist fashion.

Now we come to the second topic related to logicist AI that warrants mention herein: common logic and the intensifying quest for interoperability between logic-based systems using different logics. Only a few brief comments are offered. [ 29 ] Readers wanting more can explore the links provided in the course of the summary.

One standardization is through what is known as Common Logic (CL), and variants thereof. (CL is published as an ISO standard – ISO is the International Standards Organization.) Philosophers interested in logic, and of course logicians, will find CL to be quite fascinating. From an historical perspective, the advent of CL is interesting in no small part because the person spearheading it is none other than Pat Hayes, the same Hayes who, as we have seen, worked with McCarthy to establish logicist AI in the 1960s. Though Hayes was not at the original 1956 Dartmouth conference, he certainly must be regarded as one of the founders of contemporary AI.) One of the interesting things about CL, at least as we see it, is that it signifies a trend toward the marriage of logics, and programming languages and environments. Another system that is a logic/programming hybrid is Athena , which can be used as a programming language, and is at the same time a form of MSL. Athena is based on formal systems known as denotational proof languages (Arkoudas 2000).

How is interoperability between two systems to be enabled by CL? Suppose one of these systems is based on logic \(L\), and the other on \(L'\). (To ease exposition, assume that both logics are first-order.) The idea is that a theory \(\Phi_L\), that is, a set of formulae in \(L\), can be translated into CL, producing \(\Phi_{CL}\), and then this theory can be translated into \(\Phi_L'\). CL thus becomes an inter lingua . Note that what counts as a well-formed formula in \(L\) can be different than what counts as one in \(L'\). The two logics might also have different proof theories. For example, inference in \(L\) might be based on resolution, while inference in \(L'\) is of the natural deduction variety. Finally, the symbol sets will be different. Despite these differences, courtesy of the translations, desired behavior can be produced across the translation. That, at any rate, is the hope. The technical challenges here are immense, but federal monies are increasingly available for attacks on the problem of interoperability.

Now for the third topic in this section: what can be called encoding down . The technique is easy to understand. Suppose that we have on hand a set \(\Phi\) of first-order axioms. As is well-known, the problem of deciding, for arbitrary formula \(\phi\), whether or not it’s deducible from \(\Phi\) is Turing-undecidable: there is no Turing machine or equivalent that can correctly return “Yes” or “No” in the general case. However, if the domain in question is finite, we can encode this problem down to the propositional calculus. An assertion that all things have \(F\) is of course equivalent to the assertion that \(Fa\), \(Fb\), \(Fc\), as long as the domain contains only these three objects. So here a first-order quantified formula becomes a conjunction in the propositional calculus. Determining whether such conjunctions are provable from axioms themselves expressed in the propositional calculus is Turing-decidable, and in addition, in certain clusters of cases, the check can be done very quickly in the propositional case; very quickly . Readers interested in encoding down to the propositional calculus should consult recent DARPA-sponsored work by Bart Selman . Please note that the target of encoding down doesn’t need to be the propositional calculus. Because it’s generally harder for machines to find proofs in an intensional logic than in straight first-order logic, it is often expedient to encode down the former to the latter. For example, propositional modal logic can be encoded in multi-sorted logic (a variant of FOL); see (Arkoudas & Bringsjord 2005). Prominent usage of such an encoding down can be found in a set of systems known as Description Logics , which are a set of logics less expressive than first-order logic but more expressive than propositional logic (Baader et al. 2003). Description logics are used to reason about ontologies in a given domain and have been successfully used, for example, in the biomedical domain (Smith et al. 2007).

It’s tempting to define non-logicist AI by negation: an approach to building intelligent agents that rejects the distinguishing features of logicist AI. Such a shortcut would imply that the agents engineered by non-logicist AI researchers and developers, whatever the virtues of such agents might be, cannot be said to know that \(\phi\); – for the simple reason that, by negation, the non-logicist paradigm would have not even a single declarative proposition that is a candidate for \(\phi\);. However, this isn’t a particularly enlightening way to define non-symbolic AI. A more productive approach is to say that non-symbolic AI is AI carried out on the basis of particular formalisms other than logical systems, and to then enumerate those formalisms. It will turn out, of course, that these formalisms fail to include knowledge in the normal sense. (In philosophy, as is well-known, the normal sense is one according to which if \(p\) is known, \(p\) is a declarative statement.)

From the standpoint of formalisms other than logical systems, non-logicist AI can be partitioned into symbolic but non-logicist approaches, and connectionist/neurocomputational approaches. (AI carried out on the basis of symbolic, declarative structures that, for readability and ease of use, are not treated directly by researchers as elements of formal logics, does not count. In this category fall traditional semantic networks, Schank’s (1972) conceptual dependency scheme, frame-based schemes, and other such schemes.) The former approaches, today, are probabilistic, and are based on the formalisms (Bayesian networks) covered below . The latter approaches are based, as we have noted, on formalisms that can be broadly termed “neurocomputational.” Given our space constraints, only one of the formalisms in this category is described here (and briefly at that): the aforementioned artificial neural networks . [ 30 ] . Though artificial neural networks, with an appropriate architecture, could be used for arbitrary computation, they are almost exclusively used for building learning systems.

Neural nets are composed of units or nodes designed to represent neurons, which are connected by links designed to represent dendrites, each of which has a numeric weight .

A “Neuron” Within an Artificial Neural Network (from AIMA3e)

It is usually assumed that some of the units work in symbiosis with the external environment; these units form the sets of input and output units. Each unit has a current activation level , which is its output, and can compute, based on its inputs and weights on those inputs, its activation level at the next moment in time. This computation is entirely local: a unit takes account of but its neighbors in the net. This local computation is calculated in two stages. First, the input function , \(in_i\), gives the weighted sum of the unit’s input values, that is, the sum of the input activations multiplied by their weights:

In the second stage, the activation function , \(g\), takes the input from the first stage as argument and generates the output, or activation level, \(a_i\):

One common (and confessedly elementary) choice for the activation function (which usually governs all units in a given net) is the step function, which usually has a threshold \(t\) that sees to it that a 1 is output when the input is greater than \(t\), and that 0 is output otherwise. This is supposed to be “brain-like” to some degree, given that 1 represents the firing of a pulse from a neuron through an axon, and 0 represents no firing. A simple three-layer neural net is shown in the following picture.

A Simple Three-Layer Artificial Neural Network (from AIMA3e)

As you might imagine, there are many different kinds of neural networks. The main distinction is between feed-forward and recurrent networks. In feed-forward networks like the one pictured immediately above, as their name suggests, links move information in one direction, and there are no cycles; recurrent networks allow for cycling back, and can become rather complicated. For a more detailed presentation, see the

Supplement on Neural Nets .

Neural networks were fundamentally plagued by the fact that while they are simple and have theoretically efficient learning algorithms, when they are multi-layered and thus sufficiently expressive to represent non-linear functions, they were very hard to train in practice. This changed in the mid 2000s with the advent of methods that exploit state-of-the-art hardware better (Rajat et al. 2009). The backpropagation method for training multi-layered neural networks can be translated into a sequence of repeated simple arithmetic operations on a large set of numbers. The general trend in computing hardware has favored algorithms that are able to do a large of number of simple operations that are not that dependent on each other, versus a small of number of complex and intricate operations.

Another key recent observation is that deep neural networks can be pre-trained first in an unsupervised phase where they are just fed data without any labels for the data. Each hidden layer is forced to represent the outputs of the layer below. The outcome of this training is a series of layers which represent the input domain with increasing levels of abstraction. For example, if we pre-train the network with images of faces, we would get a first layer which is good at detecting edges in images, a second layer which can combine edges to form facial features such as eyes, noses etc., a third layer which responds to groups of features, and so on (LeCun et al. 2015).

Perhaps the best technique for teaching students about neural networks in the context of other statistical learning formalisms and methods is to focus on a specific problem, preferably one that seems unnatural to tackle using logicist techniques. The task is then to seek to engineer a solution to the problem, using any and all techniques available . One nice problem is handwriting recognition (which also happens to have a rich philosophical dimension; see e.g. Hofstadter & McGraw 1995). For example, consider the problem of assigning, given as input a handwritten digit \(d\), the correct digit, 0 through 9. Because there is a database of 60,000 labeled digits available to researchers (from the National Institute of Science and Technology), this problem has evolved into a benchmark problem for comparing learning algorithms. It turns out that neural networks currently reign as the best approach to the problem according to a recent ranking by Benenson (2016).

Readers interested in AI (and computational cognitive science) pursued from an overtly brain-based orientation are encouraged to explore the work of Rick Granger (2004a, 2004b) and researchers in his Brain Engineering Laboratory and W. H. Neukom Institute for Computational Sciences . The contrast between the “dry”, logicist AI started at the original 1956 conference, and the approach taken here by Granger and associates (in which brain circuitry is directly modeled) is remarkable. For those interested in computational properties of neural networks, Hornik et al. (1989) address the general representation capability of neural networks independent of learning.

At this point the reader has been exposed to the chief formalisms in AI, and may wonder about heterogeneous approaches that bridge them. Is there such research and development in AI? Yes. From an engineering standpoint, such work makes irresistibly good sense. There is now an understanding that, in order to build applications that get the job done, one should choose from a toolbox that includes logicist, probabilistic/Bayesian, and neurocomputational techniques. Given that the original top-down logicist paradigm is alive and thriving (e.g., see Brachman & Levesque 2004, Mueller 2006), and that, as noted, a resurgence of Bayesian and neurocomputational approaches has placed these two paradigms on solid, fertile footing as well, AI now moves forward, armed with this fundamental triad, and it is a virtual certainty that applications (e.g., robots) will be engineered by drawing from elements of all three. Watson’s DeepQA architecture is one recent example of an engineering system that leverages multiple paradigms. For a detailed discussion, see the

Supplement on Watson’s DeepQA Architecture .

Google DeepMind’s AlphaGo is another example of a multi-paradigm system, although in a much narrower form than Watson. The central algorithmic problem in games such as Go or Chess is to search through a vast sequence of valid moves. For most non-trivial games, this is not feasible to do so exhaustively. The Monte Carlo tree search (MCTS) algorithm gets around this obstacle by searching through an enormous space of valid moves in a statistical fashion (Browne et al. 2012). While MCTS is the central algorithm in AlpaGo, there are two neural networks which help evaluate states in the game and help model how expert opponents play (Silver et al. 2016). It should be noted that MCTS is behind almost all the winning submissions in general game playing (Finnsson 2012).

What, though, about deep, theoretical integration of the main paradigms in AI? Such integration is at present only a possibility for the future, but readers are directed to the research of some striving for such integration. For example: Sun (1994, 2002) has been working to demonstrate that human cognition that is on its face symbolic in nature (e.g., professional philosophizing in the analytic tradition, which deals explicitly with arguments and definitions carefully symbolized) can arise from cognition that is neurocomputational in nature. Koller (1997) has investigated the marriage between probability theory and logic. And, in general, the very recent arrival of so-called human-level AI is being led by theorists seeking to genuinely integrate the three paradigms set out above (e.g., Cassimatis 2006).

Finally, we note that cognitive architectures such as Soar (Laird 2012) and PolyScheme (Cassimatis 2006) are another area where integration of different fields of AI can be found. For example, one such endeavor striving to build human-level AI is the Companions project (Forbus and Hinrichs 2006). Companions are long-lived systems that strive to be human-level AI systems that function as collaborators with humans. The Companions architecture tries to solve multiple AI problems such as reasoning and learning, interactivity, and longevity in one unifying system.

4. The Explosive Growth of AI

As we noted above, work on AI has mushroomed over the past couple of decades. Now that we have looked a bit at the content that composes AI, we take a quick look at the explosive growth of AI.

First, a point of clarification. The growth of which we speak is not a shallow sort correlated with amount of funding provided for a given sub-field of AI. That kind of thing happens all the time in all fields, and can be triggered by entirely political and financial changes designed to grow certain areas, and diminish others. Along the same line, the growth of which we speak is not correlated with the amount of industrial activity revolving around AI (or a sub-field thereof); for this sort of growth too can be driven by forces quite outside an expansion in the scientific breadth of AI. [ 31 ] Rather, we are speaking of an explosion of deep content : new material which someone intending to be conversant with the field needs to know. Relative to other fields, the size of the explosion may or may not be unprecedented. (Though it should perhaps be noted that an analogous increase in philosophy would be marked by the development of entirely new formalisms for reasoning, reflected in the fact that, say, longstanding philosophy textbooks like Copi’s (2004) Introduction to Logic are dramatically rewritten and enlarged to include these formalisms, rather than remaining anchored to essentially immutable core formalisms, with incremental refinement around the edges through the years.) But it certainly appears to be quite remarkable, and is worth taking note of here, if for no other reason than that AI’s near-future will revolve in significant part around whether or not the new content in question forms a foundation for new long-lived research and development that would not otherwise obtain. [ 32 ]

AI has also witnessed an explosion in its usage in various artifacts and applications. While we are nowhere near building a machine with capabilities of a human or one that acts rationally in all scenarios according to the Russell/Hutter definition above, algorithms that have their origins in AI research are now widely deployed for many tasks in a variety of domains.

A huge part of AI’s growth in applications has been made possible through invention of new algorithms in the subfield of machine learning . Machine learning is concerned with building systems that improve their performance on a task when given examples of ideal performance on the task, or improve their performance with repeated experience on the task. Algorithms from machine learning have been used in speech recognition systems, spam filters, online fraud-detection systems, product-recommendation systems, etc. The current state-of-the-art in machine learning can be divided into three areas (Murphy 2013, Alpaydin 2014):

• Supervised Learning : A form of learning in which a computer tries to learn a function \(\ff\) given examples, the training data \(T\), of its values at various points in its domain \[ T=\left\{\left\langle x_1, \ff(x_1)\right\rangle,\left\langle x_2, \ff(x_2)\right\rangle, \ldots, \left\langle x_n, \ff(x_n)\right\rangle\right\}. \] A sample task would be trying to label images of faces with a person’s name. The supervision in supervised learning comes in the form of the value of the function \(\ff(x)\) at various points \(x\) in some part of the domain of the function. This is usually given in the form of a fixed set of input and output pairs for the function. Let \(\hh\) be the “learned function.” The goal of supervised learning is have \({\hh}\) match as closely as possible the true function \({\ff}\) over the same domain. The error is usually defined in terms of an error function, for instance, \(error = \sum_{x\in T} \delta(\ff(x) - \hh(x))\), over the training data \(T\). Other forms of supervision and goals for learning are possible. For example, in active learning the learning algorithm can request the value of the function for arbitrary inputs. Supervised learning dominates the field of machine learning and has been used in almost all practical applications mentioned just above.
• Unsupervised Learning : Here the machine tries to find useful knowledge or information when given some raw data \(\left\{ x_1,x_2, \ldots, x_n \right\}\). There is no function associated with the input that has to be learned. The idea is that the machine helps uncover interesting patterns or information that could be hidden in the data. One use of unsupervised learning is data mining , where large volumes of data are searched for interesting information. PageRank , one of the earliest algorithms used by the Google search engine, can be considered to be an unsupervised learning system that ranks pages without any human supervision (Chapter 14.10, Hastie et al. 2009).
• Reinforcement Learning : Here a machine is set loose in an environment where it constantly acts and perceives (similar to the Russell/Hutter view above) and only occasionally receives feedback on its behavior in the form of rewards or punishments. The machine has to learn to behave rationally from this feedback. One use of reinforcement learning has been in building agents to play computer games. The objective here is to build agents that map sensory data from the game at every time instant to an action that would help win in the game or maximize a human player’s enjoyment of the game. In most games, we know how well we are playing only at the end of the game or only at infrequent intervals throughout the game (e.g., a chess game that we feel we are winning could quickly turn against us at the end). In supervised learning, the training data has ideal input-output pairs. This form of learning is not suitable for building agents that have to operate across a length of time and are judged not on one action but a series of actions and their effects on the environment. The field of Reinforcement Learning tries to tackle this problem through a variety of methods. Though a bit dated, Sutton and Barto (1998) provide a comprehensive introduction to the field.

In addition to being used in domains that are traditionally the ken of AI, machine-learning algorithms have also been used in all stages of the scientific process. For example, machine-learning techniques are now routinely applied to analyze large volumes of data generated from particle accelerators. CERN, for instance, generates a petabyte (\(10^{15}\) bytes) per second, and statistical algorithms that have their origins in AI are used to filter and analyze this data. Particle accelerators are used in fundamental experimental research in physics to probe the structure of our physical universe. They work by colliding larger particles together to create much finer particles. Not all such events are fruitful. Machine-learning methods have been used to select events which are then analyzed further (Whiteson & Whiteson 2009 and Baldi et al. 2014). More recently, researchers at CERN launched a machine learning competition to aid in the analysis of the Higgs Boson. The goal of this challenge was to develop algorithms that separate meaningful events from background noise given data from the Large Hadron Collider, a particle accelerator at CERN.

In the past few decades, there has been an explosion in data that does not have any explicit semantics attached to it. This data is generated by both humans and machines. Most of this data is not easily machine-processable; for example, images, text, video (as opposed to carefully curated data in a knowledge- or data-base). This has given rise to a huge industry that applies AI techniques to get usable information from such enormous data. This field of applying techniques derived from AI to large volumes of data goes by names such as “data mining,” “big data,” “analytics,” etc. This field is too vast to even moderately cover in the present article, but we note that there is no full agreement on what constitutes such a “big-data” problem. One definition, from Madden (2012), is that big data differs from traditional machine-processable data in that it is too big (for most of the existing state-of-the-art hardware), too quick (generated at a fast rate, e.g. online email transactions), or too hard. It is in the too-hard part that AI techniques work quite well. While this universe is quite varied, we use the Watson’s system later in this article as an AI-relevant exemplar. As we will see later, while most of this new explosion is powered by learning, it isn’t entirely limited to just learning. This bloom in learning algorithms has been supported by both a resurgence in neurocomputational techniques and probabilistic techniques.

One of the remarkable aspects of (Charniak & McDermott 1985) is this: The authors say the central dogma of AI is that “What the brain does may be thought of at some level as a kind of computation” (p. 6). And yet nowhere in the book is brain-like computation discussed. In fact, you will search the index in vain for the term ‘neural’ and its variants. Please note that the authors are not to blame for this. A large part of AI’s growth has come from formalisms, tools, and techniques that are, in some sense, brain-based, not logic-based. A paper that conveys the importance and maturity of neurocomputation is (Litt et al. 2006). (Growth has also come from a return of probabilistic techniques that had withered by the mid-70s and 80s. More about that momentarily, in the next “resurgence” section .)

One very prominent class of non-logicist formalism does make an explicit nod in the direction of the brain: viz., artificial neural networks (or as they are often simply called, neural networks , or even just neural nets ). (The structure of neural networks and more recent developments are discussed above ). Because Minsky and Pappert’s (1969) Perceptrons led many (including, specifically, many sponsors of AI research and development) to conclude that neural networks didn’t have sufficient information-processing power to model human cognition, the formalism was pretty much universally dropped from AI. However, Minsky and Pappert had only considered very limited neural networks. Connectionism , the view that intelligence consists not in symbolic processing, but rather non -symbolic processing at least somewhat like what we find in the brain (at least at the cellular level), approximated specifically by artificial neural networks, came roaring back in the early 1980s on the strength of more sophisticated forms of such networks, and soon the situation was (to use a metaphor introduced by John McCarthy) that of two horses in a race toward building truly intelligent agents.

If one had to pick a year at which connectionism was resurrected, it would certainly be 1986, the year Parallel Distributed Processing (Rumelhart & McClelland 1986) appeared in print. The rebirth of connectionism was specifically fueled by the back-propagation (backpropagation) algorithm over neural networks, nicely covered in Chapter 20 of AIMA . The symbolicist/connectionist race led to a spate of lively debate in the literature (e.g., Smolensky 1988, Bringsjord 1991), and some AI engineers have explicitly championed a methodology marked by a rejection of knowledge representation and reasoning. For example, Rodney Brooks was such an engineer; he wrote the well-known “Intelligence Without Representation” (1991), and his Cog Project, to which we referred above, is arguably an incarnation of the premeditatedly non-logicist approach. Increasingly, however, those in the business of building sophisticated systems find that both logicist and more neurocomputational techniques are required (Wermter & Sun 2001). [ 33 ] In addition, the neurocomputational paradigm today includes connectionism only as a proper part, in light of the fact that some of those working on building intelligent systems strive to do so by engineering brain-based computation outside the neural network-based approach (e.g., Granger 2004a, 2004b).

Another recent resurgence in neurocomputational techniques has occurred in machine learning. The modus operandi in machine learning is that given a problem, say recognizing handwritten digits \(\{0,1,\ldots,9\}\) or faces, from a 2D matrix representing an image of the digits or faces, a machine learning or a domain expert would construct a feature vector representation function for the task. This function is a transformation of the input into a format that tries to throw away irrelevant information in the input and keep only information useful for the task. Inputs transformed by \(\rr\) are termed features . For recognizing faces, irrelevant information could be the amount of lighting in the scene and relevant information could be information about facial features. The machine is then fed a sequence of inputs represented by the features and the ideal or ground truth output values for those inputs. This converts the learning challenge from that of having to learn the function \(\ff\) from the examples: \(\left\{\left\langle x_1, \ff(x_1)\right\rangle,\left\langle x_2, \ff(x_2)\right\rangle, \ldots, \left\langle x_n, \ff(x_n)\right\rangle \right\}\) to having to learn from possibly easier data: \(\left\{\left\langle \rr(x_1), \ff(x_1)\right\rangle,\left\langle \rr(x_2), \ff(x_2)\right\rangle, \ldots, \left\langle \rr(x_n), \ff(x_n)\right\rangle \right\}\). Here the function \(\rr\) is the function that computes the feature vector representation of the input. Formally, \(\ff\) is assumed to be a composition of the functions \(\gg\) and \(\rr\). That is, for any input \(x\), \(f(x) = \gg\left(\rr\left(x\right)\right)\). This is denoted by \(\ff=\gg\circ \rr\). For any input, the features are first computed, and then the function \(\gg\) is applied. If the feature representation \(\rr\) is provided by the domain expert, the learning problem becomes simpler to the extent the feature representation takes on the difficulty of the task. At one extreme, the feature vector could hide an easily extractable form of the answer in the input and in the other extreme the feature representation could be just the plain input.

For non-trivial problems, choosing the right representation is vital. For instance, one of the drastic changes in the AI landscape was due to Minsky and Papert’s (1969) demonstration that the perceptron cannot learn even the binary XOR function, but this function can be learnt by the perceptron if we have the right representation. Feature engineering has grown to be one of the most labor intensive tasks of machine learning, so much so that it is considered to be one of the “black arts” of machine learning. The other significant black art of learning methods is choosing the right parameters. These black arts require significant human expertise and experience, which can be quite difficult to obtain without significant apprenticeship (Domingos 2012). Another bigger issue is that the task of feature engineering is just knowledge representation in a new skin.

Given this state of affairs, there has been a recent resurgence in methods for automatically learning a feature representation function \(\rr\); such methods potentially bypass a large part of human labor that is traditionally required. Such methods are based mostly on what are now termed deep neural networks . Such networks are simply neural networks with two or more hidden layers. These networks allow us to learn a feature function \(\rr\) by using one or more of the hidden layers to learn \(\rr\). The general form of learning in which one learns from the raw sensory data without much hand-based feature engineering has now its own term: deep learning . A general and yet concise definition (Bengio et al. 2015) is:

Deep learning can safely be regarded as the study of models that either involve a greater amount of composition of learned functions or learned concepts than traditional machine learning does. (Bengio et al. 2015, Chapter 1)

Though the idea has been around for decades, recent innovations leading to more efficient learning techniques have made the approach more feasible (Bengio et al. 2013). Deep-learning methods have recently produced state-of-the-art results in image recognition (given an image containing various objects, label the objects from a given set of labels), speech recognition (from audio input, generate a textual representation), and the analysis of data from particle accelerators (LeCun et al. 2015). Despite impressive results in tasks such as these, minor and major issues remain unresolved. A minor issue is that significant human expertise is still needed to choose an architecture and set up the right parameters for the architecture; a major issue is the existence of so-called adversarial inputs , which are indistinguishable from normal inputs to humans but are computed in a special manner that makes a neural network regard them as different than similar inputs in the training data. The existence of such adversarial inputs, which remain stable across training data, has raised doubts about how well performance on benchmarks can translate into performance in real-world systems with sensory noise (Szegedy et al. 2014).

