Newell, Allen

(b. 19 March 1927 in San Francisco, California; d. 19 July 1992 in Pittsburgh, Pennsylvania), pioneer of research into artificial intelligence (AI).

Newell was the second child of Robert R. Newell, a radiologist, and Jeanette Levalley, a homemaker. Graduating from Lowell High School in 1945 after a career marked more by his ability in football than by intellectual pursuits, Newell enlisted in the U.S. Navy. Assigned to the chore of making maps of the radiation emitted during the Bikini Atoll nuclear tests, Newell realized that he loved doing scientific work. He accordingly, after his discharge in 1946, enrolled at Stanford University, where he copublished an article on X-ray optics. In December 1947 Newell married his high school sweetheart, Noel McKenna. They had one son. Newell earned a bachelor’s degree in 1949 and began studying mathematics at Princeton University in New Jersey.

At Princeton, Newell’s interests soon turned away from theoretical math and toward experimentation. Consequently, he took a leave of absence from Princeton and went to work for the newly formed Rand Corporation in Santa Monica, California. One of the papers he wrote during this period, “Observations on the Science of Supply” (Technical Report D-926, Rand Corporation, 1951), focuses directly on the weakness of theoretical models to comprehend the realities of production and distribution systems. While working over the next few years on a Rand study of the U.S. Air Force early warning system, Newell and a colleague, Cliff Shaw, solved a basic problem: how to use a computer to simulate radar images on an early warning screen. Their achievement represented a major step not only toward Newell’s ultimate work in artificial intelligence (AI) but also toward the development of modern video games.

In September 1954 Newell experienced what he later described as a “conversion experience.” At a Rand seminar he learned that computer systems could not only create images but recognize them as well. His reaction was embodied in a classic paper, “The Chess Machine: An Example of Dealing with a Complex Task by Adaptation” (Proceedings of the Western Joint Computer Conference, 1955), which embodied nearly all the key features of what became known as an expert system: aspiration levels, “good enough” moves, generation of subgoals, and a coherent symbol system with which to represent the playing surface. Newell moved to the Carnegie Institute of Technology (now Carnegie-Mellon University) in Pittsburgh, where he developed for Rand the logic theory machine in 1956. He worked with the Nobel Prize winner Herbert Simon toward a Ph.D., which Newell received in 1957. Perhaps the culmination of this fertile period was his invention of Information Processing Languages (IPL), the first list-processing languages. These languages introduced lists, associations, frames, streams, and associative retrieval, without which the achievements in AI of the next decade might have been impossible. His 1961 manual for the IPL-V was the first to advocate “top-down” programming. Another prime goal of AI research had been reached: a language that “relies heavily on heuristic methods similar to those that have been found in human problem solving activity.”

Newell continued his dual role as Carnegie researcher and Rand employee until 1961, when he became a full-time professor in charge of developing Carnegie-Mellon’s School of Computer Science. His own doctoral degree had been awarded in industrial administration because no appropriate title for his work existed at that time. Working closely with Shaw and Simon, he produced the effective chess-playing program he had theorized earlier. Although it was no IBM Deep Blue—a ten-year-old once beat it— it provided the experimental validation that Newell always demanded from theoretical Schemas. In 1972 he and Simon finally put all their accumulated insights into a book, Human Problem Solving, which remained a standard in its field into the twenty-first century.

Newell believed that the physical architecture of the computer was not the key to machine intelligence, although he acknowledged that the development of massively parallel computing systems made the kind of processing he was designing far simpler. He and Gordon Bell coauthored a book on such systems, Computer Structures: Readings and Examples (1971). Again pushing theoretical insights toward their practical applications, Newell collaborated on L*, a new language based on the insights of his book, and a new type of menu system called ZOG, which ultimately was used on the aircraft carrier Carl Vinson in 1982.

During this period Newell also tackled a Xerox PARC project analyzing cognitive skills, the heart of which was embodied in the acronym GOMS: goals, operators, methods, and selection. His exposition of this project in The Psychology of Human-Computer Interaction (1983), which he wrote with Stuart K. Card and Thomas P. Moran, was a key step in the attainment of Newell’s ultimate goal, a broad-based theory of cognition. His first real breakthrough in this direction was Soar, a cognitive architecture capable of solving problems and learning in a human way. Other models of cognition focus on verbal memory, sense perceptions, or the creation of concepts, but Soar focuses on problem solving. As Simon said, “The Soar program is a production system.”

After Newell’s death the Soar project continued at Carnegie-Mellon University, the University of Southern California, the University of Michigan, and other centers of research on artificial intelligence. Soar works effectively because of two innovations, “chunking” and the use of a “weak” method of problem solving. With its memory defined as a set of relatively discrete chunks rather than as an unbreakable whole, the system can adapt to new information or concepts without rebuilding the entire memory. By relying on a problem-solving formula that is relatively weak, the system becomes less “stubborn,” turning away from time-wasting problems and trying to resolve them a different way. Invited to deliver the William James Lectures at Harvard University in 1987, Newell sketched out the advantages of the Soar system. He also used Soar as the model in his final book, Unified Theories of Cognition (1990).

When the American Association for Artificial Intelligence was founded in 1979, Newell was elected its first president, and he also headed the Cognitive Science Society. A month before his death he received the National Medal of Science. In 1975 he and Simon received the A. M. Turing Award from the Association for Computing Machinery, which established an annual award in Newell’s honor after his death. Newell’s other honors included awards from the American Federation of Information Processing Societies (1971), the American Psychological Association (1985), the Institute of Electrical and Electronics Engineers (1990), the Franklin Institute (1992), and honorary degrees from the University of Pennsylvania and Groningen University. Newell died of cancer in 1992.

With his muscular build and six-foot, one-inch frame, Newell stood out among computer scientists, who are often more dedicated to intellectual, rather than physical, development. He was renowned for his work ethic, often working through the night to solve a problem. As he said in 1991, “There is no substitute for working hard—very hard.” One of his few leisure pursuits was watching football games on television on the weekends, a reminder of his high school avocation. A man of prodigious energy and endurance, focused on the impalpable workings of the brain, Allen Newell was a rare combination.

For more information on Newell see Pamela McCorduck, Machines Who Think (1979), and J. Laird and P. Rosenbloom, “In Pursuit of Mind: The Research of Allen Newell,” AI Magazine 13, no. 4 (1992): 17-45. An obituary is in the New York Times (20 July 1992).

Hartley S. Spatt