Allen Newell

Allen Newell (March 19, 1927 - July 19, 1992) was an American computer scientist and cognitive psychologist who co-created the Logic Theorist, widely described as the first artificial intelligence program, and spent the following four decades trying to explain human thinking by building computer programs that reproduced it. Working first at the RAND Corporation and then at Carnegie Mellon University, he and his long-time collaborator Herbert A. Simon produced a string of foundational systems and ideas: the Information Processing Languages (the first list-processing languages), the General Problem Solver and means-ends analysis, the physical symbol system hypothesis, the book Human Problem Solving (1972), and the Soar cognitive architecture ^[1][4].

Newell and Simon shared the 1975 ACM Turing Award for "basic contributions to artificial intelligence, the psychology of human cognition, and list processing" ^[9]. Newell was also the first president of the American Association for Artificial Intelligence (AAAI), the 1989 recipient of the IJCAI Award for Research Excellence, and a 1992 recipient of the National Medal of Science, awarded the year he died ^[1][3][4]. Beyond his own research he was one of the principal builders of Carnegie Mellon's computer science department, a central figure in ARPA's speech-recognition program of the 1970s, and, with Stuart Card and Thomas Moran, a founder of the scientific study of human-computer interaction ^[1].

Early life and education

Newell was born in San Francisco on March 19, 1927, the son of Robert R. Newell, a professor of radiology at Stanford Medical School, and Jeanette Le Valley Newell. He attended Lowell High School in San Francisco, where he met Noël McKenna; they married when he was 20 and remained married for 45 years, raising a son, Paul ^[1].

He graduated from high school as World War II was ending, worked a summer in a shipyard, and enlisted in the U.S. Navy. Serving on a ship that carried scientific observers to the Bikini Atoll nuclear tests, he was assigned to map the radiation distribution over the atolls, an experience he later credited with turning him toward science ^[1]. He then studied physics at Stanford University, publishing his first paper, on X-ray optics, in 1949, the year he took his BS. At Stanford he also took classes from the mathematician George Polya, whose 1945 book How to Solve It introduced him to heuristic reasoning, the art of guided search and discovery. That idea stayed at the center of his work for the rest of his life ^[1][2].

After a year of graduate work in mathematics at Princeton (1949-50), where he encountered the new game theory of von Neumann and Morgenstern, Newell decided he preferred experimental and applied research to pure mathematics. He left for the RAND Corporation in Santa Monica in 1950 ^[1][2].

RAND and the turn to artificial intelligence

At RAND, Newell first worked on Air Force logistics problems, coauthoring technical reports on formal organization theory with Joseph B. Kruskal. Field visits convinced him that axiomatic models were too far from organizational reality, and he moved to laboratory experiments on decision making in small groups. With John Kennedy, Robert Chapman, and William Biel he conceived a full-scale simulation of an Air Force early warning station, funded in 1952, to study how radar crews processed information under load. The project created RAND's Systems Research Laboratory, which was later spun off as the System Development Corporation ^[1].

The project shaped Newell's future in two ways. Herbert Simon, then a professor at Carnegie Institute of Technology, met Newell in 1952 when he became a consultant to the laboratory, and the two quickly found common ground in treating information processing as the core of human decision making. And the need to simulate radar displays led Newell to Cliff Shaw, a RAND systems programmer, with whom he generated simulated radar maps on a Card-Programmed Calculator. The exercise convinced both men that computers were general symbol manipulators, not just calculators ^[1].

In September 1954 Newell attended a RAND seminar in which Oliver Selfridge described a running pattern-recognition program. Newell later called it his "conversion experience": it became instantly clear to him, he said, "that intelligent adaptive systems could be built that were far more complex than anything yet done." From that day he committed himself to understanding human thinking by simulating it ^[1]. Within months he had written "The Chess Machine: An Example of Dealing with a Complex Task by Adaptation," a design for a chess program built around goals, subgoals, and satisficing evaluation, which he presented at the Western Joint Computer Conference in March 1955. Both Alan Turing and Claude Shannon had described chess-playing strategies on paper around 1950, but Newell's aim was different: chess was a vehicle for understanding the mechanisms needed for "ultracomplicated problems" ^[1].

The Logic Theorist and the first list-processing languages

Early in 1955 Newell moved to Pittsburgh to work with Simon at Carnegie Institute of Technology and to complete a doctorate, while RAND kept him on staff as a one-man Pittsburgh outpost, an arrangement that lasted until 1961. The planned chess project turned instead to proving theorems in symbolic logic. With Shaw contributing systems programming at RAND, the three built the Logic Theory Machine, better known as the Logic Theorist: it existed as a hand simulation by December 15, 1955, and ran as a program on RAND's JOHNNIAC computer in the summer of 1956 ^[1].