There is a second dimension to the explosive growth of AI: the explosion in popularity of probabilistic methods that aren’t neurocomputational in nature, in order to formalize and mechanize a form of non-logicist reasoning in the face of uncertainty. Interestingly enough, it is Eugene Charniak himself who can be safely considered one of the leading proponents of an explicit, premeditated turn away from logic to statistical techniques. His area of specialization is natural language processing, and whereas his introductory textbook of 1985 gave an accurate sense of his approach to parsing at the time (as we have seen, write computer programs that, given English text as input, ultimately infer meaning expressed in FOL), this approach was abandoned in favor of purely statistical approaches (Charniak 1993). At the AI@50 conference, Charniak boldly proclaimed, in a talk tellingly entitled “Why Natural Language Processing is Now Statistical Natural Language Processing,” that logicist AI is moribund, and that the statistical approach is the only promising game in town – for the next 50 years. [ 34 ]

The chief source of energy and debate at the conference flowed from the clash between Charniak’s probabilistic orientation, and the original logicist orientation, upheld at the conference in question by John McCarthy and others.

AI’s use of probability theory grows out of the standard form of this theory, which grew directly out of technical philosophy and logic. This form will be familiar to many philosophers, but let’s review it quickly now, in order to set a firm stage for making points about the new probabilistic techniques that have energized AI.

Just as in the case of FOL, in probability theory we are concerned with declarative statements, or propositions , to which degrees of belief are applied; we can thus say that both logicist and probabilistic approaches are symbolic in nature. Both approaches also agree that statements can either be true or false in the world. In building agents, a simplistic logic-based approach requires agents to know the truth-value of all possible statements. This is not realistic, as an agent may not know the truth-value of some proposition \(p\) due to either ignorance, non-determinism in the physical world, or just plain vagueness in the meaning of the statement. More specifically, the fundamental proposition in probability theory is a random variable , which can be conceived of as an aspect of the world whose status is initially unknown to the agent. We usually capitalize the names of random variables, though we reserve \(p,q,r, \ldots\) as such names as well. For example, in a particular murder investigation centered on whether or not Mr. Barolo committed the crime, the random variable \(Guilty\) might be of concern. The detective may be interested as well in whether or not the murder weapon – a particular knife, let us assume – belongs to Barolo. In light of this, we might say that \(\Weapon = \true\) if it does, and \(\Weapon = \false\) if it doesn’t. As a notational convenience, we can write \(weapon\) and \(\lnot weapon\) and for these two cases, respectively; and we can use this convention for other variables of this type.

The kind of variables we have described so far are \(\mathbf{Boolean}\), because their \(\mathbf{domain}\) is simply \(\{true,false\}.\) But we can generalize and allow \(\mathbf{discrete}\) random variables, whose values are from any countable domain. For example, \(\PriceTChina\) might be a variable for the price of (a particular, presumably) tea in China, and its domain might be \(\{1,2,3,4,5\}\), where each number here is in US dollars. A third type of variable is \(\mathbf{continous}\); its domain is either the reals, or some subset thereof.

We say that an atomic event is an assignment of particular values from the appropriate domains to all the variables composing the (idealized) world. For example, in the simple murder investigation world introduced just above, we have two Boolean variables, \(\Guilty\) and \(\Weapon\), and there are just four atomic events. Note that atomic events have some obvious properties. For example, they are mutually exclusive, exhaustive, and logically entail the truth or falsity of every proposition. Usually not obvious to beginning students is a fourth property, namely, any proposition is logically equivalent to the disjunction of all atomic events that entail that proposition.

Prior probabilities correspond to a degree of belief accorded to a proposition in the complete absence of any other information. For example, if the prior probability of Barolo’s guilt is \(0.2\), we write \[ P\left(\Guilty=true\right)=0.2 \]

or simply \(\P(guilty)=0.2\). It is often convenient to have a notation allowing one to refer economically to the probabilities of all the possible values for a random variable. For example, we can write \[ \P\left(\PriceTChina\right) \]

as an abbreviation for the five equations listing all the possible prices for tea in China. We can also write \[ \P\left(\PriceTChina\right)=\langle 1,2,3,4,5\rangle \]

In addition, as further convenient notation, we can write \( \mathbf{P}\left(\Guilty, \Weapon\right)\) to denote the probabilities of all combinations of values of the relevant set of random variables. This is referred to as the joint probability distribution of \(\Guilty\) and \(\Weapon\). The full joint probability distribution covers the distribution for all the random variables used to describe a world. Given our simple murder world, we have 20 atomic events summed up in the equation \[ \mathbf{P}\left(\Guilty, \Weapon, \PriceTChina\right) \]

The final piece of the basic language of probability theory corresponds to conditional probabilities. Where \(p\) and \(q\) are any propositions, the relevant expression is \(P\!\left(p\given q\right)\), which can be interpreted as “the probability of \(p\), given that all we know is \(q\).” For example, \[ P\left(guilty\ggiven weapon\right)=0.7 \]

says that if the murder weapon belongs to Barolo, and no other information is available, the probability that Barolo is guilty is \(0.7.\)

Andrei Kolmogorov showed how to construct probability theory from three axioms that make use of the machinery now introduced, viz.,

• All probabilities fall between \(0\) and \(1.\) I.e., \(\forall p. 0 \leq P(p) \leq 1\).
• Valid (in the traditional logicist sense) propositions have a probability of \(1\); unsatisfiable (in the traditional logicist sense) propositions have a probability of \(0\).
• \(P(p\lor q) = P(p) +P(q) - P(p\land q)\)

These axioms are clearly at bottom logicist. The remainder of probability theory can be erected from this foundation (conditional probabilities are easily defined in terms of prior probabilities). We can thus say that logic is in some fundamental sense still being used to characterize the set of beliefs that a rational agent can have. But where does probabilistic inference enter the picture on this account, since traditional deduction is not used for inference in probability theory?

Probabilistic inference consists in computing, from observed evidence expressed in terms of probability theory, posterior probabilities of propositions of interest. For a good long while, there have been algorithms for carrying out such computation. These algorithms precede the resurgence of probabilistic techniques in the 1990s. (Chapter 13 of AIMA presents a number of them.) For example, given the Kolmogorov axioms, here is a straightforward way of computing the probability of any proposition, using the full joint distribution giving the probabilities of all atomic events: Where \(p\) is some proposition, let \(\alpha(p)\) be the disjunction of all atomic events in which \(p\) holds. Since the probability of a proposition (i.e., \(P(p)\)) is equal to the sum of the probabilities of the atomic events in which it holds, we have an equation that provides a method for computing the probability of any proposition \(p\), viz.,

Unfortunately, there were two serious problems infecting this original probabilistic approach: One, the processing in question needed to take place over paralyzingly large amounts of information (enumeration over the entire distribution is required). And two, the expressivity of the approach was merely propositional. (It was by the way the philosopher Hilary Putnam (1963) who pointed out that there was a price to pay in moving to the first-order level. The issue is not discussed herein.) Everything changed with the advent of a new formalism that marks the marriage of probabilism and graph theory: Bayesian networks (also called belief nets ). The pivotal text was (Pearl 1988). For a more detailed discussion, see the

Supplement on Bayesian Networks .

Before concluding this section, it is probably worth noting that, from the standpoint of philosophy, a situation such as the murder investigation we have exploited above would often be analyzed into arguments , and strength factors, not into numbers to be crunched by purely arithmetical procedures. For example, in the epistemology of Roderick Chisholm, as presented his Theory of Knowledge (1966, 1977), Detective Holmes might classify a proposition like Barolo committed the murder. as counterbalanced if he was unable to find a compelling argument either way, or perhaps probable if the murder weapon turned out to belong to Barolo. Such categories cannot be found on a continuum from 0 to 1, and they are used in articulating arguments for or against Barolo’s guilt. Argument-based approaches to uncertain and defeasible reasoning are virtually non-existent in AI. One exception is Pollock’s approach, covered below. This approach is Chisholmian in nature.

It should also be noted that there have been well-established formalisms for dealing with probabilistic reasoning as an instance of logic-based reasoning. E.g., the activity a researcher in probabilistic reasoning undertakes when she proves a theorem \(\phi\) about their domain (e.g. any theorem in (Pearl 1988)) is purely within the realm of traditional logic. Readers interested in logic-flavored approaches to probabilistic reasoning can consult (Adams 1996, Hailperin 1996 & 2010, Halpern 1998). Formalisms marrying probability theory, induction and deductive reasoning, placing them on an equal footing, have been on the rise, with Markov logic (Richardson and Domingos 2006) being salient among these approaches.

Probabilistic Machine Learning

Machine learning, in the sense given above , has been associated with probabilistic techniques. Probabilistic techniques have been associated with both the learning of functions (e.g. Naive Bayes classification) and the modeling of theoretical properties of learning algorithms. For example, a standard reformulation of supervised learning casts it as a Bayesian problem . Assume that we are looking at recognizing digits \([0{-}9]\) from a given image. One way to cast this problem is to ask what the probability that the hypothesis \(H_x\): “ the digit is \(x\) ” is true given the image \(d\) from a sensor. Bayes theorem gives us:

\(P(d\given H_x)\) and \(P(H_x)\) can be estimated from the given training dataset. Then the hypothesis with the highest posterior probability is then given as the answer and is given by: \(\argmax_{x}P\left(d\ggiven H_x\right)*P\left(H_x\right) \) In addition to probabilistic methods being used to build algorithms, probability theory has also been used to analyze algorithms which might not have an overt probabilistic or logical formulation. For example, one of the central classes of meta-theorems in learning, probably approximately correct (PAC) theorems, are cast in terms of lower bounds of the probability that the mismatch between the induced/learnt f L function and the true function f T being less than a certain amount, given that the learnt function f L works well for a certain number of cases (see Chapter 18, AIMA).

From at least its modern inception, AI has always been connected to gadgets, often ones produced by corporations, and it would be remiss of us not to say a few words about this phenomenon. While there have been a large number of commercial in-the-wild success stories for AI and its sister fields, such as optimization and decision-making, some applications are more visible and have been thoroughly battle-tested in the wild. In 2014, one of the most visible such domains (one in which AI has been strikingly successful) is information retrieval, incarnated as web search. Another recent success story is pattern recognition. The state-of-the-art in applied pattern recognition (e.g., fingerprint/face verification, speech recognition, and handwriting recognition) is robust enough to allow “high-stakes” deployment outside the laboratory. As of mid 2018, several corporations and research laboratories have begun testing autonomous vehicles on public roads, with even a handful of jurisdictions making self-driving cars legal to operate. For example, Google’s autonomous cars have navigated hundreds of thousands of miles in California with minimal human help under non-trivial conditions (Guizzo 2011).

Computer games provide a robust test bed for AI techniques as they can capture important parts that might be necessary to test an AI technique while abstracting or removing details that might beyond the scope of core AI research, for example, designing better hardware or dealing with legal issues (Laird and VanLent 2001). One subclass of games that has seen quite fruitful for commercial deployment of AI is real-time strategy games. Real-time strategy games are games in which players manage an army given limited resources. One objective is to constantly battle other players and reduce an opponent’s forces. Real-time strategy games differ from strategy games in that players plan their actions simultaneously in real-time and do not have to take turns playing. Such games have a number of challenges that are tantalizing within the grasp of the state-of-the-art. This makes such games an attractive venue in which to deploy simple AI agents. An overview of AI used in real-time strategy games can be found in (Robertson and Watson 2015).

Some other ventures in AI, despite significant success, have been only chugging slowly and humbly along, quietly. For instance, AI-related methods have achieved triumphs in solving open problems in mathematics that have resisted any solution for decades. The most noteworthy instance of such a problem is perhaps a proof of the statement that “ All Robbins algebras are Boolean algebras. ” This was conjectured in the 1930s, and the proof was finally discovered by the Otter automatic theorem-prover in 1996 after just a few months of effort (Kolata 1996, Wos 2013). Sister fields like formal verification have also bloomed to the extent that it is now not too difficult to semi-automatically verify vital hardware/software components (Kaufmann et al. 2000 and Chajed et al. 2017).

Other related areas, such as (natural) language translation, still have a long way to go, but are good enough to let us use them under restricted conditions. The jury is out on tasks such as machine translation, which seems to require both statistical methods (Lopez 2008) and symbolic methods (España-Bonet 2011). Both methods now have comparable but limited success in the wild. A deployed translation system at Ford that was initially developed for translating manufacturing process instructions from English to other languages initially started out as rule-based system with Ford and domain-specific vocabulary and language. This system then evolved to incorporate statistical techniques along with rule-based techniques as it gained new uses beyond translating manuals, for example, lay users within Ford translating their own documents (Rychtyckyj and Plesco 2012).

AI’s great achievements mentioned above so far have all been in limited, narrow domains. This lack of any success in the unrestricted general case has caused a small set of researchers to break away into what is now called artificial general intelligence (Goertzel and Pennachin 2007). The stated goals of this movement include shifting the focus again to building artifacts that are generally intelligent and not just capable in one narrow domain.

Computer Ethics has been around for a long time. In this sub-field, typically one would consider how one ought to act in a certain class of situations involving computer technology, where the “one” here refers to a human being (Moor 1985). So-called “robot ethics” is different. In this sub-field (which goes by names such as “moral AI,” “ethical AI,” “machine ethics,” “moral robots,” etc.) one is confronted with such prospects as robots being able to make autonomous and weighty decisions – decisions that might or might not be morally permissible (Wallach & Allen 2010). If one were to attempt to engineer a robot with a capacity for sophisticated ethical reasoning and decision-making, one would also be doing Philosophical AI, as that concept is characterized elsewhere in the present entry. There can be many different flavors of approaches toward Moral AI. Wallach and Allen (2010) provide a high-level overview of the different approaches. Moral reasoning is obviously needed in robots that have the capability for lethal action. Arkin (2009) provides an introduction to how we can control and regulate machines that have the capacity for lethal behavior. Moral AI goes beyond obviously lethal situations, and we can have a spectrum of moral machines. Moor (2006) provides one such spectrum of possible moral agents. An example of a non-lethal but ethically-charged machine would be a lying machine. Clark (2010) uses a computational theory of the mind , the ability to represent and reason about other agents, to build a lying machine that successfully persuades people into believing falsehoods. Bello & Bringsjord (2013) give a general overview of what might be required to build a moral machine, one of the ingredients being a theory of mind.

The most general framework for building machines that can reason ethically consists in endowing the machines with a moral code . This requires that the formal framework used for reasoning by the machine be expressive enough to receive such codes. The field of Moral AI, for now, is not concerned with the source or provenance of such codes. The source could be humans, and the machine could receive the code directly (via explicit encoding) or indirectly (reading). Another possibility is that the code is inferred by the machine from a more basic set of laws. We assume that the robot has access to some such code, and we then try to engineer the robot to follow that code under all circumstances while making sure that the moral code and its representation do not lead to unintended consequences. Deontic logics are a class of formal logics that have been studied the most for this purpose. Abstractly, such logics are concerned mainly with what follows from a given moral code. Engineering then studies the match of a given deontic logic to a moral code (i.e., is the logic expressive enough) which has to be balanced with the ease of automation. Bringsjord et al. (2006) provide a blueprint for using deontic logics to build systems that can perform actions in accordance with a moral code. The role deontic logics play in the framework offered by Bringsjord et al (which can be considered to be representative of the field of deontic logic for moral AI) can be best understood as striving towards Leibniz’s dream of a universal moral calculus:

When controversies arise, there will be no more need for a disputation between two philosophers than there would be between two accountants [computistas]. It would be enough for them to pick up their pens and sit at their abacuses, and say to each other (perhaps having summoned a mutual friend): ‘Let us calculate.’

Deontic logic-based frameworks can also be used in a fashion that is analogous to moral self-reflection. In this mode, logic-based verification of the robot’s internal modules can done before the robot ventures out into the real world. Govindarajulu and Bringsjord (2015) present an approach, drawing from formal-program verification , in which a deontic-logic based system could be used to verify that a robot acts in a certain ethically-sanctioned manner under certain conditions. Since formal-verification approaches can be used to assert statements about an infinite number of situations and conditions, such approaches might be preferred to having the robot roam around in an ethically-charged test environment and make a finite set of decisions that are then judged for their ethical correctness. More recently, Govindarajulu and Bringsjord (2017) use a deontic logic to present a computational model of the Doctrine of Double Effect , an ethical principle for moral dilemmas that has been studied empirically and analyzed extensively by philosophers. [ 35 ] The principle is usually presented and motivated via dilemmas using trolleys and was first presented in this fashion by Foot (1967).

While there has been substantial theoretical and philosophical work, the field of machine ethics is still in its infancy. There has been some embryonic work in building ethical machines. One recent such example would be Pereira and Saptawijaya (2016) who use logic programming and base their work in machine ethics on the ethical theory known as contractualism , set out by Scanlon (1982). And what about the future? Since artificial agents are bound to get smarter and smarter, and to have more and more autonomy and responsibility, robot ethics is almost certainly going to grow in importance. This endeavor might not be a straightforward application of classical ethics. For example, experimental results suggest that humans hold robots to different ethical standards than they expect from humans under similar conditions (Malle et al. 2015). [ 36 ]

Notice that the heading for this section isn’t Philosophy of AI. We’ll get to that category momentarily. (For now it can be identified with the attempt to answer such questions as whether artificial agents created in AI can ever reach the full heights of human intelligence.) Philosophical AI is AI, not philosophy; but it’s AI rooted in and flowing from, philosophy. For example, one could engage, using the tools and techniques of philosophy, a paradox, work out a proposed solution, and then proceed to a step that is surely optional for philosophers: expressing the solution in terms that can be translated into a computer program that, when executed, allows an artificial agent to surmount concrete instances of the original paradox. [ 37 ] Before we ostensively characterize Philosophical AI of this sort courtesy of a particular research program, let us consider first the view that AI is in fact simply philosophy, or a part thereof.

Daniel Dennett (1979) has famously claimed not just that there are parts of AI intimately bound up with philosophy, but that AI is philosophy (and psychology, at least of the cognitive sort). (He has made a parallel claim about Artificial Life (Dennett 1998)). This view will turn out to be incorrect, but the reasons why it’s wrong will prove illuminating, and our discussion will pave the way for a discussion of Philosophical AI.

What does Dennett say, exactly? This:

I want to claim that AI is better viewed as sharing with traditional epistemology the status of being a most general, most abstract asking of the top-down question: how is knowledge possible? (Dennett 1979, 60)

Elsewhere he says his view is that AI should be viewed “as a most abstract inquiry into the possibility of intelligence or knowledge” (Dennett 1979, 64).

In short, Dennett holds that AI is the attempt to explain intelligence, not by studying the brain in the hopes of identifying components to which cognition can be reduced, and not by engineering small information-processing units from which one can build in bottom-up fashion to high-level cognitive processes, but rather by – and this is why he says the approach is top-down – designing and implementing abstract algorithms that capture cognition. Leaving aside the fact that, at least starting in the early 1980s, AI includes an approach that is in some sense bottom-up (see the neurocomputational paradigm discussed above, in Non-Logicist AI: A Summary ; and see, specifically, Granger’s (2004a, 2004b) work, hyperlinked in text immediately above, a specific counterexample), a fatal flaw infects Dennett’s view. Dennett sees the potential flaw, as reflected in:

It has seemed to some philosophers that AI cannot plausibly be so construed because it takes on an additional burden: it restricts itself to mechanistic solutions, and hence its domain is not the Kantian domain of all possible modes of intelligence, but just all possible mechanistically realizable modes of intelligence. This, it is claimed, would beg the question against vitalists, dualists, and other anti-mechanists. (Dennett 1979, 61)

Dennett has a ready answer to this objection. He writes:

But … the mechanism requirement of AI is not an additional constraint of any moment, for if psychology is possible at all, and if Church’s thesis is true, the constraint of mechanism is no more severe than the constraint against begging the question in psychology, and who would wish to evade that? (Dennett 1979, 61)

Unfortunately, this is acutely problematic; and examination of the problems throws light on the nature of AI.

First, insofar as philosophy and psychology are concerned with the nature of mind, they aren’t in the least trammeled by the presupposition that mentation consists in computation. AI, at least of the “Strong” variety (we’ll discuss “Strong” versus “Weak” AI below ) is indeed an attempt to substantiate, through engineering certain impressive artifacts, the thesis that intelligence is at bottom computational (at the level of Turing machines and their equivalents, e.g., Register machines). So there is a philosophical claim, for sure. But this doesn’t make AI philosophy, any more than some of the deeper, more aggressive claims of some physicists (e.g., that the universe is ultimately digital in nature) make their field philosophy. Philosophy of physics certainly entertains the proposition that the physical universe can be perfectly modeled in digital terms (in a series of cellular automata, e.g.), but of course philosophy of physics can’t be identified with this doctrine.

Second, we now know well (and those familiar with the relevant formal terrain knew at the time of Dennett’s writing) that information processing can exceed standard computation, that is, can exceed computation at and below the level of what a Turing machine can muster ( Turing-computation , we shall say). (Such information processing is known as hypercomputation , a term coined by philosopher Jack Copeland, who has himself defined such machines (e.g., Copeland 1998). The first machines capable of hypercomputation were trial-and-error machines , introduced in the same famous issue of the Journal of Symbolic Logic (Gold 1965; Putnam 1965). A new hypercomputer is the infinite time Turing machine (Hamkins & Lewis 2000).) Dennett’s appeal to Church’s thesis thus flies in the face of the mathematical facts: some varieties of information processing exceed standard computation (or Turing-computation). Church’s thesis, or more precisely, the Church-Turing thesis, is the view that a function \(f\) is effectively computable if and only if \(f\) is Turing-computable (i.e., some Turing machine can compute \(f\)). Thus, this thesis has nothing to say about information processing that is more demanding than what a Turing machine can achieve. (Put another way, there is no counter-example to CTT to be automatically found in an information-processing device capable of feats beyond the reach of TMs.) For all philosophy and psychology know, intelligence, even if tied to information processing, exceeds what is Turing-computational or Turing-mechanical. [ 38 ] This is especially true because philosophy and psychology, unlike AI, are in no way fundamentally charged with engineering artifacts, which makes the physical realizability of hypercomputation irrelevant from their perspectives. Therefore, contra Dennett, to consider AI as psychology or philosophy is to commit a serious error, precisely because so doing would box these fields into only a speck of the entire space of functions from the natural numbers (including tuples therefrom) to the natural numbers. (Only a tiny portion of the functions in this space are Turing-computable.) AI is without question much, much narrower than this pair of fields. Of course, it’s possible that AI could be replaced by a field devoted not to building computational artifacts by writing computer programs and running them on embodied Turing machines. But this new field, by definition, would not be AI. Our exploration of AIMA and other textbooks provide direct empirical confirmation of this.

Third, most AI researchers and developers, in point of fact, are simply concerned with building useful, profitable artifacts, and don’t spend much time reflecting upon the kinds of abstract definitions of intelligence explored in this entry (e.g., What Exactly is AI? ).

Though AI isn’t philosophy, there are certainly ways of doing real implementation-focussed AI of the highest caliber that are intimately bound up with philosophy. The best way to demonstrate this is to simply present such research and development, or at least a representative example thereof. While there have been many examples of such work, the most prominent example in AI is John Pollock’s OSCAR project, which stretched over a considerable portion of his lifetime. For a detailed presentation and further discussion, see the

Supplement on the OSCAR Project.

It’s important to note at this juncture that the OSCAR project, and the information processing that underlies it, are without question at once philosophy and technical AI. Given that the work in question has appeared in the pages of Artificial Intelligence , a first-rank journal devoted to that field, and not to philosophy, this is undeniable (see, e.g., Pollock 2001, 1992). This point is important because while it’s certainly appropriate, in the present venue, to emphasize connections between AI and philosophy, some readers may suspect that this emphasis is contrived: they may suspect that the truth of the matter is that page after page of AI journals are filled with narrow, technical content far from philosophy. Many such papers do exist. But we must distinguish between writings designed to present the nature of AI, and its core methods and goals, versus writings designed to present progress on specific technical issues.

