JOHN MCCARTHY (1927-2011)

BY NILS J. NILSSON

JOHN MCCARTHY, widely known as the father of artificial intelligence, died on Oct. 24, 2011, at his home in Stanford, California. He was 84 years old.

He was born on Sept. 4, 1927, in Boston, Massachusetts. His father, John Patrick McCarthy, was an Irish Catholic immigrant. He worked as a carpenter, a fisherman, and a union organizer, moving the family from Boston to New York City and then to Los Angeles. In Los Angeles, John’s father found work as an organizer for the Amalgamated Clothing Workers. John’s mother, Ida Glatt, a Lithuanian Jewish immigrant, worked as a journalist for The Federated Press wire service, then for a Communist newspaper, and finally as a social worker. Both parents were active members of the Communist Party during the 1930s. John, along with his younger brother, Patrick, grew up in a politically charged household that encouraged learning and critical thinking. John reported that even before he attended high school he got interested in science by reading a translation of a Russian popular science book for children called 100,000 Whys.¹

In 1943, John graduated two years early from Belmont High School in Los Angeles. While in high school, he taught himself calculus from books used for freshmen and sophomores at Caltech, which was nearby. In 1944 John entered Caltech as a mathematics major and was, not surprisingly, able to skip the first two years of mathematics courses. John was reportedly suspended from Caltech for failure to attend physical education courses. After serving in the U.S. Army, he was readmitted to Caltech, receiving a B.S. in mathematics in 1948.

John became fascinated with the idea of developing machines that could think like people while taking some graduate courses at Caltech and attending the Hixon Symposium on Cerebral Mechanisms in Behavior in September 1948.

After one year of graduate work at Caltech, McCarthy enrolled at Princeton University in 1949 because he thought it was the best place for his chosen field of mathematics. Within his first year he independently decided on a thesis topic involving a problem in partial differential equations, which, as he said, made him “relatively popular, because the usual thing is that the students pester the professors to suggest topics.” He finished his Ph.D. work under Solomon Lefschetz in 1951 with a dissertation entitled “Projection Operators and Partial Differential Equations.” In addition to his thesis work at Princeton, McCarthy pursued his fascination with mechanizing intelligence, which had begun at Caltech.

McCarthy first met Marvin Minsky (NAE 1989), a beginning graduate student at Princeton, in 1951. They shared an interest in mechanizing intelligence and collaborated over the next decade, although their approaches to making machines intelligent ultimately diverged significantly.

After finishing his Ph.D. work, McCarthy stayed on at Princeton for two years as an instructor. He worked at Bell Labs during the summer of 1952, where he met Claude Shannon (NAE 1985). They collaborated on a volume of papers on automata, published as Automata Studies (Princeton University Press, 1956). McCarthy was disappointed that most of the papers were on the theory of automata and only one or two related to his chief interest, machine intelligence.

McCarthy became an acting assistant professor of mathematics at Stanford University in 1953, where he continued his research on differential equations and analysis. He moved again in 1955, this time to Dartmouth College.

Even with his interest in all things computational, McCarthy didn’t actually try to program a computer until 1955, when IBM decided to make an IBM 704 computer available for research and education at MIT, other New England colleges, and IBM. As the Dartmouth representative, McCarthy met Nathaniel Rochester, head of IBM’s Information Research Department in Poughkeepsie, New York, and was invited to spend the summer of 1955 with his group at IBM.

During that summer McCarthy and Rochester persuaded Claude Shannon and Marvin Minsky, then a Harvard junior fellow in mathematics and neurology, to join them in proposing a workshop to the Rockefeller Foundation to be held at Dartmouth the following summer: “Summer Research Project on Artificial Intelligence.” Among other things, the proposal stated that the

“study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it …For the present purpose the artificial intelligence problem is taken to be that of making a machine behave in ways that would be called intelligent if a human were so behaving.”

While the workshop did not immediately lead to great breakthroughs, it provided a critical venue for a few researchers working in the field to meet, talk, and plan future research in artificial intelligence, cementing their commitment and laying the groundwork for their seminal contributions to defining the discipline. Although others had written earlier about mechanizing intelligence, the Dartmouth workshop is generally considered to be the beginning of serious work in the field — a field to which McCarthy devoted his scientific career.

While at Dartmouth, McCarthy became a Sloan fellow in physical science and spent his fellowship at MIT, where he had access to an IBM computer and could interact with Marvin Minsky, then at Harvard. This led to his appointment in 1958 as assistant professor of communication science at MIT.