The Logic Theorist proved 38 of the first 52 theorems in chapter 2 of Whitehead and Russell's Principia Mathematica, and found a proof of Theorem 2.85 shorter than the one in the book, which reportedly delighted Bertrand Russell. It is widely described as the first artificial intelligence program: it searched for proofs selectively, using heuristics to prune a search tree that would otherwise explode combinatorially ^[5]. Newell and Simon presented the work at the 1956 Dartmouth summer workshop, where John McCarthy had coined the term "artificial intelligence"; Simon recalled that the reception was muted ^[5]. (For reasons no longer clear, Shaw was not a coauthor of the first Logic Theorist paper, but Simon always described him as a full partner in the research ^[1].)

To implement the program, the team invented the Information Processing Languages (IPLs), the first list-processing programming languages. The IPLs introduced ideas that became fundamental to computer science: lists, associations, schemas, dynamic memory allocation, recursion, functions as arguments, and generators. Lisp, which McCarthy developed in 1958, embedded list processing in the lambda calculus, improved the syntax, added garbage collection, and soon displaced IPL as the standard AI language ^[1][5]. Newell's doctoral work incorporated the Logic Theorist research; he received his PhD in industrial administration from Carnegie Institute of Technology in 1957 ^[1][2].

General Problem Solver and Human Problem Solving

During a 1957 workshop at Carnegie Tech, Newell and Simon extracted from the recorded protocol of a person solving logic problems a mechanism they called means-ends analysis: compare the current situation with the goal, find a difference, retrieve an operator known to reduce differences of that kind, apply it, and repeat. The idea became the core of the General Problem Solver (GPS), first described in a conference report published in 1960. GPS separated general problem-solving strategy from domain knowledge: given a problem space, a set of operators, and a table connecting operators to differences, it could attack problems in many domains ^[1].

The Newell-Shaw-Simon group also produced the NSS chess program (named for its authors' initials), described in 1958. It was not the first chess program to run, and it was not strong; critics of AI liked to note that it once lost to a ten-year-old. Its purpose was to show that highly selective, heuristic-guided search could produce intelligent behavior in a complex task ^[1]. Alongside the programs, the group pioneered thinking-aloud protocol analysis, a method for testing theories of cognition against second-by-second traces of human reasoning, launching systematic experiments by 1957 and developing formal encodings such as problem behavior graphs ^[1].

Newell and Simon's last major joint project summarized this research program in Human Problem Solving (1972), a book that presented their theory of human cognition as heuristic search through problem spaces, grounded in protocol data. Their collaboration tapered off afterward without any rift; the two talked weekly until a few days before Newell's death ^[1].

Carnegie Mellon, computer structures, and human-computer interaction

Newell joined the Carnegie Institute of Technology faculty in 1961 and spent the rest of his career there as the institution became Carnegie Mellon University. He held the Helen Whitaker professorship in computer science ^[2]. He was, with Simon and Alan Perlis, a principal figure in building what became one of the leading computer science departments in the United States, obtaining and renewing the large ARPA grants that funded it and chairing the task force whose February 1982 report led, in collaboration with IBM, to the Andrew system, one of the first campus-wide computing networks in the country ^[1].

Newell described several major lines of his work as "diversions" from his central goal of understanding the mind, though most produced lasting results of their own ^[1]:

Scroll sideways for more →

The physical symbol system hypothesis

Newell and Simon used their 1975 Turing Award lecture, published in 1976 as "Computer Science as Empirical Inquiry: Symbols and Search," to state the working hypothesis behind their whole research program: "A physical symbol system has the necessary and sufficient means for general intelligent action" ^[6][12]. On this view, intelligence, human or artificial, consists in the manipulation of symbol structures, and computers and minds are two instances of the same kind of system. The companion heuristic search hypothesis held that symbol systems solve problems by generating and selectively searching candidate solutions ^[12]. The hypothesis became the philosophical backbone of what is now called symbolic AI ^[6].

It also became one of the most debated claims in the field. Hubert Dreyfus argued that human expertise rests on unconscious, embodied skill rather than symbol manipulation; Rodney Brooks built robots that behaved usefully without symbolic representations, arguing that "the world is its own best model"; and the post-2012 success of deep neural networks was widely read as evidence that sub-symbolic learning outperforms hand-built symbol systems on perception and many other tasks ^[6]. Newell, for his part, insisted the question would be settled by building systems rather than by philosophical argument; Simon wrote that both of them believed the fate of AI "would be determined not by debates with philosophers about what was possible, a priori, but by our success or failure in building programs" that actually simulate thought ^[1].