Writings in the latter category are more often than not quite narrow, but, as the example of Pollock shows, sometimes these specific issues are inextricably linked to philosophy. And of course Pollock’s work is a representative example (albeit the most substantive one). One could just as easily have selected work by folks who don’t happen to also produce straight philosophy. For example, for an entire book written within the confines of AI and computer science, but which is epistemic logic in action in many ways, suitable for use in seminars on that topic, see (Fagin et al. 2004). (It is hard to find technical work that isn’t bound up with philosophy in some direct way. E.g., AI research on learning is all intimately bound up with philosophical treatments of induction, of how genuinely new concepts not simply defined in terms of prior ones can be learned. One possible partial answer offered by AI is inductive logic programming , discussed in Chapter 19 of AIMA .)

What of writings in the former category? Writings in this category, while by definition in AI venues, not philosophy ones, are nonetheless philosophical. Most textbooks include plenty of material that falls into this latter category, and hence they include discussion of the philosophical nature of AI (e.g., that AI is aimed at building artificial intelligences, and that’s why, after all, it’s called ‘AI’).

8. Philosophy of Artificial Intelligence

Recall that we earlier discussed proposed definitions of AI, and recall specifically that these proposals were couched in terms of the goals of the field. We can follow this pattern here: We can distinguish between “Strong” and “Weak” AI by taking note of the different goals that these two versions of AI strive to reach. “Strong” AI seeks to create artificial persons: machines that have all the mental powers we have, including phenomenal consciousness. “Weak” AI, on the other hand, seeks to build information-processing machines that appear to have the full mental repertoire of human persons (Searle 1997). “Weak” AI can also be defined as the form of AI that aims at a system able to pass not just the Turing Test (again, abbreviated as TT), but the Total Turing Test (Harnad 1991). In TTT, a machine must muster more than linguistic indistinguishability: it must pass for a human in all behaviors – throwing a baseball, eating, teaching a class, etc.

It would certainly seem to be exceedingly difficult for philosophers to overthrow “Weak” AI (Bringsjord and Xiao 2000). After all, what philosophical reason stands in the way of AI producing artifacts that appear to be animals or even humans? However, some philosophers have aimed to do in “Strong” AI, and we turn now to the most prominent case in point.

Without question, the most famous argument in the philosophy of AI is John Searle’s (1980) Chinese Room Argument (CRA), designed to overthrow “Strong” AI. We present a quick summary here and a “report from the trenches” as to how AI practitioners regard the argument. Readers wanting to further study CRA will find an excellent next step in the entry on the Chinese Room Argument and (Bishop & Preston 2002).

CRA is based on a thought-experiment in which Searle himself stars. He is inside a room; outside the room are native Chinese speakers who don’t know that Searle is inside it. Searle-in-the-box, like Searle-in-real-life, doesn’t know any Chinese, but is fluent in English. The Chinese speakers send cards into the room through a slot; on these cards are written questions in Chinese. The box, courtesy of Searle’s secret work therein, returns cards to the native Chinese speakers as output. Searle’s output is produced by consulting a rulebook: this book is a lookup table that tells him what Chinese to produce based on what is sent in. To Searle, the Chinese is all just a bunch of – to use Searle’s language – squiggle-squoggles. The following schematic picture sums up the situation. The labels should be obvious. \(O\) denotes the outside observers, in this case the Chinese speakers. Input is denoted by \(i\) and output by \(o\). As you can see, there is an icon for the rulebook, and Searle himself is denoted by \(P\).

The Chinese Room, Schematic View

Now, what is the argument based on this thought-experiment? Even if you’ve never heard of CRA before, you doubtless can see the basic idea: that Searle (in the box) is supposed to be everything a computer can be, and because he doesn’t understand Chinese, no computer could have such understanding. Searle is mindlessly moving squiggle-squoggles around, and (according to the argument) that’s all computers do, fundamentally. [ 39 ]

Where does CRA stand today? As we’ve already indicated, the argument would still seem to be alive and well; witness (Bishop & Preston 2002). However, there is little doubt that at least among AI practitioners , CRA is generally rejected. (This is of course thoroughly unsurprising.) Among these practitioners, the philosopher who has offered the most formidable response out of AI itself is Rapaport (1988), who argues that while AI systems are indeed syntactic, the right syntax can constitute semantics. It should be said that a common attitude among proponents of “Strong” AI is that CRA is not only unsound, but silly, based as it is on a fanciful story (CR) far removed from the practice of AI – practice which is year by year moving ineluctably toward sophisticated robots that will once and for all silence CRA and its proponents. For example, John Pollock (as we’ve noted, philosopher and practitioner of AI) writes:

Once [my intelligent system] OSCAR is fully functional, the argument from analogy will lead us inexorably to attribute thoughts and feelings to OSCAR with precisely the same credentials with which we attribute them to human beings. Philosophical arguments to the contrary will be passé. (Pollock 1995, p. 6)

To wrap up discussion of CRA, we make two quick points, to wit:

• Despite the confidence of the likes of Pollock about the eventual irrelevance of CRA in the face of the eventual human-level prowess of OSCAR (and, by extension, any number of other still-improving AI systems), the brute fact is that deeply semantic natural-language processing (NLP) is rarely even pursued these days, so proponents of CRA are certainly not the ones feeling some discomfort in light of the current state of AI. In short, Searle would rightly point to any of the success stories of AI, including the Watson system we have discussed, and still proclaim that understanding is nowhere to be found – and he would be well within his philosophical rights in saying this.
• It would appear that the CRA is bubbling back to a level of engagement not seen for a number of years, in light of the empirical fact that certain thinkers are now issuing explicit warnings to the effect that future conscious, malevolent machines may well wish to do in our species. In reply, Searle (2014) points out that since CRA is sound, there can’t be conscious machines; and if there can’t be conscious machines, there can’t be malevolent machines that wish anything. We return to this at the end of our entry; the chief point here is that CRA continues to be quite relevant, and indeed we suspect that Searle’s basis for have-no-fear will be taken up energetically by not only philosophers, but AI experts, futurists, lawyers, and policy-makers.

Readers may wonder if there are philosophical debates that AI researchers engage in, in the course of working in their field (as opposed to when they might attend a philosophy conference). Surely, AI researchers have philosophical discussions amongst themselves, right?

Generally, one finds that AI researchers do discuss among themselves topics in philosophy of AI, and these topics are usually the very same ones that occupy philosophers of AI. However, the attitude reflected in the quote from Pollock immediately above is by far the dominant one. That is, in general, the attitude of AI researchers is that philosophizing is sometimes fun, but the upward march of AI engineering cannot be stopped, will not fail, and will eventually render such philosophizing otiose.

We will return to the issue of the future of AI in the final section of this entry.

Four decades ago, J.R. Lucas (1964) argued that Gödel’s first incompleteness theorem entails that no machine can ever reach human-level intelligence. His argument has not proved to be compelling, but Lucas initiated a debate that has produced more formidable arguments. One of Lucas’ indefatigable defenders is the physicist Roger Penrose, whose first attempt to vindicate Lucas was a Gödelian attack on “Strong” AI articulated in his The Emperor’s New Mind (1989). This first attempt fell short, and Penrose published a more elaborate and more fastidious Gödelian case, expressed in Chapters 2 and 3 of his Shadows of the Mind (1994).

In light of the fact that readers can turn to the entry on the Gödel’s Incompleteness Theorems , a full review here is not needed. Instead, readers will be given a decent sense of the argument by turning to an online paper in which Penrose, writing in response to critics (e.g., the philosopher David Chalmers, the logician Solomon Feferman, and the computer scientist Drew McDermott) of his Shadows of the Mind , distills the argument to a couple of paragraphs. [ 40 ] Indeed, in this paper Penrose gives what he takes to be the perfected version of the core Gödelian case given in SOTM . Here is this version, verbatim:

We try to suppose that the totality of methods of (unassailable) mathematical reasoning that are in principle humanly accessible can be encapsulated in some (not necessarily computational) sound formal system \(F\). A human mathematician, if presented with \(F\), could argue as follows (bearing in mind that the phrase “I am \(F\)” is merely a shorthand for “\(F\) encapsulates all the humanly accessible methods of mathematical proof”): (A) “Though I don’t know that I necessarily am \(F\), I conclude that if I were, then the system \(F\) would have to be sound and, more to the point, \(F'\) would have to be sound, where \(F'\) is \(F\) supplemented by the further assertion “I am \(F\).” I perceive that it follows from the assumption that I am \(F\) that the Gödel statement \(G(F')\) would have to be true and, furthermore, that it would not be a consequence of \(F'\). But I have just perceived that “If I happened to be \(F\), then \(G(F')\) would have to be true,” and perceptions of this nature would be precisely what \(F'\) is supposed to achieve. Since I am therefore capable of perceiving something beyond the powers of \(F'\), I deduce that I cannot be \(F\) after all. Moreover, this applies to any other (Gödelizable) system, in place of \(F\).” (Penrose 1996, 3.2)

Does this argument succeed? A firm answer to this question is not appropriate to seek in the present entry. Interested readers are encouraged to consult four full-scale treatments of the argument (LaForte et. al 1998; Bringsjord and Xiao 2000; Shapiro 2003; Bowie 1982).

In addition to the Gödelian and Searlean arguments covered briefly above, a third attack on “Strong” AI (of the symbolic variety) has been widely discussed (though with the rise of statistical machine learning has come a corresponding decrease in the attention paid to it), namely, one given by the philosopher Hubert Dreyfus (1972, 1992), some incarnations of which have been co-articulated with his brother, Stuart Dreyfus (1987), a computer scientist. Put crudely, the core idea in this attack is that human expertise is not based on the explicit, disembodied, mechanical manipulation of symbolic information (such as formulae in some logic, or probabilities in some Bayesian network), and that AI’s efforts to build machines with such expertise are doomed if based on the symbolic paradigm. The genesis of the Dreyfusian attack was a belief that the critique of (if you will) symbol-based philosophy (e.g., philosophy in the logic-based, rationalist tradition, as opposed to what is called the Continental tradition) from such thinkers as Heidegger and Merleau-Ponty could be made against the rationalist tradition in AI. After further reading and study of Dreyfus’ writings, readers may judge whether this critique is compelling, in an information-driven world increasingly managed by intelligent agents that carry out symbolic reasoning (albeit not even close to the human level).

For readers interested in exploring philosophy of AI beyond what Jim Moor (in a recent address – “The Next Fifty Years of AI: Future Scientific Research vs. Past Philosophical Criticisms” – as the 2006 Barwise Award winner at the annual eastern American Philosophical Association meeting) has called “the big three” criticisms of AI, there is no shortage of additional material, much of it available on the Web. The last chapter of AIMA provides a compressed overview of some additional arguments against “Strong” AI, and is in general not a bad next step. Needless to say, Philosophy of AI today involves much more than the three well-known arguments discussed above, and, inevitably, Philosophy of AI tomorrow will include new debates and problems we can’t see now. Because machines, inevitably, will get smarter and smarter (regardless of just how smart they get), Philosophy of AI, pure and simple, is a growth industry. With every human activity that machines match, the “big” questions will only attract more attention.

If past predictions are any indication, the only thing we know today about tomorrow’s science and technology is that it will be radically different than whatever we predict it will be like. Arguably, in the case of AI, we may also specifically know today that progress will be much slower than what most expect. After all, at the 1956 kickoff conference (discussed at the start of this entry), Herb Simon predicted that thinking machines able to match the human mind were “just around the corner” (for the relevant quotes and informative discussion, see the first chapter of AIMA ). As it turned out, the new century would arrive without a single machine able to converse at even the toddler level. (Recall that when it comes to the building of machines capable of displaying human-level intelligence, Descartes, not Turing, seems today to be the better prophet.) Nonetheless, astonishing though it may be, serious thinkers in the late 20th century have continued to issue incredibly optimistic predictions regarding the progress of AI. For example, Hans Moravec (1999), in his Robot: Mere Machine to Transcendent Mind , informs us that because the speed of computer hardware doubles every 18 months (in accordance with Moore’s Law, which has apparently held in the past), “fourth generation” robots will soon enough exceed humans in all respects, from running companies to writing novels. These robots, so the story goes, will evolve to such lofty cognitive heights that we will stand to them as single-cell organisms stand to us today. [ 41 ]

Moravec is by no means singularly Pollyannaish: Many others in AI predict the same sensational future unfolding on about the same rapid schedule. In fact, at the aforementioned AI@50 conference, Jim Moor posed the question “Will human-level AI be achieved within the next 50 years?” to five thinkers who attended the original 1956 conference: John McCarthy, Marvin Minsky, Oliver Selfridge, Ray Solomonoff, and Trenchard Moore. McCarthy and Minsky gave firm, unhesitating affirmatives, and Solomonoff seemed to suggest that AI provided the one ray of hope in the face of fact that our species seems bent on destroying itself. (Selfridge’s reply was a bit cryptic. Moore returned a firm, unambiguous negative, and declared that once his computer is smart enough to interact with him conversationally about mathematical problems, he might take this whole enterprise more seriously.) It is left to the reader to judge the accuracy of such risky predictions as have been given by Moravec, McCarthy, and Minsky. [ 42 ]

The judgment of the reader in this regard ought to factor in the stunning resurgence, very recently, of serious reflection on what is known as “The Singularity,” (denoted by us simply as S ) the future point at which artificial intelligence exceeds human intelligence, whereupon immediately thereafter (as the story goes) the machines make themselves rapidly smarter and smarter and smarter, reaching a superhuman level of intelligence that, stuck as we are in the mud of our limited mentation, we can’t fathom. For extensive, balanced analysis of S , see Eden et al. (2013).

Readers unfamiliar with the literature on S may be quite surprised to learn the degree to which, among learned folks, this hypothetical event is not only taken seriously, but has in fact become a target for extensive and frequent philosophizing [for a mordant tour of the recent thought in question, see Floridi (2015)]. What arguments support the belief that S is in our future? There are two main arguments at this point: the familiar hardware-based one [championed by Moravec, as noted above, and again more recently by Kurzweil (2006)]; and the – as far as we know – original argument given by mathematician I. J. Good (1965). In addition, there is a recent and related doomsayer argument advanced by Bostrom (2014), which seems to presuppose that S will occur. Good’s argument, nicely amplified and adjusted by Chalmers (2010), who affirms the tidied-up version of the argument, runs as follows:

• Premise 1 : There will be AI (created by HI and such that AI = HI).
• Premise 2 : If there is AI, there will be AI\(^+\) (created by AI).
• Premise 3 : If there is AI\(^+\), there will be AI\(^{++}\) (created by AI\(^+\)).
• Conclusion : There will be AI\(^{++}\) (= S will occur).

In this argument, ‘AI’ is artificial intelligence at the level of, and created by, human persons, ‘AI\(^+\)’ artificial intelligence above the level of human persons, and ‘AI\(^{++}\)’ super-intelligence constitutive of S . The key process is presumably the creation of one class of machine by another. We have added for convenience ‘HI’ for human intelligence; the central idea is then: HI will create AI, the latter at the same level of intelligence as the former; AI will create AI\(^+\); AI\(^+\) will create AI\(^{++}\); with the ascension proceeding perhaps forever, but at any rate proceeding long enough for us to be as ants outstripped by gods.

The argument certainly appears to be formally valid. Are its three premises true? Taking up such a question would fling us far beyond the scope of this entry. We point out only that the concept of one class of machines creating another, more powerful class of machines is not a transparent one, and neither Good nor Chalmers provides a rigorous account of the concept, which is ripe for philosophical analysis. (As to mathematical analysis, some exists, of course. It is for example well-known that a computing machine at level \(L\) cannot possibly create another machine at a higher level \(L'\). For instance, a linear-bounded automaton can’t create a Turing machine.)

The Good-Chalmers argument has a rather clinical air about it; the argument doesn’t say anything regarding whether machines in the AI\(^{++}\) category will be benign, malicious, or munificent. Many others gladly fill this gap with dark, dark pessimism. The locus classicus here is without question a widely read paper by Bill Joy (2000): “Why The Future Doesn’t Need Us.” Joy believes that the human race is doomed, in no small part because it’s busy building smart machines. He writes:

The 21st-century technologies – genetics, nanotechnology, and robotics (GNR) – are so powerful that they can spawn whole new classes of accidents and abuses. Most dangerously, for the first time, these accidents and abuses are widely within the reach of individuals or small groups. They will not require large facilities or rare raw materials. Knowledge alone will enable the use of them. Thus we have the possibility not just of weapons of mass destruction but of knowledge-enabled mass destruction (KMD), this destructiveness hugely amplified by the power of self-replication. I think it is no exaggeration to say we are on the cusp of the further perfection of extreme evil, an evil whose possibility spreads well beyond that which weapons of mass destruction bequeathed to the nation-states, on to a surprising and terrible empowerment of extreme individuals. [ 43 ]

Philosophers would be most interested in arguments for this view. What are Joy’s? Well, no small reason for the attention lavished on his paper is that, like Raymond Kurzweil (2000), Joy relies heavily on an argument given by none other than the Unabomber (Theodore Kaczynski). The idea is that, assuming we succeed in building intelligent machines, we will have them do most (if not all) work for us. If we further allow the machines to make decisions for us – even if we retain oversight over the machines –, we will eventually depend on them to the point where we must simply accept their decisions. But even if we don’t allow the machines to make decisions, the control of such machines is likely to be held by a small elite who will view the rest of humanity as unnecessary – since the machines can do any needed work (Joy 2000).

This isn’t the place to assess this argument. (Having said that, the pattern pushed by the Unabomber and his supporters certainly appears to be flatly invalid. [ 44 ] ) In fact, many readers will doubtless feel that no such place exists or will exist, because the reasoning here is amateurish. So then, what about the reasoning of professional philosophers on the matter?

Bostrom has recently painted an exceedingly dark picture of a possible future. He points out that the “first superintelligence” could have the capability

to shape the future of Earth-originating life, could easily have non-anthropomorphic final goals, and would likely have instrumental reasons to pursue open-ended resource acquisition. If we now reflect that human beings consist of useful resources (such as conveniently located atoms) and that we depend on many more local resources, we can see that the outcome could easily be one in which humanity quickly becomes extinct. (Bostrom 2014, p. 416)

Clearly, the most vulnerable premise in this sort of argument is that the “first superintelligence” will arrive indeed arrive. Here perhaps the Good-Chalmers argument provides a basis.

Searle (2014) thinks Bostrom’s book is misguided and fundamentally mistaken, and that we needn’t worry. His rationale is dirt-simple: Machines aren’t conscious; Bostrom is alarmed at the prospect of malicious machines who do us in; a malicious machine is by definition a conscious machine; ergo, Bostrom’s argument doesn’t work. Searle writes:

If the computer can fly airplanes, drive cars, and win at chess, who cares if it is totally nonconscious? But if we are worried about a maliciously motivated superintelligence destroying us, then it is important that the malicious motivation should be real. Without consciousness, there is no possibiity of its being real.

The positively remarkable thing here, it seems to us, is that Searle appears to be unaware of the brute fact that most AI engineers are perfectly content to build machines on the basis of the AIMA view of AI we presented and explained above: the view according to which machines simply map percepts to actions. On this view, it doesn’t matter whether the machine really has desires; what matters is whether it acts suitably on the basis of how AI scientists engineer formal correlates to desire. An autonomous machine with overwhelming destructive power that non-consciously “decides” to kill doesn’t become just a nuisance because genuine, human-level, subjective desire is absent from the machine. If an AI can play the game of chess, and the game of Jeopardy! , it can certainly play the game of war. Just as it does little good for a human loser to point out that the victorious machine in a game of chess isn’t conscious, it will do little good for humans being killed by machines to point out that these machines aren’t conscious. (It is interesting to note that the genesis of Joy’s paper was an informal conversation with John Searle and Raymond Kurzweil. According to Joy, Searle didn’t think there was much to worry about, since he was (and is) quite confident that tomorrow’s robots can’t be conscious. [ 45 ] )

There are some things we can safely say about tomorrow. Certainly, barring some cataclysmic events (nuclear or biological warfare, global economic depression, a meteorite smashing into Earth, etc.), we now know that AI will succeed in producing artificial animals . Since even some natural animals (mules, e.g.) can be easily trained to work for humans, it stands to reason that artificial animals, designed from scratch with our purposes in mind, will be deployed to work for us. In fact, many jobs currently done by humans will certainly be done by appropriately programmed artificial animals. To pick an arbitrary example, it is difficult to believe that commercial drivers won’t be artificial in the future. (Indeed, Daimler is already running commercials in which they tout the ability of their automobiles to drive “autonomously,” allowing human occupants of these vehicles to ignore the road and read.) Other examples would include: cleaners, mail carriers, clerical workers, military scouts, surgeons, and pilots. (As to cleaners, probably a significant number of readers, at this very moment, have robots from iRobot cleaning the carpets in their homes.) It is hard to see how such jobs are inseparably bound up with the attributes often taken to be at the core of personhood – attributes that would be the most difficult for AI to replicate. [ 46 ]

Andy Clark (2003) has another prediction: Humans will gradually become, at least to an appreciable degree, cyborgs, courtesy of artificial limbs and sense organs, and implants. The main driver of this trend will be that while standalone AIs are often desirable, they are hard to engineer when the desired level of intelligence is high. But to let humans “pilot” less intelligent machines is a good deal easier, and still very attractive for concrete reasons. Another related prediction is that AI would play the role of a cognitive prosthesis for humans (Ford et al. 1997; Hoffman et al. 2001). The prosthesis view sees AI as a “great equalizer” that would lead to less stratification in society, perhaps similar to how the Hindu-Arabic numeral system made arithmetic available to the masses, and to how the Guttenberg press contributed to literacy becoming more universal.

Even if the argument is formally invalid, it leaves us with a question – the cornerstone question about AI and the future: Will AI produce artificial creatures that replicate and exceed human cognition (as Kurzweil and Joy believe)? Or is this merely an interesting supposition?

This is a question not just for scientists and engineers; it is also a question for philosophers. This is so for two reasons. One, research and development designed to validate an affirmative answer must include philosophy – for reasons rooted in earlier parts of the present entry. (E.g., philosophy is the place to turn to for robust formalisms to model human propositional attitudes in machine terms.) Two, philosophers might well be able to provide arguments that answer the cornerstone question now, definitively. If a version of either of the three arguments against “Strong” AI alluded to above (Searle’s CRA; the Gödelian attack; the Dreyfus argument) are sound, then of course AI will not manage to produce machines having the mental powers of persons. No doubt the future holds not only ever-smarter machines, but new arguments pro and con on the question of whether this progress can reach the human level that Descartes declared to be unreachable.