McCarthy spent the summer of 1958 at IBM, where he first began to recognize the need for a new programming language that could support recursion and dynamic storage. Back at MIT in the fall of 1958, he began work on a new language, one he called LISP (for “list processor”). Besides recursion, programs written in LISP could process arbitrary symbolic structures and could treat programs (written in LISP) as data (expressed as lists in LISP). John’s 1960 paper, “Recursive Functions of Symbolic Expressions and Their Computation by Machine, Part I,” established the theoretical foundations of LISP as a universal computational formalism.² LISP soon became the language of choice for AI research.

In 1957 and 1958 at MIT, McCarthy provided the first suggestions for implementing timesharing, allowing a single computer to be simultaneously shared among several users.³ In 1962 these ideas were embodied in MIT’s Compatible Time-Sharing System (CTSS) as well as in later systems that McCarthy helped build at Bolt, Beranek and Newman and at Stanford. CTSS led directly to the creation of Project MAC, which revolutionized computing at MIT.

In 1959 John invented a technique he called “garbage collection,” in which random access memory is freed up by removing code that subsequent computations won’t need. That technique was added to LISP and is still routinely used in Java and other programming languages. McCarthy also made substantial contributions to the algebraic languages ALGOL 58 and 60.

Even before the Dartmouth workshop, McCarthy became interested in formulating a system that could be given statements (corresponding to English sentences) in some suitable language and then reason with these statements. At a 1958 conference in Teddington, England, he gave a paper titled “Programs with Commonsense” (often referred to as the Advice Taker paper) proposing just such a system.⁴  Although his proposal was called “half-baked” at the conference by the philosopher and linguist Yehoshua Bar-Hillel, it would go on to constitute significant parts of the subsequent AI literature and occupy the research lives of McCarthy and many others pursuing what has been called the “logicist” approach to AI. The paper is now seen by many as marking the birth of the field of knowledge representation.⁵

Complementing McCarthy’s work in AI were his efforts at establishing a mathematical theory of computation. Although Alan Turing and others had done earlier work on what could be computed and by what processes, McCarthy wanted to be able to treat computer programs as mathematical objects that could be proved “correct,” that is, the mathematical proof that programs meet the specifications set for them. He wrote three important papers on this topic, culminating in a paper describing the first proof of the correctness of a compiler.⁶

In the fall of 1962 John returned to Stanford as a full professor in the Computer Sciences Division of the Mathematics Department. In 1965 he became an inaugural member of Stanford’s new Department of Computer Science.

As McCarthy’s various AI and related projects grew, he formed the Stanford Artificial Intelligence Laboratory (SAIL). For his own research at the lab, McCarthy mainly followed up on the ideas first proposed in his earlier paper, “Programs with Commonsense.” Most of his AI work at Stanford focused on the problem of implementing commonsense reasoning, using first-order logic with a number of extensions.

Although not personally involved in robotics, McCarthy encouraged research on it at SAIL. One of the most successful robotic achievements at SAIL was the assembly of a Model T Ford racing water pump using a vision system and an electromechanical hand.⁷

During the late 1960s and 1970s, SAIL was a shining example of what a community of very bright faculty, students, and staff could accomplish with adequate funding, powerful computers, peripherals, and associated software, much of it developed at SAIL. John’s philosophy in “managing” SAIL was to let a thousand flowers bloom. He often gave good advice to people about how to tackle a problem, but if they took an alternative approach that was successful, he was equally pleased. The foundations for graphical user interfaces and printers, computer typesetting and publishing, speech recognition, computer vision and robotics, computer music, and other technologies that are now ubiquitous all got their start at SAIL’s facilities. SAIL was also one of the first nodes on the ARPAnet, a precursor to the modern internet. SAIL produced many Ph.D.s and other graduates. Sixteen Turing awards from the Association for Computing Machinery were given to people who had been affiliated with SAIL.

John was a philosopher in the old Greek sense of the word: one who loves wisdom. Like the ancient Greeks debating philosophy in the marketplace, John’s arena was the electronic marketplace of bulletin boards, newsletters, and the web. He was particularly passionate about free speech. When the Stanford administration curtailed access on some Stanford computers to the jokes newsletter called rec.humor.funny, John organized a successful petition to reinstate it. Much of John’s writing is contained in memos and other material on his many web pages, which continue to be accessible from www-formal.stanford.edu/jmc.