Soar and unified theories of cognition

From the early 1960s Newell's publications track a single preoccupation: the control structures and architectures that could carry problem-solving programs past their brittleness. He led development of a succession of production-system languages, in which all knowledge takes the form of condition-action rules; the best known, OPS5, supplied the central ideas for the language used to program Soar ^[1].

Soar, developed with his graduate students John Laird and Paul Rosenbloom at Carnegie Mellon starting in 1983, was the culmination. The name originally stood for State, Operator And Result ^[7]. Soar is a production system organized around problem spaces, inheriting the GPS view that all deliberate behavior is search. Two mechanisms defined it: a universal weak method (Laird and Newell, 1983), under which the system declares an impasse whenever its knowledge fails to select a unique next operator and recursively sets up a subgoal in a new problem space to resolve it; and chunking (Rosenbloom and Newell, 1982), which compiles the results of subgoal processing into new productions, a single learning mechanism that reproduced the empirically observed power law of practice ^[1][7].

In 1987 Newell delivered the William James Lectures at Harvard, published in 1990 as Unified Theories of Cognition. The book argued that cognitive psychology already had enough data to support broad theories of the mind's architecture, and offered Soar, carefully labeled "a candidate unified theory" rather than the answer, as a worked example ^[1]. The Soar project outlived him: work continued at Carnegie Mellon, the University of Southern California, and above all the University of Michigan, where Laird's group still maintains the architecture as an open-source system, with releases (version 9.6.5, published in May 2026) adding refinements to chunking, episodic memory, and semantic memory ^[7][8].

Honors and death

Scroll sideways for more →

Newell was a member of the National Academy of Sciences ^[1] and held honorary doctorates from the University of Pennsylvania (1986) and the University of Groningen (1989) ^[2][4]. His doctoral students included computer chess pioneer Hans Berliner, HCI researcher Stuart Card, Soar architect John Laird, and multi-agent systems researcher Milind Tambe ^[4].

Newell died of cancer in Pittsburgh on July 19, 1992, at age 65, still actively working on Soar ^[1]. In December 1991, seven months before his death, he gave a talk at Carnegie Mellon called "Desires and Diversions," reflecting on his research strategy; among the maxims he offered was "Choose a final project to outlast you." Simon's memoir for the National Academy closes on it: "For Allen, Soar was that project" ^[1]. His memory is carried by the ACM/AAAI Allen Newell Award, established in 1993 and given for career contributions that have breadth within computer science or that bridge computer science and other disciplines ^[10], and by Newell-Simon Hall, a Carnegie Mellon computer science building that pairs his name with Simon's ^[11].

See also

• History of artificial intelligence
• Symbolic AI
• Dartmouth conference
• John McCarthy
• Marvin Minsky
• Turing Award

References

1. ^Simon, Herbert A. "Allen Newell 1927-1992." Biographical Memoirs, Vol. 71. National Academy of Sciences, 1997. nasonline.org/...newell-allen.pdf
2. ^Lee, J. A. N. "Allen Newell." Computer Pioneers, IEEE Computer Society. history.computer.org/...newell
3. ^National Science & Technology Medals Foundation. "Allen Newell: National Medal of Science, 1992." nationalmedals.org/...allen-newell
4. ^Wikipedia. "Allen Newell." en.wikipedia.org/...Allen_Newell
5. ^Wikipedia. "Logic Theorist." en.wikipedia.org/...Logic_Theorist
6. ^Wikipedia. "Physical symbol system." en.wikipedia.org/...Physical_symbol_system
7. ^Wikipedia. "Soar (cognitive architecture)." en.wikipedia.org/...Soar_(cognitive_architecture)
8. ^Soar Cognitive Architecture home page. University of Michigan. soar.eecs.umich.edu
9. ^Wikipedia. "Turing Award." en.wikipedia.org/...Turing_Award
10. ^AAAI. "AAAI Awards: Allen Newell Award." aaai.org/aaai-awards
11. ^Wikipedia. "Carnegie Mellon School of Computer Science." en.wikipedia.org/...lon_School_of_Computer_Science
12. ^Newell, Allen, and Herbert A. Simon. "Computer Science as Empirical Inquiry: Symbols and Search." Communications of the ACM 19, no. 3 (1976): 113-126. doi.org/...360018.360022