• Adams, E. W., 1996, A Primer of Probability Logic , Stanford, CA: CSLI.
• Almeida, J., Frade, M., Pinto, J. & de Sousa, S., 2011, Rigorous Software Development: An Introduction to Program Verification , New York, NY: Spinger.
• Alpaydin, E., 2014, Introduction to Machine Learning , Cambridge, MA: MIT Press.
• Amir, E. & Maynard-Reid, P., 1999, “Logic-Based Subsumption Architecture,” in Proceedings of the 16 th International Joint Conference on Artificial Intelligence (IJCAI-1999) , (San Francisco, CA: MIT Morgan Kaufmann), pp. 147–152.
• Amir, E. & Maynard-Reid, P., 2000, “Logic-Based Subsumption Architecture: Empirical Evaluation,” in Proceedings of the AAAI Fall Symposium on Parallel Architectures for Cognition .
• Amir, E. & Maynard-Reid, P., 2001, “LiSA: A Robot Driven by Logical Subsumption,” in Proceedings of the Fifth Symposium on the Logical Formalization of Commonsense Reasoning , (New York, NY).
• Anderson, C. A., 1983, “The Paradox of the Knower,” The Journal of Philosophy , 80.6: 338–355.
• Anderson, J. & Lebiere, C., 2003, “The Newell Test for a Theory of Cognition,” Behavioral and Brain Sciences , 26: 587–640.
• Ashcraft, M., 1994, Human Memory and Cognition , New York, NY: HarperCollins.
• Arkin, R., 2009, Governing Lethal Behavior in Autonomous Robots , London: Chapman and Hall/CRC Imprint, Taylor and Francis Group.
• Arkoudas, K. & Bringsjord, S., 2005, “Vivid: A Framework for Heterogeneous Problem Solving,” Artificial Intelligence , 173.15: 1367–1405.
• Arkoudas, K. & Bringsjord, S., 2005, “Metareasoning for Multi-agent Epistemic Logics,” in Fifth International Conference on Computational Logic In Multi-Agent Systems (CLIMA 2004) , in the series Lecture Notes in Artificial Intelligence (LNAI) , volume 3487, New York, NY: Springer-Verlag, pp. 111–125.
• Arkoudas, K., 2000, Denotational Proof Languages , PhD dissertation, Massachusetts Institute of Technology (Computer Science).
• Baader, F., Calvanese, D., McGuinness, D. L., Nardi, D., & Patel-Schneider, P. F., eds., 2003, The Description Logic Handbook: Theory, Implementation, and Applications , New York, NY: Cambridge University Press.
• Smith, B., Ashburner, M., Rosse, C., Bard, J., Bug, W., Ceusters, W., Goldberg, L. J., Eilbeck, K., Ireland, A., Mungall, C. J., The OBI Consortium, Leontis, N., Rocca-Serra, P., Ruttenberg, A., Sansone, S., Scheuermann, R. H., Shah, N., Whetzel, P. L. & Lewis, S., 2007, “The OBO Foundry: Coordinated Evolution of Ontologies to Support Biomedical Data Integration,” Nature Biotechnology 25, 1251–1255.
• Barwise, J. & Etchemendy, J., 1999, Language, Proof, and Logic , New York, NY: Seven Bridges Press.
• Barwise, J. & Etchemendy, J., 1995, “Heterogeneous Logic,” in Diagrammatic Reasoning: Cognitive and Computational Perspectives , J. Glasgow, N.H. Narayanan, & B. Chandrasekaran, eds., Cambridge, MA: MIT Press, pp. 211–234.
• Baldi, P., Sadowski P. & Whiteson D., 2014, “Searching for Exotic Particles in High-energy Physics with Deep Learning,” Nature Communications . [ Available online ]
• Barwise, J. & Etchemendy, J., 1994, Hyperproof , Stanford, CA: CSLI.
• Barwise, J. & Etchemendy, J., 1990, “Infons and Inference,” in Situation Theory and its Applications, (Vol 1) , Cooper, Mukai, and Perry (eds), CSLI Lecture Notes #22, CSLI Press, pp. 33–78.
• Bello, P. & Bringsjord S., 2013, “On How to Build a Moral Machine,” Topoi , 32.2: 251–266.
• Bengio, Y., Goodfellow, I., & Courville, A., 2016, Deep Learning , Cambridge: MIT Press. [ Available online ]
• Bengio, Y., Courville, A. & Vincent, P., 2013, “Representation Learning: A Review and New Perspectives,” Pattern Analysis and Machine Intelligence, IEEE Transactions , 35.8: 1798–1828.
• Berners-Lee, T., Hendler, J. & Lassila, O., 2001, “The Semantic Web,” Scientific American , 284: 34–43.
• Bishop, M. & Preston, J., 2002, Views into the Chinese Room: New Essays on Searle and Artificial Intelligence , Oxford, UK: Oxford University Press.
• Boden, M., 1994, “Creativity and Computers,” in Artificial Intelligence and Computers , T. Dartnall, ed., Dordrecht, The Netherlands: Kluwer, pp. 3–26.
• Boolos, G. S., Burgess, J.P., & Jeffrey., R.C., 2007, Computability and Logic 5th edition , Cambridge: Cambridge University Press.
• Bostrom, N., 2014, Superintelligence: Paths, Dangers, Strategies , Oxford, UK: Oxford University Press.
• Bowie, G.L., 1982, “Lucas’ Number is Finally Up,” Journal of Philosophical Logic , 11: 279–285.
• Brachman, R. & Levesque, H., 2004, Knowledge Representation and Reasoning , San Francisco, CA: Morgan Kaufmann/Elsevier.
• Bringsjord, S., Arkoudas K. & Bello P., 2006, “Toward a General Logicist Methodology for Engineering Ethically Correct Robots,” IEEE Intelligent Systems, 21.4: 38–44.
• Bringsjord, S. & Ferrucci, D., 1998, “Logic and Artificial Intelligence: Divorced, Still Married, Separated…?” Minds and Machines , 8: 273–308.
• Bringsjord, S. & Schimanski, B., 2003, “What is Artificial Intelligence? Psychometric AI as an Answer,” Proceedings of the 18 th International Joint Conference on Artificial Intelligence (IJCAI-2003) , (San Francisco, CA: MIT Morgan Kaufmann), pp. 887–893.
• Bringsjord, S. & Ferrucci, D., 2000, Artificial Intelligence and Literary Creativity: Inside the Mind of Brutus, a Storytelling Machine , Mahwah, NJ: Lawrence Erlbaum.
• Bringsjord, S. & van Heuveln, B., 2003, “The Mental Eye Defense of an Infinitized Version of Yablo’s Paradox,” Analysis 63.1: 61–70.
• Bringsjord S. & Xiao, H., 2000, “A Refutation of Penrose’s Gödelian Case Against Artificial Intelligence,” Journal of Experimental and Theoretical Artificial Intelligence , 12: 307–329.
• Bringsjord, S. & Zenzen, M., 2002, “Toward a Formal Philosophy of Hypercomputation,” Minds and Machines , 12: 241–258.
• Bringsjord, S., 2000, “Animals, Zombanimals, and the Total Turing Test: The Essence of Artificial Intelligence,” Journal of Logic, Language, and Information , 9: 397–418.
• Bringsjord, S., 1998, “Philosophy and ‘Super’ Computation,” The Digital Phoenix: How Computers are Changing Philosophy , J. Moor and T. Bynam, eds., Oxford, UK: Oxford University Press, pp. 231–252.
• Bringsjord, S., 1991, “Is the Connectionist-Logicist Clash one of AI’s Wonderful Red Herrings?” Journal of Experimental & Theoretical AI , 3.4: 319–349.
• Bringsjord, S., Govindarajulu N. S., Eberbach, E. & Yang, Y., 2012, “Perhaps the Rigorous Modeling of Economic Phenomena Requires Hypercomputation,” International Journal of Unconventional Computing , 8.1: 3–32. [ Preprint available online ]
• Bringsjord, S., 2011, “Psychometric Artificial Intelligence,” Journal of Experimental and Theoretical Artificial Intelligence , 23.3: 271–277.
• Bringsjord, S. & Govindarajulu N. S., 2012, “Given the Web, What is Intelligence, Really?” Metaphilosophy 43.12: 464–479.
• Brooks, R. A., 1991, “Intelligence Without Representation,” Artificial Intelligence , 47: 139–159.
• Browne, C. B., Powley, E. & Whitehouse, D., 2012, “A Survey of Monte Carlo Tree Search Methods,” A Survey of Monte Carlo Tree Search Methods , 4.1: 1–43.
• Buchanan, B. G., 2005, “A (Very) Brief History of Artificial Intelligence,” AI Magazine , 26.4: 53–60.
• Carroll, L., 1958, Symbolic Logic; Game of Logic , New York, NY: Dover.
• Cassimatis, N., 2006, “Cognitive Substrate for Human-Level Intelligence,” AI Magazine , 27.2: 71–82.
• Chajed, T., Chen, H., Chlipala, A., Kaashoek, F., Zeldovich, N., & Ziegler, D., 2017, “Research Highlight: Certifying a File System using Crash Hoare Logic: Correctness in the Presence of Crashes,” Communications of the ACM (CACM) , 60.4: 75–84.
• Chalmers, D., 2010, “The Singularity: A Philosophical Analysis,” Journal of Consciousness Studies , 17: 7–65.
• Charniak, E., 1993, Statistical Language Learning , Cambridge: MIT Press.
• Charniak, E. & McDermott, D., 1985, Introduction to Artificial Intelligence , Reading, MA: Addison Wesley.
• Chellas, B., 1980, Modal Logic: An Introduction , Cambridge, UK: Cambridge University Press.
• Chisholm, R., 1957, Perceiving , Ithaca, NY: Cornell University Press.
• Chisholm, R., 1966, Theory of Knowledge , Englewood Cliffs, NJ: Prentice-Hall.
• Chisholm, R., 1977, Theory of Knowledge 2nd ed , Englewood Cliffs, NJ: Prentice-Hall.
• Clark, A., 2003, Natural-Born Cyborgs , Oxford, UK: Oxford University Press.
• Clark, M. H., 2010, Cognitive Illusions and the Lying Machine: A Blueprint for Sophistic Mendacity , PhD dissertation, Rensselaer Polytechnic Institute (Cognitive Science).
• Copeland, B. J., 1998, “Super Turing Machines,” Complexity , 4: 30–32.
• Copi, I. & Cohen, C., 2004, Introduction to Logic , Saddle River, NJ: Prentice-Hall.
• Dennett, D., 1998, “Artificial Life as Philosophy,” in his Brainchildren: Essays on Designing Minds , Cambridge, MA: MIT Press, pp. 261–263.
• Dennett, D., 1994, “The Practical Requirements for Making a Conscious Robot,” Philosophical Transactions of the Royal Society of London , 349: 133–146.
• Dennett, D., 1979, “Artificial Intelligence as Philosophy and as Psychology,” Philosophical Perspectives in Artificial Intelligence , M. Ringle, ed., Atlantic Highlands, NJ: Humanities Press, pp. 57–80.
• Descartes, 1637, R., in Haldane, E. and Ross, G.R.T., translators, 1911, The Philosophical Works of Descartes, Volume 1 , Cambridge, UK: Cambridge University Press.
• Dick, P. K., 1968, Do Androids Dream of Electric Sheep? , New York, NY: Doubleday.
• Domingos, P., 2012, “A Few Useful Things to Know about Machine Learning,” Communications of the ACM , 55.10: 78–87.
• Dreyfus, H., 1972, What Computers Can’t Do , Cambridge, MA: MIT Press.
• Dreyfus, H., 1992, What Computers Still Can’t Do , Cambridge, MA: MIT Press.
• Dreyfus, H. & Dreyfus, S., 1987, Mind Over Machine: The Power of Human Intuition and Expertise in the Era of the Computer , New York, NY: Free Press.
• Ebbinghaus, H., Flum, J. & Thomas, W., 1984, Mathematical Logic , New York, NY: Springer-Verlag.
• Eden, A., Moor, J., Soraker, J. & Steinhart, E., 2013, Singularity Hypotheses: A Scientific and Philosophical Assessment , New York, NY: Springer.
• España-Bonet, C., Enache, R., Slaski, A., Ranta, A., Màrquez L. & Gonzàlez, M., 2011, “Patent Translation within the MOLTO project,” in Proceedings of the 4th Workshop on Patent Translation, MT Summit XIII , pp. 70–78.
• Evans, G., 1968, “A Program for the Solution of a Class of Geometric-Analogy Intelligence-Test Questions,” in M. Minsky, ed., Semantic Information Processing , Cambridge, MA: MIT Press, pp. 271–353.
• Fagin, R., Halpern, J. Y., Moses, Y. & Vardi, M., 2004, Reasoning About Knowledge , Cambridge, MA: MIT Press.
• Ferrucci, D. & Lally, A., 2004, “UIMA: An Architectural Approach to Unstructured Information Processing in the Corporate Research Environment,” Natural Language Engineering , 10.3–4: 327–348. Cambridge, UK: Cambridge University Press.
• Ferrucci, D., Brown, E., Chu-Carroll, J., Fan, J., Gondek, D., Kalyanpur, A., Lally, A., Murdock, J., Nyberg, E., Prager, J., Schlaefer, N. & Welty, C., 2010, “Building Watson: An Overview of the DeepQA Project,” AI Magazine , 31.3: 59–79.
• Finnsson, H., 2012, “Generalized Monte-Carlo Tree Search Extensions for General Game Playing,” in Proceedings of the Twenty-Sixth AAAI Conference on Artificial Intelligence (AAAI-2012) , Toronto, Canda, pp. 1550–1556.
• Fitelson, B., 2005, “Inductive Logic,” in Pfeifer, J. and Sarkar, S., eds., Philosophy of Science: An Encyclopedia , London, UK: Routledge, pp. 384–394.
• Floridi, L., 2015, “Singularitarians, AItheists, and Why the Problem with Artificial Intelligence is H.A.L. (Humanity At Large), not HAL,” APA Newsletter: Philosophy and Computers , 14.2: 8–11.
• Foot, P., 1967, “The Problem of Abortion and the Doctrine of the Double Effect,” Oxford Review , 5: 5–15.
• Forbus, K. D. & Hinrichs, T. R., 2006, “Companion Cognitive Systems: A Step toward Human-Level AI,” AI Magazine , 27.2: 83.
• Ford, K. M., Glymour C. & Hayes P., 1997, “On the Other Hand … Cognitive Prostheses,” AI Magazine , 18.3: 104.
• Friedland, N., Allen, P., Matthews, G., Witbrock, M., Baxter, D., Curtis, J., Shepard, B., Miraglia, P., Angele, J., Staab, S., Moench, E., Oppermann, H., Wenke, D., Israel, D., Chaudhri, V., Porter, B., Barker, K., Fan, J., Yi Chaw, S., Yeh, P., Tecuci, D. & Clark, P., 2004, “Project Halo: Towards a Digital Aristotle,” AI Magazine , 25.4: 29–47.
• Genesereth, M., Love, N. & Pell B., 2005, “General Game Playing: Overview of the AAAI Competition,” AI Magazine , 26.2: 62–72. [ Available online ]
• Ginsberg, M., 1993, Essentials of Artificial Intelligence , New York, NY: Morgan Kaufmann.
• Glymour, G., 1992, Thinking Things Through , Cambridge, MA: MIT Press.
• Goertzel, B. & Pennachin, C., eds., 2007, Artificial General Intelligence , Berlin, Heidelberg: Springer-Verlag.
• Gold, M., 1965, “Limiting Recursion,” Journal of Symbolic Logic , 30.1: 28–47.
• Goldstine, H. & von Neumann, J., 1947, “Planning and Coding of Problems for an Electronic Computing Instrument,” IAS Reports Institute for Advanced Study, Princeton, NJ. [This remarkable work is available online from the Institute for Advanced Study. Please note that this paper is Part II of a three-volume set. The first volume was devoted to a preliminary discussion, and the first author on it was Arthur Burks, joining Goldstine and von Neumann.]
• Good, I., 1965, “Speculations Concerning the First Ultraintelligent Machines,” in Advances in Computing (vol. 6), F. Alt and M. Rubinoff, eds., New York, NY: Academic Press, pp. 31–38.
• Govindarajulu, N. S., Bringsjord, S. & Licato J., 2013, “On Deep Computational Formalization of Natural Language,” in Proceedings of the Workshop “Formalizing Mechanisms for Artificial General Intelligence and Cognition (Formal MAGiC),” Osnabrück, Germany: PICS.
• Govindarajulu, N. S., & Bringsjord, S., 2015, “Ethical Regulation of Robots Must Be Embedded in Their Operating Systems” in Trappl, R., ed., A Construction Manual for Robot’s Ethical Systems: Requirements, Methods, Implementations , Berlin, DE: Springer.
• Govindarajulu, N. S., & Bringsjord, S., 2017, “On Automating the Doctrine of Double Effect,” in Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI-17) , pp. 4722–4730. doi:10.24963/ijcai.2017/658
• Granger, R., 2004a, “Derivation and Analysis of Basic Computational Operations of Thalamocortical Circuits,” Journal of Cognitive Neuroscience 16: 856–877.
• Granger, R., 2004b, “Brain Circuit Implementation: High-precision Computation from Low-Precision Components,” in Toward Replacement Parts for the Brain , T. Berger and D. Glanzman, eds., Cambridge, MA: MIT Press, pp. 277–294.
• Griewank, A., 2000, Evaluating Derivatives: Principles and Techniques of Algorithmic Differentiation , Philadlphia, PA: Society for Industrial and Applied Mathematics (SIAM).
• Guizzo, E., 2011, “How Google’s Self-driving Car Works,” IEEE Spectrum Online . [ Available online]
• Hailperin, T., 1996, Sentential Probability Logic: Origins, Development, Current Status, and Technical Applications , Bethlehem, United States: Lehigh University Press.
• Hailperin, T., 2010, Logic with a Probability Semantics , Bethlehem, United States: Lehigh University Press.
• Halpern, J. Y., 1990, “An Analysis of First-order Logics of Probability,” Artificial Intelligence , 46: 311–350.
• Halpern, J., Harper, R., Immerman, N., Kolaitis, P. G., Vardi, M. & Vianu, V., 2001, “On the Unusual Effectiveness of Logic in Computer Science,” The Bulletin of Symbolic Logic , 7.2: 213–236.
• Hamkins, J. & Lewis, A., 2000, “Infinite Time Turing Machines,” Journal of Symbolic Logic , 65.2: 567–604.
• Harnad, S., 1991, “Other Bodies, Other Minds: A Machine Incarnation of an Old Philosophical Problem,” Minds and Machines , 1.1: 43–54.
• Haugeland, J., 1985, Artificial Intelligence: The Very Idea , Cambridge, MA: MIT Press.
• Hendler, J. & Jennifer G., 2008, “Metcalfe’s Law, Web 2.0, and the Semantic Web,” Web Semantics: Science, Services and Agents on the World Wide Web , 6.1: 14–20.
• Hinton, G., Deng, L., Yu, D., Dahl, G.E., Mohamed, A. R., Jaitly, N., Senior, A., Vanhoucke, V., Nguyen, P., Sainath, T. & Kingsbury, B., 2012, “Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups,” IEEE Signal Processing Magazine , 29.6: 82–97.
• Hoffman, R. R., Hayes, P. J. & Ford, K. M., 2001, “Human-Centered Computing: Thinking In and Out of the Box,” IEEE Intelligent Systems , 16.5: 76–78.
• Hoffman, R. R., Bradshaw J. M., Hayes P. J. & Ford K. M., 2003, “ The Borg Hypothesis,” IEEE Intelligent Systems , 18.5: 73–75.
• Hofstadter, D. & McGraw, G., 1995, “Letter Spirit: Esthetic Perception and Creative Play in the Rich Microcosm of the Roman Alphabet,” in Hofstadter’s Fluid Concepts and Creative Analogies: Computer Models of the Fundamental Mechanisms of Thought , New York, NY: Basic Books, pp. 407–488.
• Hornik, K., Stinchcombe, M. & White, H., 1989, “Multilayer Feedforward Networks are Universal Approximators,” Neural Networks , 2.5: 359–366.
• Hutter, M., 2005, Universal Artificial Intelligence , Berlin: Springer.
• Joy, W., 2000, “Why the Future Doesn’t Need Us,” Wired 8.4. [ Available online ]
• Kahneman, D., 2013. Thinking, Fast and Slow , New York, NY: Farrar, Straus, and Giroux.
• Kaufmann, M., Manolios, P. & Moore, J. S., 2000, Computer-Aided Reasoning: ACL2 Case Studies , Dordrecht, The Netherlands: Kluwer Academic Publishers.
• Klenk, M., Forbus, K., Tomai, E., Kim,H. & Kyckelhahn, B., 2005, “Solving Everyday Physical Reasoning Problems by Analogy using Sketches,” in Proceedings of 20th National Conference on Artificial Intelligence (AAAI-05), Pittsburgh, PA.
• Kolata, G., 1996, “Computer Math Proof Shows Reasoning Power,” in New York Times . [ Availabe online ]
• Koller, D., Levy, A. & Pfeffer, A., 1997, “P-CLASSIC: A Tractable Probablistic Description Logic,” in Proceedings of the AAAI 1997 Meeting , 390–397.
• Kurzweil, R., 2006, The Singularity Is Near: When Humans Transcend Biology , New York, NY: Penguin USA.
• Kurzweil, R., 2000, The Age of Spiritual Machines: When Computers Exceed Human Intelligence , New York, NY: Penguin USA.
• LaForte, G., Hayes P. & Ford, K., 1998, “Why Gödel’s Theorem Cannot Refute Computationslism,” Artificial Intelligence , 104: 265–286.
• Laird, J. E., 2012, The Soar Cognitive Architecture , Cambridge, MA: MIT Press.
• Laird, J. & VanLent M., 2001, “Human-level AI’s Killer Application: Interactive Computer Games,” AI Magazine 22.2:15–26.
• LeCun, Y., Bengio, Y. & Hinton G., 2015, “Deep Learning,” Nature , 521: 436–444.
• Lenat, D., 1983, “EURISKO: A Program that Learns New Heuristics and Domain Concepts,” Artificial Intelligence , 21(1-2): 61–98. doi:10.1016/s0004-3702(83)80005-8
• Lenat, D., & Guha, R. V., 1990, Building Large Knowledge-Based Systems: Representation and Inference in the Cyc Project , Reading, MA: Addison Wesley.
• Lenzen, W., 2004, “Leibniz’s Logic,” in Gabbay, D., Woods, J. and Kanamori, A., eds., Handbook of the History of Logic , Elsevier, Amsterdam, The Netherlands, pp. 1–83.
• Lewis, H. & Papadimitriou, C., 1981, Elements of the Theory of Computation , Prentice Hall, Englewood Cliffs, NJ: Prentice Hall.
• Litt, A., Eliasmith, C., Kroon, F., Weinstein, S. & Thagard, P., 2006, “Is the Brain a Quantum Computer?” Cognitive Science 30: 593–603.
• Lucas, J. R., 1964, “Minds, Machines, and Gödel,” in Minds and Machines , A. R. Anderson, ed., Prentice-Hall, NJ: Prentice-Hall, pp. 43–59.
• Luger, G., 2008, Artificial Intelligence: Structures and Strategies for Complex Problem Solving , New York, NY: Pearson.
• Luger, G. & Stubblefield, W., 1993, Artificial Intelligence: Structures and Strategies for Complex Problem Solving , Redwood, CA: Benjamin Cummings.
• Lopez, A., 2008, “Statistical Machine Translation,” ACM Computing Surveys , 40.3: 1–49.
• Malle, B. F., Scheutz, M., Arnold, T., Voiklis, J. & Cusimano, C., 2015, “Sacrifice One For the Good of Many?: People Apply Different Moral Norms to Human and Robot Agents,” in Proceedings of the Tenth Annual ACM/IEEE International Conference on Human-Robot Interaction (HRI ’15) (New York, NY: ACM), pp. 117–124.
• Manzano, M., 1996, Extensions of First Order Logic , Cambridge, UK: Cambridge University Press.
• Marcus, G., 2013, “Why Can’t My Computer Understand Me?,” in The New Yorker , August 2013. [ Available online ]
• McCarthy, J. & Hayes, P., 1969, “Some Philosophical Problems from the Standpoint of Artificial Intelligence,” in Machine Intelligence 4 , B. Meltzer and D. Michie, eds., Edinburgh: Edinburgh University Press, 463–502.
• Mueller, E., 2006, Commonsense Reasoning , San Francisco, CA: Morgan Kaufmann.
• Murphy, K. P., 2012, Machine Learning: A Probabilistic Perspective , Cambridge, MA: MIT Press.
• Minsky, M. & Pappert, S., 1969, Perceptrons: An Introduction to Computational Geometry , Cambridge, MA: MIT Press.
• Montague, R., 1970, “Universal Grammar,” Theoria , 36, 373–398.
• Moor, J., 2006, “The Nature, Importance, and Difficulty of Machine Ethics”, IEEE Intelligent Systems 21.4: 18–21.
• Moor, J., 1985, “What is Computer Ethics?” Metaphilosophy 16.4: 266–274.
• Moor, J., ed., 2003, The Turing Test: The Elusive Standard of Artificial Intelligence , Dordrecht, The Netherlands: Kluwer Academic Publishers.
• Moravec, H., 1999, Robot: Mere Machine to Transcendant Mind , Oxford, UK: Oxford University Press,
• Naumowicz, A. & Kornilowicz., A., 2009, “A Brief Overview of Mizar,” in Theorem Proving in Higher Order Logics , S. Berghofer, T. Nipkow, C. Urban & M. Wenzel, eds., Berlin: Springer, pp. 67–72.
• Newell, N., 1973, “You Can’t Play 20 Questions with Nature and Win: Projective Comments on the Papers of this Symposium”, in Visual Information Processing , W. Chase, ed., New York, NY: Academic Press, pp. 283–308.
• Nilsson, N., 1998, Artificial Intelligence: A New Synthesis , San Francisco, CA: Morgan Kaufmann.
• Nilsson, N., 1987, Principles of Artificial Intelligence , New York, NY: Springer-Verlag.
• Nilsson, N., 1991, “Logic and Artificial Intelligence,” Artificial Intelligence , 47: 31–56.
• Nozick, R., 1970, “Newcomb’s Problem and Two Principles of Choice,” in Essays in Honor of Carl G. Hempel , N. Rescher, ed., Highlands, NJ: Humanities Press, pp. 114–146. This appears to be the very first published treatment of NP – though the paradox goes back to its creator: William Newcomb, a physicist.
• Osherson, D., Stob, M. & Weinstein, S., 1986, Systems That Learn , Cambridge, MA: MIT Press.
• Pearl, J., 1988, Probabilistic Reasoning in Intelligent Systems , San Mateo, CA: Morgan Kaufmann.
• Pennington, J., Socher R., & Manning C. D., 2014, “GloVe: Global Vectors for Word Representation,” in Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP 2014) , pp. 1532–1543. [ Available online ]
• Penrose, R., 1989, The Emperor’s New Mind , Oxford, UK: Oxford University Press.
• Penrose, R., 1994, Shadows of the Mind , Oxford, UK: Oxford University Press.
• Penrose, R., 1996, “Beyond the Doubting of a Shadow: A Reply to Commentaries on Shadows of the Mind ,” Psyche , 2.3. This paper is available online.
• Pereira, L., & Saptawijaya A., 2016, Programming Machine Ethics , Berlin, Germany: Springer
• Pinker, S., 1997, How the Mind Works , New York, NY: Norton.
• Pollock, J., 2006, Thinking about Acting: Logical Foundations for Rational Decision Making , Oxford, UK: Oxford University Press.
• Pollock, J., 2001, “Defeasible Reasoning with Variable Degrees of Justification,” Artificial Intelligence , 133, 233–282.
• Pollock, J., 1995, Cognitive Carpentry: A Blueprint for How to Build a Person , Cambridge, MA: MIT Press.
• Pollock, J., 1992, “How to Reason Defeasibly,” Artificial Intelligence , 57, 1–42.
• Pollock, J., 1989, How to Build a Person: A Prolegomenon , Cambridge, MA: MIT Press.
• Pollock, J., 1974, Knowledge and Justification , Princeton, NJ: Princeton University Press.
• Pollock, J., 1967, “Criteria and our Knowledge of the Material World,” Philosophical Review , 76, 28–60.
• Pollock, J., 1965, Analyticity and Implication , PhD dissertation, University of California at Berkeley (Philosophy).
• Potter, M.D., 2004, Set Theory and its Philosophy , Oxford, UK: Oxford University Press
• Preston, J. & Bishop, M., 2002, Views into the Chinese Room: New Essays on Searle and Artificial Intelligence , Oxford, UK: Oxford University Press.
• Putnam, H., 1965, “Trial and Error Predicates and a Solution to a Problem of Mostowski,” Journal of Symbolic Logic , 30.1, 49–57.
• Putnam, H., 1963, “Degree of Confirmation and Inductive Logic,” in The Philosophy of Rudolf Carnap , Schilipp, P., ed., Open Court, pp. 270–292.
• Rajat, R., Anand, M. & Ng, A. Y., 2009, “Large-scale Deep Unsupervised Learning Using Graphics Processors,” in Proceedings of the 26th Annual International Conference on Machine Learning , ACM, pp. 873–880.
• Rapaport, W., 1988, “Syntactic Semantics: Foundations of Computational Natural-Language Understanding,” in Aspects of Artificial Intelligence , J. H. Fetzer ed., Dordrecht, The Netherlands: Kluwer Academic Publishers, 81–131.
• Rapaport, W. & Shapiro, S., 1999, “Cognition and Fiction: An Introduction,” Understanding Language Understanding: Computational Models of Reading , A. Ram & K. Moorman, eds., Cambridge, MA: MIT Press, 11–25. [ Available online ]
• Reeke, G. & Edelman, G., 1988, “Real Brains and Artificial Intelligence,” in The Artificial Intelligence Debate: False Starts, Real Foundations , Cambridge, MA: MIT Press, pp. 143–173.
• Richardson, M. & Domingos, P., 2006, “Markov Logic Networks,” Machine Learning , 62.1–2:107–136.
• Robertson, G. & Watson, I., 2015, “A Review of Real-Time Strategy Game AI,” AI Magazine , 35.4: 75–104.
• Rosenschein, S. & Kaelbling, L., 1986, “The Synthesis of Machines with Provable Epistemic Properties,” in Proceedings of the 1986 Conference on Theoretical Aspects of Reasoning About Knowledge , San Mateo, CA: Morgan Kaufmann, pp. 83–98.
• Rumelhart, D. & McClelland, J., 1986, eds., Parallel Distributed Processing , Cambridge, MA: MIT Press.
• Russell, S., 1997, “Rationality and Intelligence,” Artificial Intelligence , 94: 57–77. [ Version available online from author ]
• Russell, S. & Norvig, P., 1995, Artificial Intelligence: A Modern Approach , Saddle River, NJ: Prentice Hall.
• Russell, S. & Norvig, P., 2002, Artificial Intelligence: A Modern Approach 2nd edition , Saddle River, NJ: Prentice Hall.
• Russell, S. & Norvig, P., 2009, Artificial Intelligence: A Modern Approach 3rd edition , Saddle River, NJ: Prentice Hall.
• Rychtyckyj, N. & Plesco, C., 2012, “Applying Automated Language Translation at a Global Enterprise Level,” AI Magazine , 34.1: 43–54.
• Scanlon, T. M., 1982, “Contractualism and Utilitarianism,” in A. Sen and B. Williams, eds., Utilitarianism and Beyond , Cambridge: Cambridge University Press, pp. 103–128.
• Schank, R., 1972, “Conceptual Dependency: A Theory of Natural Language Understanding,” Cognitive Psychology , 3.4: 532–631.
• Schaul, T. & Schmidhüber, J., 2010, “Metalearning,” Scholarpedia 5(6): 4650. URL: http://www.scholarpedia.org/article/Metalearning
• Schmidhüber, J., 2009, “Ultimate Cognition à la Gödel,” Cognitive Computation 1.2: 177–193.
• Searle, J., 1997, The Mystery of Consciousness , New York, NY: New York Review of Books.
• Searle, J., 1980, “Minds, Brains and Programs,” Behavioral and Brain Sciences , 3: 417–424.
• Searle, J., 1984, Minds, Brains and Science , Cambridge, MA: Harvard University Press. The Chinese Room Argument is covered in Chapter Two, “Can Computers Think?”.
• Searle, J., 2014, “What Your Computer Can’t Know,” New York Review of Books , October 9.
• Shapiro, S., 2000, “An Introduction to SNePS 3,” in Conceptual Structures: Logical, Linguistic, and Computational Issues. Lecture Notes in Artificial Intelligence 1867 , B. Ganter & G. W. Mineau, eds., Springer-Verlag, 510–524.
• Shapiro, S., 2003, “Mechanism, Truth, and Penrose’s New Argument,” Journal of Philosophical Logic , 32.1: 19–42.
• Siegelmann, H., 1999, Neural Networks and Analog Computation: Beyond the Turing Limit , Boston, MA: Birkhauser.
• Siegelmann, H. & and Sontag, E., 1994, “Analog Computation Via Neural Nets,” Theoretical Computer Science , 131: 331–360.
• Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., Dieleman, S., Grewe, D., Nham, J., Kalchbrenner, N., Sutskever, I., Lillicrap, T., Leach, M., Kavukcuoglu, K., Graepel T. & Hassabis D., 2016, “Mastering the Game of Go with Deep Neural Networks and Tree Search,” Nature , 529: 484–489.
• Shin, S-J, 2002, The Iconic Logic of Peirce’s Graphs , Cambridge, MA: MIT Press.
• Smolensky, P., 1988, “On the Proper Treatment of Connectionism,” Behavioral & Brain Sciences , 11: 1–22.
• Somers, J., 2013, “The Man Who Would Teach Machines to Think,” in The Atlantic . [ Available online ]
• Stanovich, K. & West, R., 2000, “Individual Differences in Reasoning: Implications for the Rationality Debate,” Behavioral and Brain Sciences , 23.5: 645–665.
• Strzalkowski, T. & Harabagiu, M. S., 2006, eds., Advances in Open Domain Question Answering ; in the series Text, Speech and Language Technology, volume 32, Dordrecht, The Netherlands: Springer-Verlag.
• Sun, R., 2002, Duality of the Mind: A Bottom Up Approach Toward Cognition , Mahwah, NJ: Lawrence Erlbaum.
• Sun, R., 1994, Integrating Rules and Connectionism for Robust Commonsense Reasoning , New York, NY: John Wiley and Sons.
• Sutton R. S. & Barto A. G., 1998, Reinforcement Learning: An Introduction , Cambridge, MA: MIT Press.
• Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I. & Fergus, R., 2014, “Intriguing Properties of Neural Networks,” in Second International Conference on Learning Representations , Banff, Canada. [ Available online ]
• Hastie, T., Tibshirani, R., & Jerome, F., 2009, The Elements of Statistical Learning , in the series Springer Series in Statistics , New York: Springer.
• Turing, A., 1950, “Computing Machinery and Intelligence,” Mind , LIX: 433–460.
• Turing, A., 1936, “On Computable Numbers with Applications to the Entscheidung-Problem,” Proceedings of the London Mathematical Society , 42: 230–265.
• Vilalta, R. & Drissi, Y., 2002, “A Perspective View and Survey of Meta-learning,” Artificial Intelligence Review , 18.2:77–95.
• Voronkov, A., 1995, “The Anatomy of Vampire: Implementing Bottom-Up Procedures with Code Trees,” Journal of Automated Reasoning , 15.2.
• Wallach, W. & Allen, C., 2010, Moral Machines: Teaching Robots Right from Wrong , Oxford, UK: Oxford University Press.
• Wermter, S. & Sun, R., 2001 (Spring), “The Present and the Future of Hybrid Neural Symbolic Systems: Some Reflections from the Neural Information Processing Systems Workshop,” AI Magazine , 22.1: 123–125.
• Suppes, P., 1972, Axiomatic Set Theory , New York, NY: Dover.
• Whiteson, S. & Whiteson, D., 2009, “Machine Learning for Event Selection in High Energy Physics,” Engineering Applications of Artificial Intelligence 22.8: 1203–1217.
• Williams, D. E., Hinton G. E., & Williams R. J., 1986 “Learning Representations by Back-propagating Errors,” Nature , 323.10: 533–536.
• Winston, P., 1992, Artificial Intelligence , Reading, MA: Addison-Wesley.
• Wos, L., Overbeek, R., Lusk R. & Boyle, J., 1992, Automated Reasoning: Introduction and Applications (2nd edition) , New York, NY: McGraw-Hill.
• Wos, L., 2013, “The Legacy of a Great Researcher,” in Automated Reasoning and Mathematics: Essays in Memory of William McCune , Bonacina, M.P. & Stickel, M.E., eds., 1–14. Berlin: Springer.
• Zalta, E., 1988, Intensional Logic and the Metaphysics of Intentionality , Cambridge, MA: Bradford Books.