Along with his work in AI and computer science, McCarthy was deeply interested in the connections between AI, philosophy, and cognitive science. He wrote on one of his web pages, “It turns out that many philosophical problems take new forms when thought about in terms of how to design a robot.”⁸

McCarthy, along with most AI researchers and many philosophers, believed that all animals, including humans, were deterministic machines. Some people worry, for example, that machines cannot have free will. According to McCarthy, free will involves considering different courses of action and having the ability to choose among them. He summarized his position about human free will by quoting his daughter, Sarah, who said at age 4, “I can, but I won’t.” Even some chess programs have this kind of free will (even though they are completely deterministic). In two memos McCarthy claimed that useful robots would also need free will.⁹ About consciousness, McCarthy was of the opinion that “thinking about consciousness with a view to designing it provides a new approach to some of the problems of consciousness studied by philosophers.” In particular, “From the AI point of view, consciousness must be regarded as a collection of interacting processes rather than the unitary object of much philosophical speculation.”¹⁰ But he also argued that “robots should not be programmed to have emotions or to behave so as to have emotion ascribed to them.”¹¹

John McCarthy’s many awards included the Association for Computing Machinery’s Turing Award (1971), the first International Joint Conference on Artificial Intelligence Award for Research Excellence (1985), the Kyoto Prize (1988), the National Medal of Science (1990), the Benjamin Franklin Medal in Computer and Cognitive Sciences (2003), and memberships in the American Academy of Arts and Sciences (1974), the National Academy of Engineering (1987), and the National Academy of Sciences (1989). He was a founding fellow and past president of the Association for the Advancement of Artificial Intelligence. He was named the Charles M. Pigott Professor of Engineering by Stanford in 1987.

He is survived by his third wife, Carolyn Talcott of Stanford; two daughters, Susan McCarthy of San Francisco and Sarah McCarthy of Nevada City, California; a son, Timothy Talcott McCarthy of Stanford; a brother, Patrick, of Los Angeles; two grandchildren; and his first wife, Martha Coyote. His second wife, Vera Watson, died in 1978 in a mountain-climbing accident attempting to scale Annapurna in Nepal.

John McCarthy’s genius, puckish humor, and presence, along with his provocations to think more deeply, will be greatly missed by his colleagues, family, and many friends.

__________________________
¹Much of the information in this tribute is from the transcript, “Oral History of John McCarthy,” of an interview by Nils Nilsson on Sept. 12, 2007, at and for the Computer History Museum, Mountain View, California.
²McCarthy J. 1960. Recursive functions of symbolic expressions and their computation by machine. I. Commun. ACM 3(4):184-195.
³McCarthy’s recollections about his suggestions for timesharing can be found at www-formal.stanford.edu/jmc/history/timesharing/timesharing.html.
⁴McCarthy J. 1959. Programs with commonsense. In: Proceedings of the Symposium on Mechanisation of Thought Processes, vols. 1 and 2. Blake DV, Uttley AM, eds. London: Her Majesty’s Stationary Office.
⁵Hayes PJ, Morgenstern L. 2007. On John McCarthy’s 80th birthday, in honor of his contributions. AI Mag. 28(4):93-102.
⁶McCarthy J. 1963. A basis for a mathematical theory of computation. In: Computer Programming and Formal Systems. Braffort P,  Hirschenberg D, eds. Amsterdam: North-Holland; McCarthy J. 1962. Towards a mathematical science of computation. In: Information Processing 62: Proceedings of IFIP Congress 1962, 21-28; McCarthy J, Painter J. 1967. Correctness of a compiler for arithmetic expressions. In: Mathematical Aspects of Computer Science 1. Proc. Symp. Appl. Math., vol. 19. American Mathematical Society.
⁷Bolles R, Paul R. 1973. The use of sensory feedback in a programmable assembly system. Stanford AI Laboratory Memo AIM-220, Stanford Computer Science Department Report STAN-CS-396, Oct.
⁸McCarthy J. 1996. What has AI in common with philosophy. Unpublished memo, April 23.
⁹See: McCarthy J. 2000. Free will—even for robots, unpublished memo, Feb 14; McCarthy J. 2002. Simple deterministic free will, unpublished memo, May 16. Online at www-formal.stanford.edu/jmc/freewill.html.
¹⁰McCarthy J. 1995. Todd Moody’s zombies. J. Consciousness Stud. 2(4):345-347.
¹¹McCarthy J. 1995. Making robots conscious of their mental states. Paper presented at Machine Intelligence 15 workshop, Oxford University, UK, July 24. Online at jmc.stanford.edu/articles/consciousness.html.