How to cite this entry . Preview the PDF version of this entry at the Friends of the SEP Society . Look up topics and thinkers related to this entry at the Internet Philosophy Ontology Project (InPhO). Enhanced bibliography for this entry at PhilPapers , with links to its database.

Other Internet Resources

• Artificial Intelligence Positioned to be a Game-changer , an excellent segment on AI from CBS’s esteemed 60 Minutes program, this gives a popular science level overview of the current state of AI (as of Ocotober, 2016). The videos in the segment covers applications of AI, Watson’s evolution from winning Jeopardy! to fighting cancer and advances in robotics.
• MacroVU’s Map Coverage of the Great Debates of AI
• web site for first edition (1995)
• web site for second edition (2002)
• web site for the third edition (2009)
• Association for the Advancement of Artificial Intelligence
• Cognitive Science Society
• International Joint Conference on Artificial Intelligence
• Artificial General Intelligence (AGI) Conference
• An introduction and a collection of resources on Artificial General Intelligence
• AGI 2010 Workshop Call for a Serious Computational Science of Intelligence

Cited Resources

• Baydin A.G., Pearlmutter, B. A., Radul, A. A. & Siskind J. M., 2015, “Automatic Differentiation in Machine Learning: A Survey,” arXiv:1502.05767 [cs.SC]. URL: http://arxiv.org/abs/1502.05767
• Benenson, 2016, “ Classification Datasets Results, ” URL = http://rodrigob.github.io/are_we_there_yet/build/classification_datasets_results.html (Last accessed in July 2018).
• LeCun, Y., Cortes, C. and Burges, C. J.C, 2017, “THE MNIST DATABASE of handwritten digits,” URL = http://yann.lecun.com/exdb/mnist/ (Last accessed in July 2018).
• Levesque, J. H., 2013, “ On Our Best Behaviour ,” Speech for the IJCAI 2013 Award for Research Excellence , Beijing.
• Artificial Intelligence: Principles and Techniques (Stanford University)
• Artifical Intelligence (online course from Udacity).
• Artificial Intelligence (Columbia University).
• Artificial Intelligence MIT (Fall 2010).
• Artificial Intelligence for Robotics: Programming a Robotic Car (online course on AI formalisms that are used in mobile robots).

artificial intelligence: logic-based | causation: probabilistic | Chinese room argument | cognitive science | computability and complexity | computing: modern history of | connectionism | epistemology: Bayesian | frame problem | information technology: and moral values | language of thought hypothesis | learning theory, formal | linguistics: computational | mind: computational theory of | reasoning: automated | reasoning: defeasible | statistics, philosophy of | Turing test

Acknowledgments

Thanks are due to Peter Norvig and Prentice-Hall for allowing figures from AIMA to be used in this entry. Thanks are due as well to the many first-rate (human) minds who have read earlier drafts of this entry, and provided helpful feedback. Without the support of our AI research and development from both ONR and AFOSR, our knowledge of AI and ML would confessedly be acutely narrow, and we are grateful for the support. We are also very grateful to the anonymous referees who provided us with meticulous reviews in our reviewing round in late 2015 to early 2016. Special acknowledgements are due to the SEP editors and, in particular, Uri Nodelman for patiently working with us throughout and for providing technical and insightful editorial help.

Copyright © 2018 by Selmer Bringsjord < Selmer . Bringsjord @ gmail . com > Naveen Sundar Govindarajulu < Naveen . Sundar . G @ gmail . com >

• Accessibility

Support SEP

Mirror sites.

View this site from another server:

• Info about mirror sites

The Stanford Encyclopedia of Philosophy is copyright © 2024 by The Metaphysics Research Lab , Department of Philosophy, Stanford University

Library of Congress Catalog Data: ISSN 1095-5054

Links and Functions

• www.en.lmu.de

Breadcrumb Navigation

• Philosophy of Artificial Intelligence

Main Navigation

• For Students
• Decision Theory & Philosophy of Social Science
• Epistemology
• Philosophy of Language
• Philosophy of Mathematics
• Philosophy of Physics
• Philosophy of Science
• Research Projects
• Publications
• Journal: Erkenntnis

Our research in the philosophy of Artificial Intelligence (AI) addresses general questions concerning the methodology and foundations of AI and the role of AI in the sciences, philosophy, society, and industry. Such general questions are concerned with the following issues:

• Epistemology of Machine Learning: In what sense do machine learning algorithms learn? What is the rational way to learn from evidence? How can we get from examples to hypotheses? Does machine learning solve the problem of induction? How should we represent causal reasoning and reasoning under uncertainty formally? Is deep learning epistemically well-grounded? Does inference in artificial neural networks conform to a logical system?
• Explainable AI: What defines good explanations? Is there a difference between explanations, interpretations, and justifications? In which situations do we demand explanations for AI decisions? Are all (of these) explanations causal? In what sense are some algorithms black-boxes and others not? Is there a trade-off between explainability and accuracy? Which methods are there to make AI more transparent? How can we bridge traditional logical AI and contemporary machine learning?
• Artificial Agency and Ethics of AI: What do we mean by autonomous agents? Must autonomous AI have consciousness? Is there a way of defining consciousness in formal terms? What defines an intelligent agent, action or reasoning? Must an intelligent agent also be a moral agent? Can we implement ethical rules into AI systems? Should we rely on algorithms in safety-critical domains? What are the biases of machine learning algorithms? How can we make machine learning algorithms fair and secure?
• AI, Philosophy, and Science: How do AI and philosophy of science relate to each other? What can they learn from each other? Which philosophical challenges does modern AI research face? Is machine learning a buzzword for statistics/cognitive science? If not, what does it add? What role does machine learning play in other sciences? Will machine learning substitute statistics in the near future? Or science as we know it? Can we automatize causal discovery?

Members of faculty working on philosophy of artificial intelligence:

• Mario Günther
• Ulrike Hahn
• Stephan Hartmann
• Martin King
• Hannes Leitgeb
• Alessandra Marra
• Robert Prentner
• Tom Sterkenburg
• Borut Trpin
• Naftali Weinberger

Doctoral fellows working in philosophy of artificial intelligence:

• Omid Charrakh
• Matteo De Benedetto
• Timo Freiesleben
• Conrad Friedrich
• Johannes Leonhard Keller
• Johannes Kleiner
• José A. Leyva Galano
• Rolf Pfister
• Minkyung Wang
• Harry Waterstone
• Privacy Policy
• Accessibility

• Doctor of Philosophy in Computer Science (PhD)
• Graduate School
• Prospective Students
• Graduate Degree Programs

Go to programs search

PhD students in the Department of Computer Science may focus their research in the following areas:

• Artificial Intelligence:  computer vision, decision theory/game theory, knowledge representation and reasoning, intelligent user interfaces, machine learning, natural language understanding and generation, robotics and haptics.
• Computer Graphics:  animation, imaging, modeling, rendering, visualization.
• Data Management and Mining:  business intelligence, data integration, genomic analysis, text mining, web databases.
• Formal Verification and Analysis of Systems:  analog, digital and hybrid systems, VLSI, protocols, software.
• Human Centered Technologies:  human computer interaction (HCI), visual, haptic and multimodal interfaces, computer-supported cooperative work (CSCW), visual analytics.
• Networks, Systems, and Security:  high performance computing/parallel processing, networking, operating systems and virtualization, security.
• Scientific Computing:  numerical methods and software, differential equations, linear algebra, optimization.
• Software Engineering and Programming Languages:  development tools, foundations of computation, middleware, programming languages, software engineering.
• Theory: algorithm design and analysis (including empirical), algorithmic game theory, discrete optimization, graph theory, computational geometry

For specific program requirements, please refer to the departmental program website

What makes the program unique?

The UBC Department of Computer Science has many contacts in the computing industry. A strong rapport between the industry and research communities is beneficial to both, especially in cases where the department focuses its research to developing real-world applications.

I love Vancouver! It's the greatest city in the world. I love the integration of nature into the city; it has all of the mountains, forests, and oceans. In addition, the city is a melting pot of cultures, and that's definitely reflected at UBC. It feels like there's a place for everyone at UBC.

Michael Yin

Quick Facts

Program enquiries, admission information & requirements, 1) check eligibility, minimum academic requirements.

The Faculty of Graduate and Postdoctoral Studies establishes the minimum admission requirements common to all applicants, usually a minimum overall average in the B+ range (76% at UBC). The graduate program that you are applying to may have additional requirements. Please review the specific requirements for applicants with credentials from institutions in:

• Canada or the United States
• International countries other than the United States

Each program may set higher academic minimum requirements. Please review the program website carefully to understand the program requirements. Meeting the minimum requirements does not guarantee admission as it is a competitive process.

English Language Test

Applicants from a university outside Canada in which English is not the primary language of instruction must provide results of an English language proficiency examination as part of their application. Tests must have been taken within the last 24 months at the time of submission of your application.

Minimum requirements for the two most common English language proficiency tests to apply to this program are listed below:

TOEFL: Test of English as a Foreign Language - internet-based

Overall score requirement : 100

IELTS: International English Language Testing System

Overall score requirement : 7.0

Other Test Scores

Some programs require additional test scores such as the Graduate Record Examination (GRE) or the Graduate Management Test (GMAT). The requirements for this program are:

The GRE is not required.

2) Meet Deadlines

3) prepare application, transcripts.

All applicants have to submit transcripts from all past post-secondary study. Document submission requirements depend on whether your institution of study is within Canada or outside of Canada.

Letters of Reference

A minimum of three references are required for application to graduate programs at UBC. References should be requested from individuals who are prepared to provide a report on your academic ability and qualifications.

Statement of Interest

Many programs require a statement of interest , sometimes called a "statement of intent", "description of research interests" or something similar.

• Supervision

Students in research-based programs usually require a faculty member to function as their thesis supervisor. Please follow the instructions provided by each program whether applicants should contact faculty members.

Instructions regarding thesis supervisor contact for Doctor of Philosophy in Computer Science (PhD)

Citizenship verification.

Permanent Residents of Canada must provide a clear photocopy of both sides of the Permanent Resident card.

4) Apply Online

All applicants must complete an online application form and pay the application fee to be considered for admission to UBC.

Tuition & Financial Support

Financial Support

Applicants to UBC have access to a variety of funding options, including merit-based (i.e. based on your academic performance) and need-based (i.e. based on your financial situation) opportunities.

Program Funding Packages

All full-time PhD students will be provided with a funding package of at least $31,920 for each of the first four years of their PhD program. The funding package consists of any combination of internal or external awards, teaching-related work, research assistantships, and graduate academic assistantships. This support is contingent on full-time registration as a UBC Graduate student, satisfactory performance in assigned teaching and research assistantship duties, and good standing with satisfactory progress in your academic performance. CS students are expected to apply for fellowships or scholarship to which they are eligible.

Average Funding

• 40 students received Teaching Assistantships. Average TA funding based on 40 students was $6,950.
• 77 students received Research Assistantships. Average RA funding based on 77 students was $20,513.
• 18 students received Academic Assistantships. Average AA funding based on 18 students was $6,167.
• 81 students received internal awards. Average internal award funding based on 81 students was $11,015.
• 8 students received external awards. Average external award funding based on 8 students was $19,625.

Scholarships & awards (merit-based funding)

All applicants are encouraged to review the awards listing to identify potential opportunities to fund their graduate education. The database lists merit-based scholarships and awards and allows for filtering by various criteria, such as domestic vs. international or degree level.

Graduate Research Assistantships (GRA)

Many professors are able to provide Research Assistantships (GRA) from their research grants to support full-time graduate students studying under their supervision. The duties constitute part of the student's graduate degree requirements. A Graduate Research Assistantship is considered a form of fellowship for a period of graduate study and is therefore not covered by a collective agreement. Stipends vary widely, and are dependent on the field of study and the type of research grant from which the assistantship is being funded.

Graduate Teaching Assistantships (GTA)

Graduate programs may have Teaching Assistantships available for registered full-time graduate students. Full teaching assistantships involve 12 hours work per week in preparation, lecturing, or laboratory instruction although many graduate programs offer partial TA appointments at less than 12 hours per week. Teaching assistantship rates are set by collective bargaining between the University and the Teaching Assistants' Union .

Graduate Academic Assistantships (GAA)

Academic Assistantships are employment opportunities to perform work that is relevant to the university or to an individual faculty member, but not to support the student’s graduate research and thesis. Wages are considered regular earnings and when paid monthly, include vacation pay.

Financial aid (need-based funding)

Canadian and US applicants may qualify for governmental loans to finance their studies. Please review eligibility and types of loans .

All students may be able to access private sector or bank loans.

Foreign government scholarships

Many foreign governments provide support to their citizens in pursuing education abroad. International applicants should check the various governmental resources in their home country, such as the Department of Education, for available scholarships.

Working while studying

The possibility to pursue work to supplement income may depend on the demands the program has on students. It should be carefully weighed if work leads to prolonged program durations or whether work placements can be meaningfully embedded into a program.

International students enrolled as full-time students with a valid study permit can work on campus for unlimited hours and work off-campus for no more than 20 hours a week.

A good starting point to explore student jobs is the UBC Work Learn program or a Co-Op placement .

Tax credits and RRSP withdrawals

Students with taxable income in Canada may be able to claim federal or provincial tax credits.

Canadian residents with RRSP accounts may be able to use the Lifelong Learning Plan (LLP) which allows students to withdraw amounts from their registered retirement savings plan (RRSPs) to finance full-time training or education for themselves or their partner.

Please review Filing taxes in Canada on the student services website for more information.

Cost Estimator

Applicants have access to the cost estimator to develop a financial plan that takes into account various income sources and expenses.

Career Outcomes

111 students graduated between 2005 and 2013. Of these, career information was obtained for 106 alumni (based on research conducted between Feb-May 2016):

Sample Employers in Higher Education

Sample employers outside higher education, sample job titles outside higher education, phd career outcome survey, career options.

Our faculty and students actively interact with industry in numerous fields. Via internships, consulting and the launching of new companies, they contribute to the state-of-the-art in environmental monitoring, energy prediction, software, cloud computing, search engines, social networks, advertising, e-commerce, electronic trading, entertainment games, special effects in movies, robotics, bioinformatics, biomedical engineering, and more.

Alumni on Success

Job Title Senior Director, Product &amp; Business Development

Employer NGRAIN

Enrolment, Duration & Other Stats

These statistics show data for the Doctor of Philosophy in Computer Science (PhD). Data are separated for each degree program combination. You may view data for other degree options in the respective program profile.

ENROLMENT DATA

Completion Rates & Times

Upcoming doctoral exams, thursday, 8 august 2024 - 12:30pm - room 200, monday, 12 august 2024 - 10:30am - x836, icics building, 2366 main mall, monday, 26 august 2024 - 10:00am - x836, icics building, 2366 main mall, thursday, 29 august 2024 - 3:00pm - 146, icics building, 2366 main mall, friday, 27 september 2024 - 9:00am.

• Research Supervisors

Advice and insights from UBC Faculty on reaching out to supervisors

These videos contain some general advice from faculty across UBC on finding and reaching out to a supervisor. They are not program specific.

This list shows faculty members with full supervisory privileges who are affiliated with this program. It is not a comprehensive list of all potential supervisors as faculty from other programs or faculty members without full supervisory privileges can request approvals to supervise graduate students in this program.

• Achermann, Reto (Computing systems; Computer systems engineering; Systems software; resilient and efficient systems; intersection of operating systems, applied formal methods and hardware models.)
• Beschastnikh, Ivan (Computer and information sciences; software engineering; distributed systems; cloud computing; software analysis; Machine Learning)
• Bowman, William (Computer and information sciences; Programming languages and software engineering; Programming languages; Compilers; programming languages)
• Carenini, Giuseppe (Artificial intelligence, user modeling, decision theory, machine learning, social issues in computing, computational linguistics, information visualization)
• Clune, Jeff
• Conati, Cristina (artificial intelligence, human-computer interaction, affective computing, personalized interfaces, intelligent user interfaces, intelligent interface agents, virtual agent, user-adapted interaction, computer-assisted education, educational computer games, computers in education, user-adaptive interaction, Artificial intelligence, adaptive interfaces, cognitive systems, user modelling)
• Condon, Anne (Algorithms; Molecular Programming)
• Ding, Jiarui (Bioinformatics; Basic medicine and life sciences; Computational Biology; Machine Learning; Probabilistic Deep Learning; single-cell genomics; visualization; Cancer biology; Computational Immunology; Food Allergy; neuroscience)
• Evans, William (Computer and information sciences; Algorithms; theoretical computer science; Computer Sciences and Mathematical Tools; computational geometry; graph drawing; program compression)
• Feeley, Michael (Distributed systems, operating systems, workstation and pc clusters)
• Friedlander, Michael (numerical optimization, numerical linear algebra, scientific computing, Scientific computing)
• Friedman, Joel (Computer and information sciences; Algebraic Graph Theory; Combinatorics; Computer Science Theory)
• Garcia, Ronald (Programming languages; programming languages)
• Greenstreet, Mark (Dynamic systems, formal methods, hybrid systems, differential equations)
• Greif, Chen (Numerical computation; Numerical analysis; scientific computing; numerical linear algebra; numerical solution of elliptic partial differential equations)
• Gujarati, Arpan (Computer and information sciences; Systems)
• Harvey, Nicholas (randomized algorithms, combinatorial optimization, graph sparsification, discrepancy theory and learning theory; algorithmic problems arising in computer networking, including cache analysis, load balancing, data replication, peer-to-peer networks, and network coding.)
• Hoang, Nguyen Phong (networking; security & privacy; network security; online privacy; Internet measurement)
• Holmes, Reid (Computer and information sciences; computer science; open source software; software comprehension; software development tools; software engineering; software quality; software testing; static analysis)
• Hu, Alan (Computer and information sciences; formal methods; formal verification; model checking; nonce to detect automated mining of profiles; post-silicon validation; security; software analysis)
• Hutchinson, Norman (Computer and information sciences; Computer Systems; distributed systems; File Systems; Virtualization)
• Kiczales, Gregor (MOOCs, Blended Learning, Flexible Learning, University Strategy for Flexible and Blended Learning, Computer Science Education, Programming Languages, Programming languages, aspect-oriented programming, foundations, reflections and meta programming, software design)
• Lakshmanan, Laks (data management and data cleaning; data warehousing and OLAP; data and text mining; analytics on big graphs and news; social networks and media; recommender systems)
• Lecuyer, Mathias (Machine learning systems; Guarantees of robustness, privacy, and security)
• Lemieux, Caroline (Programming languages and software engineering; help developers improve the correctness, security, and performance of software systems; test-input generation; specification mining; program synthesis)

Doctoral Citations

Sample Thesis Submissions

• Discrete optimization problems in geometric mesh processing
• On effective learning for multimodal data
• From devices to data and back again : a tale of computationally modelling affective touch
• Towards alleviating human supervision for document-level relation extraction
• Methods for design of efficient on-device natural language processing architectures
• A formal framework for understanding run-time checking errors in gradually typed languages
• Understanding semantics and geometry of scenes
• Computational tools for complex electronic auctions
• From videos to animatable 3d neural characters
• Structured representation learning by controlling generative models
• Versatile neural approaches to more accurate and robust topic segmentation
• Machine learning for spectroscopic data analysis : challenges of limited labelled data
• Enriching block-based end-user programming with visual features
• Accelerating Bayesian inference in probabilistic programming
• Computationally efficient geometric methods for optimization and inference in machine learning

Related Programs

Same specialization.

• Master of Science in Computer Science (MSc)

Same Academic Unit

• Master of Data Science (MDS)

At the UBC Okanagan Campus

Further information, specialization.

Computer Science covers Bayesian statistics and applications, bioinformatics, computational intelligence (computational vision, automated reasoning, multi-agent systems, intelligent interfaces, and machine learning), computer communications, databases, distributed and parallel systems, empirical analysis of algorithms, computer graphics, human-computer interaction, hybrid systems, integrated systems design, networks, network security, networking and multimedia, numerical methods and geometry in computer graphics, operating systems, programming languages, robotics, scientific computation, software engineering, visualization, and theoretical aspects of computer science (computational complexity, computational geometry, analysis of complex graphs, and parallel processing).

UBC Calendar

Program website, faculty overview, academic unit, program identifier, classification, social media channels, supervisor search.

Departments/Programs may update graduate degree program details through the Faculty & Staff portal. To update contact details for application inquiries, please use this form .

My experience as a non-degree student was really positive. I loved the way lectures, tutorials, labs, assignments and projects all complemented each other. I found the lectures stimulating and the professors and TAs encouraging. I also loved just being on the UBC campus. I'm surrounded by nature (...

Geoffrey Woollard

I applied to UBC in 2020, during the pandemic. It was a close call between working with Marcus Brubaker, who co-founded my former employer Structura Biotechnology, before becoming an Assistant Professor at York University, and working with Khanh Dao Duc at UBC. Khanh introduced me to his...

Baraa Orabi

I think three factors had a differentiating effect on this decision: UBC's unique multidisciplinary environment which is key to my research as a computer scientist and bioinformatician. UBC being on the West Coast generally and Vancouver specifically and the amazing weather and nature that comes...

Experience the colours of Vancouver

Great academic programs, great location: the distinct seasons and mild climate are among the reasons why graduate students choose to study here -- from the autumn leaves to cherry blossoms, witness the many colours Vancouver has to offer.

• Why Grad School at UBC?
• Application & Admission
• Info Sessions
• Research Projects
• Indigenous Students
• International Students
• Tuition, Fees & Cost of Living
• Newly Admitted
• Student Status & Classification
• Student Responsibilities
• Managing your Program
• Health, Wellbeing and Safety
• Professional Development
• Dissertation & Thesis Preparation
• Final Doctoral Exam
• Final Dissertation & Thesis Submission
• Life in Vancouver
• Vancouver Campus
• Graduate Student Spaces
• Graduate Life Centre
• Life as a Grad Student
• Graduate Student Ambassadors
• Meet our Students
• Award Opportunities
• Award Guidelines
• Minimum Funding Policy for PhD Students
• Killam Awards & Fellowships
• Dean's Message
• Leadership Team
• Strategic Plan & Priorities
• Vision & Mission
• Equity, Diversity & Inclusion
• Initiatives, Plans & Reports
• Graduate Education Analysis & Research
• Media Enquiries
• Newsletters
• Giving to Graduate Studies

Strategic Priorities

• Strategic Plan 2019-2024
• Improving Student Funding
• Promoting Excellence in Graduate Programs
• Enhancing Graduate Supervision
• Advancing Indigenous Inclusion
• Supporting Student Development and Success
• Reimagining Graduate Education
• Enriching the Student Experience

Initiatives

• Public Scholars Initiative
• 3 Minute Thesis (3MT)
• PhD Career Outcomes

Postgraduate taught

MA Philosophy of Artificial Intelligence

Explore social, political, ethical and conceptual issues in AI

Year of entry: 2024 (September)

1 year full-time, 2 years part-time

Department of Philosophy

September 2024 ( semester dates )

Apply for this course

Join us online or in person to find out more about postgraduate study at York.

in the UK for research in Philosophy

according to the Times Higher Education's ranking of the latest REF results (2021).

for our research impact

Explore the ways in which artificial intelligence (AI) is shaping our lives and raising complex social, political, ethical and conceptual questions. 

On this MA you can study issues in AI and investigate them through your own research. You can choose from a wide range of option modules. You'll also be able to create a substantial piece of research following your own interests, with our support. You'll develop valuable transferable skills in research, analysis, critical thinking and presentation which will be essential if you are thinking of continuing to study to PhD level. The skills that you develop on the course will also equip you for a range of careers, including work with technology ethics and policy, and technology consulting.

You’ll join our postgraduate community across both the Department of Philosophy and the Humanities Research Centre, and participate in our broad and diverse research culture.

World-class research

Ranked 2nd in the UK in the Times Higher Education’s ranking of the latest REF results (2021).

Gender equality

We hold an Athena SWAN bronze award for our commitment to gender equality.

York Philosophy Society

Join a warm and welcoming student-run society.

Course content

You'll study philosophical issues surrounding artificial intelligence and machine learning. Teaching will be research-led, drawing on our strong and diverse research community. You'll learn about dissertation preparation, and will work on postgraduate research skills. Later in the year you and your peers will hold an in-house conference.

You'll attend regular research seminars (colloquia) in the department at which guest speakers will discuss their latest research.

You’ll receive encouragement, support and guidance in selecting and independently studying ideas of personal interest to you.

Course structure for part-time study

Year 1: Topics in the Philosophy of Artificial Intelligence and two option modules.

Year 2: Research Skills and Dissemination Practice, two option modules and your dissertation.

• Dissertation

Core modules 

• Research Skills and Dissemination Practice
• Topics in the Philosophy of Artificial Intelligence

Option modules 

You will also study four option modules. Option modules vary from year to year according to staff availability. The availability of each option module is subject to a minimum enrolment number. In previous years, options have covered topics such as:

• Advanced Topics in Ethics
• Advanced Topics in Political Philosophy
• Ancient Philosophy
• Buddhism, Ethics, and the Self
• Comparative Analytic Theology
• Fictionalism
• Freedom, Right, and Revolution: Post-Kantian Moral, Legal, and Political Philosophy
• Human and Machine Creativity  
• Models and Modelling in Science and Technology
• Phenomenology and Psychiatry
• Philosophical Theology after Maimonides
• Problem Based Learning: Government and Health Data
• Problem Based Learning: Surveillance Capitalism
• Reading Philosophy
• Science-Engaged Analytic Theology
• Social Justice and Political Economy
• Time, Tense, and Existence
• Topics in Consciousness and Representation
• Topics in Perception and Emotion
• Using and Abusing Language
• Wittgenstein's Tractatus

Our modules may change to reflect the latest academic thinking and expertise of our staff, and in line with Department/School academic planning.

Your 10,000 word dissertation enables you to produce a sustained piece of critical writing on a topic of your choosing. It will allow you to apply the core knowledge, skills and experience that you have gained in the previous stage of the course.

You'll attend dissertation preparation seminars to enable you to write your proposal, with further support later in the year. You'll be supervised by a member of staff with expertise in the relevant area.

The York approach

Every course at York is built on a distinctive set of learning outcomes. These will give you a clear understanding of what you will be able to accomplish at the end of the course and help you explain what you can offer employers. Our academics identify the knowledge, skills, and experiences you'll need upon graduation and then design the course to get you there.

Students who complete this course will be able to:

• Critically review current scholarship and research on key problems, issues and debates across a wide range of topics within the philosophy of artificial intelligence and related fields.
• Apply critical perspectives to current research in their field in a technically proficient yet accessible and clear manner informed by current practice, scholarship, and research.
• Work effectively and collaboratively in the planning, organization, and delivery of significant research events and reports (in a manner continuous with best practice at a professional level).
• Take full ownership of their own development as researchers and professionals, continually reflecting on their own practice, progress, and received feedback, and seeking assistance where appropriate.
• Give presentations of their ideas and arguments at a professional level (aligned with best practice) to varied audiences.
• Create detailed and persuasive project proposals at a high level (continuous with best practice in professional research proposals), and initiate, develop, and complete substantial independent projects.

Fees and funding

Annual tuition fees for 2024/25.

Students on a Student Visa are not currently permitted to study part-time at York.

For courses which are longer than one year , the tuition fees quoted are for the first year of study.

• UK (home) fees may increase in subsequent years (up to a maximum of 2%).
• International fees may increase in subsequent years in line with the prevailing Consumer Price Index (CPI) inflation rate (up to a maximum of 10%).

Fees information

UK (home) or international fees?  The level of fee that you will be asked to pay depends on whether you're classed as a UK (home) or international student.  Check your fee status .

Find out more information about tuition fees and how to pay them.

• Postgraduate taught fees and expenses

Funding information

Discover your funding options to help with tuition fees and living costs.

We'll confirm more funding opportunities for students joining us in 2024/25 throughout the year.

If you've successfully completed an undergraduate degree at York you could be eligible for a  10% Masters fee discount .

Funding opportunities

• UK government Masters loans
• Funding for UK students
• Funding for international students

Departmental scholarship information

For further information on all eligibility criteria and how to apply for our scholarships see our  funding opportunities for Philosophy .

Roger Woolhouse Prize

A prize of £500 awarded to the MA Philosophy student who achieves the highest essay mark (>72) in the MA assessment period. 

The David Efird Student Prize

A prize of £300 will be awarded to the student who achieves the highest essay mark (>72) in the field of philosophy of religion, or research on contemporary issues or themes using a philosophy of religion perspective.

Living costs

You can use our  living costs guide  to help plan your budget. It covers additional costs that are not included in your tuition fee such as expenses for accommodation and study materials.

Teaching and assessment

You’ll work with world‐leading academics who’ll challenge you to think independently and excel in all that you do. Our approach to teaching will provide you with the knowledge, opportunities, and support you need to grow and succeed in a global workplace.

Teaching format

You'll be taught by intensive seminars and individual or small-group tutorials, which will allow you and your tutors to systematically explore complex issues at the forefront of the philosophy of artificial intelligence.

You'll be part of a lively research community at the Humanities Research Centre which includes staff, postgraduate students, postdoctoral scholars and academic visitors from across the arts and humanities. 

Teaching location

You will be based in the Department of Philosophy on Campus West. Most of your contact hours will be nearby on Campus West.

About our campus

Our beautiful green campus offers a student-friendly setting in which to live and study, within easy reach of the action in the city centre. It's  easy to get around campus  - everything is within walking or pedalling distance, or you can always use the fast and frequent bus service.

Assessment and feedback

Your work will be assessed in a variety of ways:

• you'll write a 4,000 word essay for some modules
• you'll write a research proposal and complete a reflective journal for the skills element of Research Skills and Dissemination Practice
• you'll present a paper, chair a session, and complete a funding proposal for the dissemination element of Research Skills and Dissemination Practice

You will also receive assignments throughout your course which will provide constant feedback on your development, and help prepare you for your assessments.

Typical offer Undergraduate degree 2:1 or equivalent Other international qualifications

Additional requirements

English language.

If English isn't your first language you may need to provide evidence of your English language ability. We accept the following qualifications:

	Minimum requirement
IELTS (Academic and Indicator)	7.0, minimum 7.0 in writing and 6.5 in all other components
Cambridge CEFR	C1 Advanced: 185, with a minimum of 185 in Writing and no less than 176 in all other components
Oxford ELLT	8, minimum of 8 in writing and no less than 7 in all other components
Duolingo	130, minimum 130 in Production and 120 in all other components
LanguageCert SELT	C1 with 33/50 in each component
LanguageCert Academic	75 with a minimum of 75 in Writing and no less than 70 in all other components
KITE	495-526, with 495-526 in writing and 459-494 in all other components
Skills for English	C1: Pass overall, with Pass in each component
PTE Academic	67, minimum of 67 in Writing and 61 in all other components
TOEFL	96, minimum 24 in Writing and 23 in all other components
Trinity ISE III	Distinction in all components

For more information see our postgraduate English language requirements .

If you haven't met our English language requirements

You may be eligible for one of our pre-sessional English language courses . These courses will provide you with the level of English needed to meet the conditions of your offer.

The length of course you need to take depends on your current English language test scores and how much you need to improve to reach our English language requirements.

After you've accepted your offer to study at York, we'll confirm which pre-sessional course you should apply to via You@York .

You can apply and send all your documentation online. You don’t need to complete your application all at once: you can start it, save it and finish it later.

• How to apply

Get in touch if you have any questions

• Philosophy Postgraduate Admissions
• [email protected]
• +44 (0)1904 323251

Related courses

• MA Analytic Theology
• MA Philosophy
• MA Political and Legal Philosophy
• MA Philosophy (by research)
• MSc Artificial Intelligence
• MSc Artificial Intelligence for the Creative Industries

Discover York

Accommodation

We offer a range of campus accommodation to suit you and your budget, from economy to premium.

Student life

Explore campus and city life and hear what our current students have to say about living here.

Lively, full of culture and beautiful, York is regularly voted one of the best places to live and visit in the UK.

Find out more about York. Chat to staff and students and take the tour, on campus or online.

• Aviation and Astronautical Sciences
• Computer Science, Artificial Intelligence and Data Science
• Construction and Facilities
• Critical Infrastructure
• Cyber & Information Security
• Cyberpsychology
• Engineering
• Engineering Technologies
• Intelligence and Global Security Studies
• Management of Technology
• Occupational Safety and Health
• Uncrewed Systems
• Doctoral Degrees
• Master's Degrees
• Bachelor's Degrees
• Online Programs
• Associate Degrees
• Certificates
• Minor Degrees
• STEM Events
• Webinars and Podcasts
• Master's
• Undergraduate
• Transfer Students
• Military and Veterans
• International Students
• Admissions Counselor
• Capitol Connections
• Accepted Students
• Project Lead the Way
• Builder Culture
• Campus Life
• Clubs and Organizations
• Centers and Labs
• Online Classes
• The Capitol Commitment
• Top Employers
• Co-ops and Internships
• Professional Education
• Find a Mentor
• Career Services
• Capitol Online Job Board
• Recruiters and Employers
• Why Capitol Tech
• At a Glance
• Mission, Vision and Goals
• Diversity, Equity and Inclusion
• Washington, D.C.
• Capitol History
• Capitol Partners
• News and Events
• Visitors/Campus
• Accreditation
• Recognitions & Awards
• Current Students
• Faculty & Staff
• Alumni & Giving
• News & Events
• Capitology Blog
• Maps / Directions

• Degrees and Programs

Doctor of Philosophy (PhD) in Machine Learning

• Request Information

Earn a Ph.D. in Machine Learning and discover the elements of artificial intelligence, computer engineering, and data analytics involved in this evolving field.

The doctoral degree in Machine Learning explores the ways in which algorithmic data is generated and leveraged for statistical applications and computational analysis in model-based decision-making. Students will learn the current operations, international relationships, and areas of improvement in this field, as well as research methodologies and future demands of the industry.

The PhD in Machine Learning is for current or experienced professionals in a field related to machine learning, artificial intelligence, computer science, or data analytics. Students will pursue a deep proficiency in this area using interdisciplinary methodologies, cutting-edge courses, and dynamic faculty. Graduates will contribute significantly to the Machine Learning field through the creation of new knowledge and ideas, and will quickly develop the skills to engage in leadership, research, and publishing. 

As your PhD progresses, you will move through a series of progression points and review stages by your academic supervisor. This ensures that you are engaged in research that will lead to the production of a high-quality thesis and/or publications, and that you are on track to complete this in the time available. Following submission of your PhD Thesis or accepted three academic journal articles, you will have an oral presentation assessed by an external expert in your field.

Why Capitol?

Expert guidance in doctoral research

Capitol’s doctoral programs are supervised by faculty with extensive experience in chairing doctoral dissertations and mentoring students as they launch their academic careers. You’ll receive the guidance you need to successfully complete your doctoral research project and build credentials in the field.

Proven academic excellence

Study at a university that specializes in industry-focused education in technology-based fields, nationally recognized for academic excellence in our programs.

Program is 100% online

Our PhD in Machine Learning is offered 100% online, with no on-campus classes or residencies required, allowing you the flexibility needed to balance your studies and career.

Key Faculty

Vice President

Associate Chair, Director of Engineering Labs

Dissertation Chair

Career Opportunities

Market demand for machine learning expertise

Graduates will contribute significantly to the rapidly growing machine learning field through the creation of new knowledge and ideas, and will be prepared for in-demand roles such as a trusted subject matter expert, researcher, technician, manager, or professor.

This program may be completed with a minimum of 60 credit hours, but may require additional credit hours, depending on the time required to complete the dissertation/publication research. Students who are not prepared to defend after completion of the 60 credits will be required to enroll in RSC-899, a one-credit, eight-week continuation course. Students are required to be continuously enrolled/registered in the RSC-899 course until they successfully complete their dissertation defense/exegesis.

The student will produce, present, and defend a doctoral dissertation after receiving the required approvals from the student’s Committee and the PhD Review Boards.

Doctor of Philosophy in Machine Learning Courses Total Credits: 60

MACHINE LEARNING DOCTORAL CORE: 30 CREDITS

	6
	6
	6
	6
	6

OFFENSIVE MACHINE LEARNING DOCTORAL RESEARCH AND WRITING: 30 CREDITS 

Educational Objectives:  

Students will... 

1. Integrate and synthesize alternate, divergent, or contradictory perspectives within the field of Machine Learning. 2. Demonstrate advanced knowledge and competencies in ethics of Machine Learning. 3. Analyze theories, tools, and frameworks used in Machine Learning. 4. Evaluate the legal, social, economic, environmental, and ethical impact of actions within Machine Learning. 5. Implement Machine Learning plans needed for advanced global applications.

Learning Outcomes:  

Upon graduation... 

1. Graduates will integrate the theoretical basis and practical applications of Machine Learning into their professional work.  2. Graduates will demonstrate the highest mastery of the subject matter. 3. Graduates will evaluate complex problems, synthesize divergent/alternative/contradictory perspectives and ideas fully, and develop advanced solutions to Machine Learning challenges. 4. Graduates will contribute to the body of knowledge in the study of the subject. 5. Graduates will be at the forefront of Machine Learning planning and implementation.

Tuition & Fees

Tuition rates are subject to change.

The following rates are in effect for the 2024-2025 academic year, beginning in Fall 2024 and continuing through Summer 2025:

• The application fee is $100
• The per-credit charge for doctorate courses is $950. This is the same for in-state and out-of-state students.
• Retired military receive a $50 per credit hour tuition discount
• Active duty military receive a $100 per credit hour tuition discount for doctorate level coursework.
• Information technology fee $40 per credit hour.
• High School and Community College full-time faculty and full-time staff receive a 20% discount on tuition for doctoral programs.

Find additional information for 2024-2025 doctorate tuition and fees.

Need more info, or ready to apply?

Philosophy, PhD

• Program description
• At a glance
• Degree requirements
• Admission requirements
• Tuition information
• Application deadlines
• Program learning outcomes
• Career opportunities
• Contact information

cultural, ethics, logic, theory

With access to top philosophical thinkers, you can earn an interdisciplinary doctorate in your choice of fields, such as law, medicine, religion and politics, that enables you to make an impact on the world.

General areas of research include ethics, political philosophy, metaphysics, epistemology, philosophy of law, philosophy of science, philosophy of language, philosophy of religion and the history of philosophy. The program features a focus on practical and applied philosophy and an interdisciplinary coursework component related to the student's research topic.

Practical philosophy includes the fields of ethics, philosophy of law, social and political philosophy, feminist ethics and political philosophy.

Applied philosophy includes the application of theories developed within any of the subdisciplines of philosophy to everyday problems or phenomena, such as the application of the philosophy of language in relation to hate speech, or the philosophy of mind in relation to computing and artificial intelligence. Applied philosophy also includes the application of research produced by methods used in other disciplines in order for the student to understand and address philosophical questions, like the application of data-gathering instruments used in psychology to answer questions in experimental philosophy.

Students may design dissertation projects in any of the major subfields of philosophy. For their interdisciplinary coursework supporting the dissertation project, students might, for example, pursue a certificate in social transformation, gender studies, responsible innovation in sciences, or engineering and society.

Members of the faculty are involved in interdisciplinary work in a variety of fields and enjoy close ties with the Lincoln Center for Applied Ethics, the College of Law and a number of other graduate programs at the university. The ASU philosophy faculty group sponsors an active colloquium series and regular philosophical conferences on diverse topics. The Lincoln Center for Applied Ethics also sponsors a wide range of activities, including large-scale conferences, distinguished visitors and support for graduate study.

• College/school: The College of Liberal Arts and Sciences
• Location: Tempe

84 credit hours, a written comprehensive exam, an oral comprehensive exam, a prospectus and a dissertation

Required Core Areas (15 credit hours) applied philosophy (3) epistemology (3) formal methods (3) metaphysics (3) value theory (3)

Electives (39 credit hours)

Research (18 credit hours) PHI 792 Research (12)

Culminating Experience (12 credit hours) PHI 799 Dissertation (12)

Additional Curriculum Information Students should see the academic unit for the list of courses approved for each required core area.

In completing the electives requirements, at least nine credit hours and no more than 18 credit hours must be from other disciplines supporting the student's proposed dissertation area; 30 credit hours from a previously awarded master's degree may apply toward this requirement with approval by the student's supervisory committee and the Graduate College.

To ensure breadth in the traditional areas of philosophy, students must pass with a grade of "B" or better (3.00 on a 4.00 scale).

Applicants must fulfill the requirements of both the Graduate College and The College of Liberal Arts and Sciences.

Applicants are eligible to apply to the program if they have earned a bachelor's or master's degree in any field from a regionally accredited institution.

Applicants must have a minimum cumulative GPA of 3.00 (scale is 4.00 = "A") in the last 60 hours of their first bachelor's degree program, or a minimum cumulative GPA of 3.00 (scale is 4.00 = "A") in an applicable master's degree program.

All applicants must submit:

• graduate admission application and application fee
• official transcripts
• statement of purpose
• curriculum vitae
• writing sample
• three letters of recommendation
• proof of English proficiency

Additional Application Information An applicant whose native language is not English must provide proof of English proficiency , a copy of an article or research paper in their native or principal research language, as well as the English writing sample required of all students regardless of their current residency. The philosophy program requires a TOEFL iBT score of at least 100, or a score of 7.0 on the IELTS.

The statement of purpose should explain the applicant's scholarly background and training, career goals, the primary field the applicant wishes to pursue and the proposed research specialization (no more than 600 words in length).

The writing sample must be a piece of philosophical writing, preferably a seminar paper or published article of no more than 20 pages.

Session	Modality	Deadline	Type
Session A/C	In Person	01/15	Final

Program learning outcomes identify what a student will learn or be able to do upon completion of their program. This program has the following program outcomes:

• Achieve competence with philosophical literature and writing
• Achieve professional-level skills in mastering literature and philosophical writing
• Able to identify and articulate a philosophical problem or question in one of the core areas in philosophy --- that is, metaphysics, epistemology and value theory

Both the MA and doctoral programs in philosophy help students develop and hone skills that are highly marketable and easily transferable.

Philosophy teaches its students to think critically, creatively and imaginatively. Though routine jobs are increasingly being lost to advances in automation and artificial intelligence, the skills taught by philosophy are irreplaceable by technology, highly sought-after by employers and transferrable from one occupation to another. Graduates have the ability to read closely and with a critical eye; to analyze complex problems and identify all the possible solutions, including some creative solutions; to assess the merits of each possible solution; and to articulate and argue for or against various possible solutions in clear, precise and unambiguous language.

As philosophy focuses on honing certain skills rather than acquiring a particular body of knowledge, philosophy prepares its students for a wide variety of careers rather than for just one particular occupation. Indeed, philosophy prepares its students for any career requiring problem-solving; clear, critical and creative thinking; and excellent reading, writing and communication skills.

The program is designed to prepare students for careers as philosophers, as teachers of philosophy and in areas in which they may benefit from advanced training in philosophy, such as law, civil service and publishing.

Career examples include:

• businessperson
• computer programmer
• public policy analyst

Historical, Philosophical & Religious Studies, Sch | COOR 4595 [email protected] 480-965-5778

2024-25 Catalog coming August 5th!

This site is currently under construction. If you are an incoming frosh, rising sophomore or new transfer student, please check back August 5 th , when you can browse next year's IntroSems and start applying for priority enrollment in up to 3 seminars per quarter.

PHIL 20N: Philosophy of Artificial Intelligence

Quarter(s) & Preference

General education requirements.

Not currently certified for a requirement. Courses are typically considered for Ways certification a quarter in advance.

See Explore Courses for Schedule

This seminar is expected to be in high demand. If you rank it as your first choice for priority enrollment, please be sure to apply for a second and third choice seminar for the quarter. You are also encouraged to write an additional statement for your lower ranked selection(s) so those faculty learn about your interest.

Course Description

Is it really possible for an artificial system to achieve genuine intelligence: thoughts, consciousness, emotions? What would that mean? How could we know if it had been achieved? Is there a chance that we ourselves are artificial intelligences? Would artificial intelligences, under certain conditions, actually be persons? If so, how would that affect how they ought to be treated and what ought to be expected of them? Emerging technologies with impressive capacities already seem to function in ways we do not fully understand. What are the opportunities and dangers that this presents? How should the promises and hazards of these technologies be managed?

Philosophers have studied questions much like these for millennia, in scholarly debates that have increased in fervor with advances in psychology, neuroscience, and computer science. The philosophy of mind provides tools to carefully address whether genuine artificial intelligence and artificial personhood are possible. Epistemology (the philosophy of knowledge) helps us ponder how we might be able to know. Ethics provides concepts and theories to explore how all of this might bear on what ought to be done. We will read philosophical writings in these areas as well as writings explicitly addressing the questions about artificial intelligence, hoping for a deep and clear understanding of the difficult philosophical challenges the topic presents.

No background in any of this is presupposed, and you will emerge from the class having made a good start learning about computational technologies as well as a number of fields of philosophical thinking. It will also be a good opportunity to develop your skills in discussing and writing critically about complex issues.

Meet the Instructor: John Etchemendy

"In the spirit of full disclosure, I need to warn you about something: in spite of the fact that I’ve been on the Stanford faculty for longer than I care to admit, you should know that I am also somewhat a novice instructor. You see, I was once a pretty good teacher (or so I’m told), but my reward for that was I was asked to be Stanford’s provost, and that’s what I spent about 20 years doing. (If you don’t know what the provost does, that’s fine: main thing is there’s no teaching involved!)

"So here I am, teaching my First-Year Seminar after those years as provost, and you’re thinking about taking it? Cool. I like students who are intrepid. 

"I’m a philosopher and logician by training. I have been involved with researchers in artificial intelligence (AI) since graduate school, in the relatively early days of the field. Most people only became aware of AI about ten years ago, when AI algorithms started affecting their lives. But the AI we see today is the result of a lot of work that began over 50 years ago, a good bit of it done at Stanford. During those years, I’ve watched the field progress, but I’ve also seen it fall flat on its face! 

"There are few fields that raise more philosophical questions than AI: questions about ethics, minds, consciousness, and free will. Are we, as some would say, creating our own successors? And will they be obedient servants or our new masters? Can we upload our minds into silicon chips and thereby become immortal? Or does this just create an imposter? And so on.

"I’d like to explore some of these questions with a few intrepid students who aren’t afraid to come along for the ride—even if the driver is a novice. Let’s think through these problems together."

Department(s)

Go to the IntroSems' VCA

More News Topics

AMSTUD 104Q: Picturing Americans

• American Studies

APPPHYS 79N: Energy Options for the 21st Century

• Applied Physics

ARTSTUDI 150N: Queer Sculpture

Study at Cambridge

About the university, research at cambridge.

• Undergraduate courses
• Events and open days
• Fees and finance
• Postgraduate courses
• How to apply
• Postgraduate events
• Fees and funding
• International students
• Continuing education
• Executive and professional education
• Courses in education
• How the University and Colleges work
• Term dates and calendars
• Visiting the University
• Annual reports
• Equality and diversity
• A global university
• Public engagement
• Give to Cambridge
• For Cambridge students
• For our researchers
• Business and enterprise
• Colleges & departments
• Email & phone search
• Museums & collections

Postgraduate Study

• Why Cambridge overview
• Chat with our students
• Cambridge explained overview
• The supervision system
• Student life overview
• In and around Cambridge
• Leisure activities
• Student unions
• Music awards
• Student support overview
• Mental health and wellbeing
• Disabled students
• Accommodation
• Language tuition
• Skills training
• Support for refugees
• Courses overview
• Course Directory
• Department directory
• Qualification types
• Funded studentships
• Part-time study
• Research degrees
• Visiting students
• Finance overview
• Fees overview
• What is my fee status?
• Part-time fees
• Application fee
• Living costs
• Funding overview
• Funding search
• How to apply for funding
• University funding overview
• Research Councils (UKRI)
• External funding and loans overview
• Funding searches
• External scholarships
• Charities and the voluntary sector
• Funding for disabled students
• Widening participation in funding
• Colleges overview
• What is a College?
• Choosing a College
• Applying overview
• Before you apply
• Entry requirements
• Application deadlines
• How do I apply? overview
• Application fee overview
• Application fee waiver
• Life Science courses
• Terms and conditions
• Continuing students
• Disabled applicants
• Supporting documents overview
• Academic documents
• Finance documents
• Evidence of competence in English
• AI and postgraduate applications
• Terms and Conditions
• Applicant portal and self-service
• After you apply overview
• Confirmation of admission
• Student registry
• Previous criminal convictions
• Deferring an application
• Updating your personal details
• Appeals and Complaints
• Widening participation
• Postgraduate admissions fraud
• International overview
• Immigration overview
• ATAS overview
• Applying for an ATAS certificate
• Current Cambridge students
• International qualifications
• Competence in English overview
• What tests are accepted?
• International events
• International student views overview
• Akhila’s story
• Alex’s story
• Huijie’s story
• Kelsey’s story
• Nilesh’s story
• Get in touch!
• Events overview
• Upcoming events
• Postgraduate Open Days overview
• Discover Cambridge: Master’s and PhD Study webinars
• Virtual tour
• Research Internships
• How we use participant data
• Postgraduate Newsletter

MPhil in Ethics of AI, Data and Algorithms

Primary tabs.

• Overview (active tab)
• Requirements
• How To Apply

Course closed:

Ethics of AI, Data and Algorithms is no longer accepting new applications.

The programme aims to equip students with the skills and knowledge to contribute critically and constructively to cross-disciplinary research on AI, data and algorithms and their ethical and societal implications. It introduces students from diverse backgrounds to relevant research skills and specialist knowledge from a range of academic disciplines, and provides them with the opportunity to carry out focused research under close supervision by domain experts at the University.

Learning Outcomes

Knowledge and understanding

By the end of the course, students will have acquired:

• A deeper knowledge of the history, philosophy and theory of AI, data and algorithms
• an understanding of the capabilities of current and future digital technologies and their relation to wider society
• a critical perspective on the ethical challenges that arise from applications of AI, data and algorithms and how these sit within and interact with wider society
• the conceptual understanding and analytic tools to evaluate and contribute to research on the nature, impacts and governance of AI, data and algorithms

Skills and other attributes

Graduates of the course will be able to:

Synthesise and analyse research and advanced scholarship across disciplines Put theoretical and academic knowledge into practice Present their own ideas in a public forum and contribute constructively within an international and cross-disciplinary environment

Students admitted for the MPhil can apply to continue as PhD students with a relevant Faculty.  For details of the process for applying to do a PhD, and the standard required, students should consult the Faculty in question.

The Postgraduate Virtual Open Day usually takes place at the end of October. It’s a great opportunity to ask questions to admissions staff and academics, explore the Colleges virtually, and to find out more about courses, the application process and funding opportunities. Visit the  Postgraduate Open Day  page for more details.

See further the  Postgraduate Admissions Events  pages for other events relating to Postgraduate study, including study fairs, visits and international events.

Departments

This course is advertised in the following departments:

• Faculty of Philosophy
• Leverhulme Centre for the Future of intelligence

Key Information

9 months full-time, study mode : taught, master of philosophy, faculty of philosophy this course is advertised in multiple departments. please see the overview tab for more details., course - related enquiries, application - related enquiries, course on department website, dates and deadlines:, michaelmas 2024 (closed).

Some courses can close early. See the Deadlines page for guidance on when to apply.

Funding Deadlines

These deadlines apply to applications for courses starting in Michaelmas 2024, Lent 2025 and Easter 2025.

Similar Courses

• AI Ethics and Society MSt
• Antarctic Studies PhD
• Asian and Middle Eastern Studies (Korean Studies) MPhil
• Philosophy MPhil
• Asian and Middle Eastern Studies by Research (Korean Studies) MPhil

Postgraduate Admissions Office

• Admissions Statistics
• Start an Application
• Applicant Self-Service

At a glance

• Bringing a family
• Current Postgraduates
• Cambridge Students' Union (SU)

University Policy and Guidelines

Privacy Policy

Information compliance

Equality and Diversity

Terms of Study

About this site

About our website

Privacy policy

© 2024 University of Cambridge

• Contact the University
• Accessibility
• Freedom of information
• Privacy policy and cookies
• Statement on Modern Slavery
• University A-Z
• Undergraduate
• Postgraduate
• Research news
• About research at Cambridge
• Spotlight on...

• VILLANOVA SCHOOL OF BUSINESS /
• Academic Programs /
• Executive Education /
• Artificial Intelligence and Machine Learning

GENERATIVE AI AND AI/ML DIGITAL UPSKILLING

Custom Master Class

Artificial intelligence (AI) and machine learning (ML) are rapidly transforming the business landscape. For business leaders, understanding these technologies is vital for harnessing their potential and addressing the challenges they bring. This master class is designed to help you maximize the value of Gen AI and AI/ML investments for your organization.​​

This program will guide you in identifying promising Gen AI and AI/ML opportunities, customizing AI/ML solutions to fit your unique business needs, and measuring the success of your initiatives. By the end of this course, you will have a detailed plan for an AI-driven business strategy, including its strategic rationale, implementation roadmap, change management strategy, and success metrics.​

Take the next step in your AI and digital transformation journey and develop the skills to lead in the age of Gen AI and AI/ML. As a business leader, your role is crucial in shaping your organization's future in this digital era. This master class will empower you to take charge and drive your organization toward success.

PROGRAM TOPICS

Understanding AI and Gen AI

Exploring machine learning, natural language processing, predictive AI, generative AI, and their applications.

Developing AI/ML Skills

Building practical skills for applying AI/ML in various business scenarios and functions.

Competitive Analysis

Recognizing and analyzing the use of AI/ML and Gen AI across the competitive landscape.

  

AI-Powered Enterprises

Examining the characteristics and best practices of successful AI-driven organizations.

Fostering an AI/ML Culture

Encouraging AI/ML and Gen AI thinking within teams and across the organization to drive innovation.

Cross-functional Teams

Forming and managing teams for AI/ML ideation, development, and implementation.

Opportunity Identification

Assessing and prioritizing high-value AI/ML opportunities within a business context.

Business Case Development

Preparing robust business cases to justify AI/ML investments and initiatives.

Customizing AI/ML Solutions

Tailoring AI/ML technologies and solutions to address unique business needs and challenges.

Data Preparation for AI/ML

Organizing and structuring data to ensure meaningful and actionable AI/ML outcomes.

Enterprise Patterns in AI/ML

Leveraging proven enterprise patterns to deploy scalable and effective AI/ML solutions.

ROI Measurement of AI/ML

Evaluating the return on investment for AI/ML projects to ensure value creation.

WHO SHOULD PARTICIPATE

This program is ideal for executives, senior managers, and high-potential professionals in any functional area or industry seeking to understand AI/ML opportunities. The program will be customized to focus on the unique opportunities for your functional area or organization and how to integrate artificial intelligence into your business strategies and operations.

CUSTOMIZING YOUR PROGRAM

We are deeply committed to creating meaningful value for your organization through a tailored AI/ML program. Our process includes:

    

We begin by listening to you about your organization, your learners, your challenges, and your vision. We seek to carefully understand your goals.  

Our team will collaborate with you on designing a curriculum, timeline, and budget that fit your needs. Options include in-person, online, or blended formats, delivered onsite at your organization or at our executive conference center.   

Our faculty will build a tailored learning experience through an iterative development process that is bolstered by valuable insights from you.  

These programs feature interactive lectures, case studies, innovative simulations, group exercises, or action learning projects to foster rich learning experiences.   

Our team will measure and maximize your organization’s return on investment through an evaluative process to assess the outcomes of your program.   

FACULTY SPOTLIGHT

Arup Das is a distinguished expert in Artificial Intelligence and Machine Learning, with a robust background in applied AI and a significant footprint in the generative AI industry. Currently serving as the Head of AI & Gen AI Industry Specialists at UiPath, Arup leads initiatives that drive AI and generative AI-based document and communication automation across various sectors, including financial services, healthcare, manufacturing, and the public sector. His work is focused on enhancing revenue growth, operational efficiency, and agility for large enterprises.

With over two decades of experience in technology leadership roles, Arup has left an indelible mark on the industry. Arup has been instrumental in developing AI and data-driven solutions, spearheading digital transformations, and building high-performing AI/ML teams. He has also successfully led multiple venture capital raises and exits, a testament to his strategic acumen. His professional journey includes significant stints at Avenue One, Compass, and Machine Analytics, where he developed low-code AI platforms and advanced natural language processing (NLP) solutions, further solidifying his reputation as a trailblazer in the field. Arup's expertise extends to digital strategy, product management, process engineering, and the development of modern data infrastructures, making him a well-rounded and highly sought-after consultant.

Arup is also an esteemed educator, holding a position as Professor of Applied AI & Generative AI at the Villanova School of Business. He is committed to nurturing the next generation of digital business leaders and AI practitioners, and his courses span various topics including AI ethics, business applications of AI/ML, and advanced NLP techniques. Arup has received several honors, including the Villanova University Thought Leader of the Year award. In addition to his academic roles, Arup is an accomplished author with upcoming books on generative AI and AI ethics.  

Arup holds an MBA from Cornell University, a Master's in Analytics from Villanova University, and a Master's in Computer Engineering from Stony Brook University.

Arup's passion for AI and technology, combined with his strategic vision and operational expertise, positions him as a leading voice in the AI industry. His dedication to solving complex business problems and driving innovation through AI is a testament to his ability to navigate the ever-evolving landscape of technology, inspiring confidence in his peers and collaborators.

EXECUTIVE INSIGHTS

Gain perspective on top management and leadership topics shared by VSB expert faculty in these episodes of the Inspiring Minds podcast.

TIMES UP FOR TOXIC WORKPLACES MANUELA PRIESEMUTH, PhD

Coping with the ride of incivility dean emeritus joyce e. a. russell, phd, closing the gender pay gap equitably david anderson, phd, "understanding modern work teams" narda quigley, phd.

Why Philosophy Is Crucial In The Age Of Artificial Intelligence

It seems they dont want rogue artificial superintelligences waging war on humanity, as in james camerons 1984 science fiction thriller, the terminator (ominously, arnold schwarzeneggers terminator is sent back in time from 2029)..

Philosophy is crucial in the age of artificial intelligence (AI)

New scientific understanding and engineering techniques have always impressed and frightened. No doubt they will continue to. OpenAI recently announced that it anticipates “superintelligence” – AI surpassing human abilities – this decade. It is accordingly building a new team , and devoting 20% of its computing resources to ensuring that the behaviour of such AI systems will be aligned with human values.

It seems they don't want rogue artificial superintelligences waging war on humanity, as in James Cameron's 1984 science fiction thriller, The Terminator (ominously, Arnold Schwarzenegger's terminator is sent back in time from 2029). OpenAI is calling for top machine-learning researchers and engineers to help them tackle the problem.

But might philosophers have something to contribute? More generally, what can be expected of the age-old discipline in the new technologically advanced era that is now emerging?

To begin to answer this, it is worth stressing that philosophy has been instrumental to AI since its inception. One of the first AI success stories was a 1956 computer program , dubbed the the Logic Theorist, created by Allen Newell and Herbert Simon. Its job was to prove theorems using propositions from Principia Mathematica, a 1910 a three-volume work by the philosophers Alfred North Whitehead and Bertrand Russell, aiming to reconstruct all of mathematics on one logical foundation.

Indeed, the early focus on logic in AI owed a great deal to the foundational debates pursued by mathematicians and philosophers.

One significant step was the German philosopher Gottlob Frege's development of modern logic in the late 19th century. Frege introduced the use of quantifiable variables – rather than objects such as people – into logic. His approach made it possible to say not only, for example, “Joe Biden is president” but also to systematically express such general thoughts as that “there exists an X such that X is president”, where “there exists” is a quantifier, and “X” is a variable.

Other important contributors in the 1930s were the Austrian-born logician Kurt Gödel, whose theorems of completeness and incompleteness are about the limits of what one can prove, and Polish logician Alfred Tarski's “proof of the indefinability of truth”. The latter showed that “truth” in any standard formal system cannot be defined within that particular system, so that arithmetical truth, for example, cannot be defined within the system of arithmetic.

Finally, the 1936 abstract notion of a computing machine by the British pioneer Alan Turing drew on such development and had a huge impact on early AI.

It might be said, however, that even if such good old fashioned symbolic AI was indebted to high-level philosophy and logic, the “second-wave” AI , based on deep learning, derives more from the concrete engineering feats associated with processing vast quantities of data.

Still, philosophy has played a role here too. Take large language models, such as the one that powers ChatGPT, which produces conversational text. They are enormous models, with billions or even trillions of parameters, trained on vast datasets (typically comprising much of the internet). But at their heart, they track – and exploit – statistical patterns of language use. Something very much like this idea was articulated by the Austrian philosopher Ludwig Wittgenstein in the middle of the 20th century: “the meaning of a word”, he said, “is its use in the language”.

But contemporary philosophy, and not just its history, is relevant to AI and its development. Could an LLM truly understand the language it processes? Might it achieve consciousness? These are deeply philosophical questions.

Science has so far been unable to fully explain how consciousness arises from the cells in the human brain. Some philosophers even believe that this is such a “hard problem” that is beyond the scope of science , and may require a helping hand of philosophy.

In a similar vein, we can ask whether an image generating AI could be truly creative. Margaret Boden, a British cognitive scientist and philosopher of AI, argues that while AI will be able to produce new ideas, it will struggle to evaluate them as creative people do.

She also anticipates that only a hybrid (neural-symbolic) architecture – one that uses both the logical techniques and deep learning from data – will achieve artificial general intelligence.

Human values

To return to OpenAI's announcement, when prompted with our question about the role of philosophy in the age of AI, ChatGPT suggested to us that (amongst other things) it “helps ensure that the development and use of AI are aligned with human values”.

In this spirit, perhaps we can be allowed to propose that, if AI alignment is the serious issue that OpenAI believes it to be, it is not just a technical problem to be solved by engineers or tech companies, but also a social one. That will require input from philosophers, but also social scientists, lawyers, policymakers, citizen users and others.

Indeed, many people are worried about the rising power and influence of tech companies and their impact on democracy. Some argue we need a whole new way of thinking about AI – taking into account the underlying systems supporting the industry. The British barrister and author Jamie Susskind, for example, has argued it is time to build a “ digital republic ” – one which ultimately rejects the very political and economic system that has given tech companies so much influence.

Finally, let us briefly ask, how will AI affect philosophy? Formal logic in philosophy actually dates to Aristotle's work in antiquity. In the 17th century. the German philosopher Gottfried Leibniz suggested that we may one day have a “calculus ratiocinator” – a calculating machine that would help us to derive answers to philosophical and scientific questions in a quasi-oracular fashion.

Perhaps we are now beginning to realise that vision, with some authors advocating a “computational philosophy” that literally encodes assumptions and derives consequences from them. This ultimately allows factual and/or value-oriented assessments of the outcomes.

For example, the PolyGraphs project simulates the effects of information sharing on social media. This can then be used to computationally address questions about how we ought to form our opinions.

( Author: Anthony Grayling , Professor of Philosophy, Northeastern University London and Brian Ball , Associate Professor of Philosophy AI and Information Ethics, Northeastern University London )

Promoted Listen to the latest songs, only on JioSaavn.com

( Disclosure Statement: Brian Ball receives funding from the British Academy, and has previously been supported by the Royal Society, the Royal Academy of Engineering, and the Leverhulme Trust. Anthony Grayling does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment)

This article is republished from The Conversation under a Creative Commons license. Read the original article .  

Track Budget 2023 and get Latest News Live on NDTV.com.

Track Latest News Live on NDTV.com and get news updates from India and around the world .

India Elections | Read Latest News on Lok Sabha Elections 2024 Live on NDTV.com . Get Election Schedule , information on candidates, in-depth ground reports and more - #ElectionsWithNDTV

Watch Live News:

Electrical and Computer Engineering

College of engineering, ms in artificial intelligence engineering.

The Master of Science in Artificial Intelligence–Electrical and Computer Engineering is a three-semester (97-unit) program that offers students the opportunity to gain state-of-the-art artificial intelligence knowledge from an engineering perspective. Today, AI is driving significant innovation across products, services, and systems in every industry, and tomorrow’s AI engineers will have the advantage.

ECE students within the program will learn how to design and build AI-orchestrated systems capable of operating within engineering constraints. At Carnegie Mellon, we are leading this transformation by teaching students how to simultaneously design a system’s functionality and supporting AI mechanisms, including both its AI algorithms and the platform on which the AI runs, to produce systems that are more adaptable, resilient, and trustworthy.

Students pursuing the MS in AIE will be able to:

• Demonstrate knowledge of artificial intelligence methods, systems, tool chains, and cross-cutting issues, including security, privacy, and other ethical, societal, and policy challenges
• Apply ECE concepts and tools to enable AI systems and produce AI tools
• Be informed practitioners of AI methods to solve ECE and related problems, applying ECE domain knowledge whenever possible to enhance AI effectiveness
• Understand the limits of AI systems and apply these techniques within these limits
• Evaluate trade-offs involving technical capabilities and limitations, policy, and ethics in artificial intelligent systems

Helpful links

Visit Pittsburgh

Pittsburgh handbook

Endless Opportunities

Whether pursuing academia or industry, this degree uniquely positions students for the future of research and high-demand careers with a mastery of integrating engineering domain knowledge into AI solutions.

For additional information about this college-wide initiative, please visit the College of Engineering's  MS in AI Engineering website .

Investing in Artificial Intelligence (AI) ETFs

There are several options for investors looking to diversify into ai..

• AI ETFs offer broad exposure to top AI companies, so you don't have to analyze individual stocks.
• Generative AI advances are driving tech firms to intensify AI tool development.
• Investing in AI-focused ETFs gives you access to key AI stocks and sectors.
• Motley Fool Issues Rare “All In” Buy Alert

You probably interact with artificial intelligence (AI) more often than you think. AI is powering the algorithm arranging your Netflix ( NFLX -1.79% ) menu, the software expediting your Amazon ( AMZN -8.79% ) package, and the brains behind many of the smartphone apps you use every day.

If you've used ChatGPT , the OpenAI chatbot that has wowed users by writing code and instantly answering complex questions, you've gotten a glimpse into the next frontier in AI, known as generative AI . Big tech companies, including Google's Gemini and Meta AI, and others are racing to develop AI chatbots and other generative AI technologies.

Artificial Intelligence

If you want portfolio exposure to AI companies but don't want to identify individual AI stocks, you can invest in an AI-focused exchange-traded fund (ETF) . AI ETFs provide exposure to a broad range of the best AI companies , so you don't need to research and choose individual stocks on your own.

Best AI ETFs

Best ai etfs to buy in 2024.

Data source: Yahoo! Finance. Data as of Aug. 1, 2024.

AI ETF	Assets Under Management	Expense Ratio
Global X Robotics & Artificial Intelligence ETF ( )	$2.63 billion	0.68%
ROBO Global Robotics and Automation Index ETF ( )	$1.25 billion	0.95%
iShares Robotics and Artificial Intelligence ETF ( )	$659.3 million	0.47%
First Trust Nasdaq Artificial Intelligence ETF ( )	$520.75 million	0.65%

Keep reading to learn more about each of these AI ETFs.

1. Global X Robotics and Artificial Intelligence ETF

1. global x robotics & artificial intelligence etf.

Established in 2016, the Global X Robotics & Artificial Intelligence ETF ( BOTZ -2.92% ) is a fund that invests in "companies that potentially stand to benefit from increased adoption and utilization of robotics and artificial intelligence." That includes enterprises working in industrial robotics, automation, nonindustrial robots, and autonomous vehicles .

Global X currently holds 44 stocks. Its top five holdings, which account for about 45% of the fund's assets , are:

• Nvidia ( NVDA -1.78% ): A semiconductor maker whose chips are used in a wide variety of applications -- including autonomous vehicles, virtual computing, and cryptocurrency mining -- and are central to many AI technologies
• ABB ( OTCPK:ABBN.Y ): Swiss maker of industrial automation and robotics products for use in utilities and infrastructure
• Intuitive Surgical ( ISRG -0.27% ): Maker of the da Vinci robotic surgical system, which allows for minimally invasive surgeries with precise control
• Keyence  ( KYCCF -2.42% ): A Japanese company that makes factory automation products, such as sensors and scanners
• SMC Corp ( OTC:SMCAY ): Japanese manufacturer of automatic control equipment for a variety of industrial applications

As the chart below shows, shares of the ETF have underperformed the S&P 500 index since its launch in 2016. The share price fell sharply in 2022, in line with the broad sell-off in tech stocks, although Global X has rebounded since then.

Global X offered a modest dividend yield of 0.3% at the time of this writing, but it is better suited to be a growth-oriented investment . Its expense ratio of 0.68% is higher than what you'd pay for an index fund, but it's also reasonable for the fund's performance history.

2. ROBO Global Robotics and Automation Index ETF

The ROBO Global Robotics and Automation Index ETF ( ROBO -3.52% ) is focused on companies driving "transformative innovations in robotics, automation, and artificial intelligence." ROBO invests in companies primarily focused on AI, cloud computing , and other technology companies.

Cloud Computing

ROBO holds 77 different stocks, with no single holding accounting for more than 2.2% of the ETF's value. Its top five holdings comprise only about 10% of the fund's total value. These five companies include Intuitive Surgical, the maker of the da Vinci surgical robot, and four others:

• Novanta ( NOVT -4.47% ): Novanta specializes in robotics and automation solutions across medicine, manufacturing, and robotics and automation.
• Zebra Technologies ( ZBRA -5.92% ): A technology company known for products such as barcodes and scanners, RFID tags and transponders, and related tracking devices.
• Fanuc (6954.T): A Japanese company that makes factory automation products including lasers, robots, and ultraprecision machines.
• ServiceNow ( NOW -2.56% ): One of the largest cloud software providers, ServiceNow has increasingly embraced automation and AI.

Since its inception in 2013, ROBO has underperformed the return of the S&P 500 , as the chart below shows. It trails the broad market index, with dividends factored into the return. ROBO pays a dividend yield of 0.05%, and its expense ratio is 0.95%.

3. iShares Robotics and Artificial Intelligence ETF

The iShares Robotics and Artificial Intelligence ETF ( IRBO -3.71% ) aims to track the results of an index of developed and emerging market companies that could benefit from long-term opportunities in robotics companies and AI.

The ETF was formed in 2018 and has less than $1 billion of assets under management . With 103 stock holdings, it's now well diversified. Many of its top holdings also give investors exposure to fast-growing small-cap companies.

The fund's top five investments, which account for about 7% of iShares Robotics' assets, include Nvidia and four others:

• Lumen Technologies ( LUMN -7.53% ): The telecom company formerly known as CenturyLink is known for dark fiber, edge cloud, and other broadband and telecom services.
• SiriusXM Holdings ( SIRI -3.41% ): The satellite radio company known for its satellite technology.
• Genius Sports ( GENI -7.69% ): A sports betting and sports media business using automatic production and distribution to commercialize video footage of games.
• Hello Group ( MOMO -1.66% ): The parent of Chinese online dating apps like TanTan and online media business.
• Pegasystems ( PEGA -2.02% ): A cloud software company focused on automation and AI.

As you can see from the chart below, iShares Robotics has underperformed the S&P 500 since its founding. The ETF fell in 2022 when tech stocks crashed.

The expense ratio is competitive at 0.47%, and the dividend yield at the time of this writing is 0.8%. The fund's performance will likely be heavily influenced by the overall performance of cloud stocks since it seems more exposed to cloud stocks and chipmakers than to AI companies.

4. First Trust Nasdaq AI and Robotics ETF

4. first trust nasdaq artificial intelligence & robotics etf.

First Trust Nasdaq Artificial Intelligence and Robotics ETF  ( ROBT -3.3% ) tracks the Nasdaq CTA Artificial and Robotics index, which is made up of companies engaged in AI and robotics in technology, industrials, and other sectors.

The fund currently holds 107 stocks, and the top five holdings include ServiceNow and Pegasystems as well as the following:

• SentinelOne ( S -2.36% ): A cybersecurity software company focused on endpoint security.
• Appian ( APPN 1.41% ): A cloud software company focused on business process automation through low-code tools and AI.
• Palantir ( PLTR -5.14% ): A cloud software company known for handling deep research and big data analytics.

Related investing topics

Investing in tech etfs.

Here's how to get tech exposure without the single-stock risk.

Cheap ETFs to Buy in 2024

Get a list of inexpensive buys among the market's top ETFs.

AI Investing: A Complete Beginner's Guide

This beginner's look at AI investing reviews the ins and outs of AI to help you decide whether investing in AI is right for you.

How to Invest in OpenAI

All you need to know about how to invest in OpenAI, creator of ChatGPT.

The First Trust ETF offers an expense ratio of 0.65% and a dividend yield of 0.27%. Although its trading history is relatively short, you can see from the chart below that its performance has lagged the S&P 500 recently.

Should I buy?

Should i buy ai etfs.

The best way to decide which ETF to buy is to consider which stocks a fund holds and how many of them are true AI companies . A fund's expense ratio, dividend yield, and past performance are also important, and you can opt to invest in a basket of all four of these AI ETFs to maximize your diversification.

Over time, AI, like chatbots , will grow only smarter and play a greater role in our daily lives. Already AI represents a global market worth hundreds of billions of dollars, and its wide range of practical applications includes face recognition, predictive algorithms in internet search, smart home devices, and autonomous vehicles. So pay attention to the AI market now, and you may find yourself reaping the rewards in years to come.

Investing in Artificial Intelligence ETFs: FAQ

Which etf is best for ai.

AI investors have several options in ETFs. The best-known of the AI ETFs above is Global X, which holds a number of well-known AI stocks, including Nvidia and Intuitive Surgical.

AI investors may also want to consider an ETF that tracks the Nasdaq-100 , such as the Invesco QQQ ETF (NASDAQ: QQQ), because big tech companies with exposure to AI make up almost half of the fund.

Does Vanguard have an AI ETF?

Vanguard does not currently offer an AI-focused ETF. However, the asset manager offers an information technology ETF that includes several AI stocks.

What is the best AI to invest in?

The best-known AI stock right now is Nvidia, and it's also been the most successful stock in AI. Past performance does not guarantee future returns, but it makes sense to invest in ETFs with exposure to Nvidia and other AI chip stocks as they emerge.

Does Charles Schwab have an AI ETF?

Charles Schwab does not have an AI ETF. However, the brokerage firm does have an AI "theme" that contains as many as 25 AI stocks that Schwab account holders can buy together, based on Schwab's proprietary algorithms and research.

Invest Smarter with The Motley Fool

Join over half a million premium members receiving….

• New Stock Picks Each Month
• Detailed Analysis of Companies
• Model Portfolios
• Live Streaming During Market Hours
• And Much More

Motley Fool Investing Philosophy

• #1 Buy 25+ Companies
• #2 Hold Stocks for 5+ Years
• #3 Add New Savings Regularly
• #4 Hold Through Market Volatility
• #5 Let Winners Run
• #6 Target Long-Term Returns

Why do we invest this way? Learn More

Related Articles

Motley Fool Returns

Market-beating stocks from our award-winning analyst team.

Calculated by average return of all stock recommendations since inception of the Stock Advisor service in February of 2002. Returns as of 08/05/2024.

Discounted offers are only available to new members. Stock Advisor list price is $199 per year.

Calculated by Time-Weighted Return since 2002. Volatility profiles based on trailing-three-year calculations of the standard deviation of service investment returns.

Premium Investing Services

Invest better with The Motley Fool. Get stock recommendations, portfolio guidance, and more from The Motley Fool's premium services.

We use cookies to ensure best experience for you

We use cookies and other tracking technologies to improve your browsing experience on our site, show personalize content and targeted ads, analyze site traffic, and understand where our audience is coming from. You can also read our privacy policy , We use cookies to ensure the best experience for you on our website.

• SOUTHEAST ASIA
• Leaders Speak
• Brand Solutions
• Artificial Intelligence
• Philosophy is crucial in the age of AI

• Updated On Aug 3, 2024 at 03:20 AM IST

New scientific understanding and engineering techniques have always impressed and frightened. No doubt they will continue to. Open AI recently announced that it anticipates "superintelligence" - AI surpassing human abilities - this decade. It is accordingly building a new team, and devoting 20% of its computing resources to ensuring that the behaviour of such AI systems will be aligned with human values. It seems they don't want rogue artificial superintelligences waging war on humanity, as in James Cameron's 1984 science fiction thriller, The Terminator (ominously, Arnold Schwarzenegger's terminator is sent back in time from 2029). OpenAI is calling for top machine-learning researchers and engineers to help them tackle the problem. Advt But might philosophers have something to contribute? More generally, what can be expected of the age-old discipline in the new technologically advanced era that is now emerging? To begin to answer this, it is worth stressing that philosophy has been instrumental to AI since its inception. One of the first AI success stories was a 1956 computer program, dubbed the the Logic Theorist, created by Allen Newell and Herbert Simon. Its job was to prove theorems using propositions from Principia Mathematica, a 1910 a three-volume work by the philosophers Alfred North Whitehead and Bertrand Russell, aiming to reconstruct all of mathematics on one logical foundation. Indeed, the early focus on logic in AI owed a great deal to the foundational debates pursued by mathematicians and philosophers. One significant step was the German philosopher Gottlob Frege's development of modern logic in the late 19th century. Frege introduced the use of quantifiable variables - rather than objects such as people - into logic. His approach made it possible to say not only, for example, "Joe Biden is president" but also to systematically express such general thoughts as that "there exists an X such that X is president", where "there exists" is a quantifier, and "X" is a variable. Advt Other important contributors in the 1930s were the Austrian-born logician Kurt Godel, whose theorems of completeness and incompleteness are about the limits of what one can prove, and Polish logician Alfred Tarski's "proof of the indefinability of truth". The latter showed that "truth" in any standard formal system cannot be defined within that particular system, so that arithmetical truth, for example, cannot be defined within the system of arithmetic. Finally, the 1936 abstract notion of a computing machine by the British pioneer Alan Turing drew on such development and had a huge impact on early AI. Advt It might be said, however, that even if such good old fashioned symbolic AI was indebted to high-level philosophy and logic, the "second-wave" AI, based on deep learning, derives more from the concrete engineering feats associated with processing vast quantities of data. Still, philosophy has played a role here too. Take large language models, such as the one that powers ChatGPT, which produces conversational text. They are enormous models, with billions or even trillions of parameters, trained on vast datasets (typically comprising much of the internet). But at their heart, they track - and exploit - statistical patterns of language use. Something very much like this idea was articulated by the Austrian philosopher Ludwig Wittgenstein in the middle of the 20th century: "the meaning of a word", he said, "is its use in the language". Advt But contemporary philosophy, and not just its history, is relevant to AI and its development. Could an LLM truly understand the language it processes? Might it achieve consciousness? These are deeply philosophical questions. Science has so far been unable to fully explain how consciousness arises from the cells in the human brain. Some philosophers even believe that this is such a "hard problem" that is beyond the scope of science, and may require a helping hand of philosophy. In a similar vein, we can ask whether an image generating AI could be truly creative. Margaret Boden, a British cognitive scientist and philosopher of AI, argues that while AI will be able to produce new ideas, it will struggle to evaluate them as creative people do. She also anticipates that only a hybrid (neural-symbolic) architecture - one that uses both the logical techniques and deep learning from data - will achieve artificial general intelligence. Human values To return to OpenAI's announcement, when prompted with our question about the role of philosophy in the age of AI, ChatGPT suggested to us that (amongst other things) it "helps ensure that the development and use of AI are aligned with human values". In this spirit, perhaps we can be allowed to propose that, if AI alignment is the serious issue that OpenAI believes it to be, it is not just a technical problem to be solved by engineers or tech companies, but also a social one. That will require input from philosophers, but also social scientists, lawyers, policymakers, citizen users and others. Indeed, many people are worried about the rising power and influence of tech companies and their impact on democracy. Some argue we need a whole new way of thinking about AI - taking into account the underlying systems supporting the industry. The British barrister and author Jamie Susskind, for example, has argued it is time to build a "digital republic" - one which ultimately rejects the very political and economic system that has given tech companies so much influence. Finally, let us briefly ask, how will AI affect philosophy? Formal logic in philosophy actually dates to Aristotle's work in antiquity. In the 17th century. the German philosopher Gottfried Leibniz suggested that we may one day have a "calculus ratiocinator" - a calculating machine that would help us to derive answers to philosophical and scientific questions in a quasi-oracular fashion. Perhaps we are now beginning to realise that vision, with some authors advocating a "computational philosophy" that literally encodes assumptions and derives consequences from them. This ultimately allows factual and/or value-oriented assessments of the outcomes. For example, the PolyGraphs project simulates the effects of information sharing on social media. This can then be used to computationally address questions about how we ought to form our opinions. Certainly, progress in AI has given philosophers plenty to think about; it may even have begun to provide some answers. (The Conversation) NSA NSA

• Published On Aug 3, 2024 at 03:17 AM IST

All Comments

By commenting, you agree to the Prohibited Content Policy

Find this Comment Offensive?

• Foul Language
• Inciting hatred against a certain community
• Out of Context / Spam

Join the community of 2M+ industry professionals

Subscribe to our newsletter to get latest insights & analysis., download etcio app.

• Get Realtime updates
• Save your favourite articles

We've detected unusual activity from your computer network

To continue, please click the box below to let us know you're not a robot.

Why did this happen?

Please make sure your browser supports JavaScript and cookies and that you are not blocking them from loading. For more information you can review our Terms of Service and Cookie Policy .

For inquiries related to this message please contact our support team and provide the reference ID below.

• Graz University of Technology
• Posted on: 1 August 2024

7 PhD positions in the LEAD project DigiBioTech at Graz University of Technology in the field of Artificial Intelligence-driven Biotechnology

Job Information

Offer description.

The LEAD project DigiBioTech is an excellence program funded by the Graz University of Technology, Austria, and invites applications for 3 PhD candidates in the field of Artificial Intelligence and Computer Science and 7 PhD candidates in the field of Biotechnology and Molecular Biology.

DigiBioTech aims to create a leading hub for the application of Artificial Intelligence in Biotechnology. It joins the expertise of researchers from the Graz University of Technology in biotechnology, bioinformatics, process engineering, and computer science to conquer new ground in the prediction of catalytic properties of enzymes and understanding complex interactions in bioprocesses. In this interdisciplinary framework, PhD students will collaborate with scientists from various disciplines to understand the broader implications of the research. 

Computationally Designed de novo Enzymes for Bioremediation (DigiBioTech PhD-#01)

Computational protein design has an outstanding potential to create superior biological materials with customized properties and new-to-nature chemical reactions. The PhD candidate will employ cutting-edge computational protein design techniques to create and test de novo enzymes specifically aimed at dechlorinating persistent environmental pollutants. 

Unraveling Epistatic Interactions Using Deep Mutational Scanning of Enzymes (DigiBioTech PhD-#03)

This project will employ gene-site saturation mutagenesis to systematically study the effects of mutations across proteins. The PhD candidate will develop growth assays and design and screen enzyme libraries and employ a high throughput sequencing platform developed by PhD#4. The large datasets obtained in this project will be used to train machine learning models to enable targeted enzyme engineering.

Engineering Enzymes for the Removal of Toxic C-F Containing Compounds (DigiBioTech PhD-#05)

This project aims to enhance the catalytic activity of redox enzymes for degrading fluorinated pollutants. The PhD candidate will create enzyme libraries based on an evolutionary sequence space of the target enzymes, and screen them for increased activity towards carbon-fluoride bonds present in “forever chemicals” like PFAS. Data will be used for a targeted search approach with PhD#6 to predict improved enzyme variants. 

AI-guided Discovery and Prediction of Novel Enzymes in Microbiomes of Different Habitats (DigiBioTech PhD-#07)

This PhD project aims to predict the natural ability of microbiomes to degrade hazardous chemicals containing carbon-halogen bonds. The PhD candidate will compile metagenomic data from various environments to train AI models on these sequence data.  By conducting laboratory experiments they will enrich the available data basis and explore the metagenomic sequence space to discover candidate taxa that produce pollutant-degrading enzymes.

Automated Self-Optimization of Enzyme Cascade Transformations (DigiBioTech PhD-#08) 

Reaction telescoping enhances the sustainability of biocatalytic processes by performing multiple reactions in one pot. In this project, the PhD candidate will design and conduct multistep-enzymatic experiments and apply machine learning-based self-optimization methods to improve the efficiency of enzyme cascades.

Development of model-based control and estimation algorithms for CO2-fixating bioprocesses (DigiBioTech-#09)

Gas fermentation uses microorganisms to convert CO2 into valuable chemicals and fuels. Within this project, the PhD candidate will develop mathematical models for the design of automatic control and estimation algorithms suitable for gas fermentation processes. Furthermore, lab-scale gas fermentation experiments will be carried out and optimization of this complex process through the innovative integration of computational fluid dynamics and machine-learning-guided scale-up developed by PhD#10 will be considered.

Development of data-driven and physically-motivated models (DigiBioTech PhD-#10)

The PhD candidate will develop data-driven and physically motivated models (e.g., based on computational fluid dynamics) that allow real-time predictions and fast simulations of gas fermentation processes. AI-based compartment models will help to characterize processes in detail and with extreme speed. This enables - inter alia - virtual testing of different reactor designs and very early cost-benefit analysis on a production scale.

Candidate profile and requirements: 

Applicants must hold a completed master’s degree. Students who have not yet finished their masters' degree can apply but must graduate before their appointment. 

Application deadline:  August 31st, 2024

Hours per week: 30 h/w

Employment Start: November 2024

Contract Duration: 3 years

We offer a minimum annual gross salary based on full-time of € 50.103,20, overpayment possible depending on qualification and experience. 

Application:

Please note that the specific job profiles and tasks may be published separately on the job portal at Graz University of Technology. Please visit the website  jobs.tugraz.at for more information.

Graz University of Technology aims to increase the proportion of women, in particular in management and academic staff, and therefore qualified female applicants are explicitly encouraged to apply. Preference will be given to women if applicants are equally qualified.

Graz University of Technology actively promotes diversity and equal opportunities. Applicants are not to be discriminated against in personnel selection procedures on the grounds of gender, ethnicity, religion or ideology, age, sexual orientation (Anti-discrimination).

People with disabilities and who have the relevant qualifications are expressly invited to apply.

Where to apply

Requirements, additional information, work location(s), share this page.