In October 1957 the Social Science Research Council appointed a committee to examine the developments taking place in the use of electronic computers to simulate human thinking. Funds for the committee’s work were provided by a grant from the Ford Foundation. The present statement is a report on a summer research training institute that was organized by the committee and conducted in the summer of 1958. 1The authors were codirectors of the summer research training institute, which was held in Santa Monica in quarters generously provided by the RAND Corporation at its offices there. In addition to being host to the institute, the RAND Corporation contributed the time of Fred Gruenberger, Allen Newell, and J. C. Shaw, who constituted a major part of the staff, and made available to participants the JOHNNIAC computer. Mr. Newell is a staff member of the RAND Corporation, and a lecturer in the Graduate School of Industrial Administration, Carnegie Institute of Technology. Mr. Simon, Professor of Administration and Associate Dean of the Graduate School of Industrial Administration, is a member of the Council’s board of directors and of its Committee on Simulation of Cognitive Processes. The other members of the committee are Carl I. Hovland of Yale University, and George A. Miller of Harvard University; Donald G. Marquis serves as staff of the committee.
Digital computers, designed originally to do speedy arithmetic, are in fact extremely flexible devices capable of processing and manipulating all sorts of information, numerical and nonnumerical. Appropriately programmed, they can as readily handle symbols interpretable as English words or sentences as symbols interpretable as numbers.“By ‘simulate’ we mean induce the computer to accept the same inputs and to produce approximately the same symbolic outputs as a human laboratory subject would in the same situation.”
This flexibility (and its implications) only gradually impressed itself on the users of computers. It led, in time, to a succession of efforts to simulate, with computers, a variety of intellective processes—pattern recognition, problem solving, and certain kinds of learning and concept formation. By “simulate” we mean induce the computer to accept the same inputs and to produce approximately the same symbolic outputs as a human laboratory subject would in the same situation.
By 1957 these efforts had been so far successful that one computer, for example, had learned to recognize a simple pattern, another to discover proofs for theorems in logic, and a third to play an amateurish game of chess. More important, the computers were not only performing these tasks successfully, but they were providing evidence, through their symbolic outputs, that they used processes a good deal like those used by humans to do the same tasks.
Clearly here was a novel set of techniques, of unknown potential, for studying complex human intellective processes. The new committee concluded that a wider circle of social scientists should be acquainted with these techniques, in order to evaluate them and, to the extent that they appeared promising, to prepare to disseminate them at some of the leading university centers for graduate training in the social sciences. A research training institute on simulation of cognitive processes was planned for the summer of 1958 as a first step toward accomplishing this objective.
In addition to the codirectors, the staff of the institute included J. C. Shaw and Frederick Tonge of the RAND Corporation, Carl I. Hovland of Yale University, George A. Miller of Harvard University, and Marvin Minsky of Massachusetts Institute of Technology. There were 20 participants—17 psychologists and 3 sociologists—selected from a very much larger group of able applicants than could be handled:
Robert P. Abelson, Yale University
Lawrence T. Alexander, System, Development Corporation
Richard C. Atkinson, University of California, Los Angeles
William F. Battig, University of Virginia
Daniel E. Berlyne, University of California, Berkeley
Joseph D. Birch, University of Michigan
Jack Block, University of California, Berkeley
James S. Coleman, University of Chicago
Ward Edwards, University of Michigan
Bert F. Green, Jr., Lincoln Laboratory, Massachusetts Institute of Technology
Robert L. Hamblin, Washington University
Lyle V. Jones, University of North Carolina
Edmund T. Klemmer, International Business Machines Corporation Research Center
Gilbert K. Krulee, Case Institute of Technology
Nissim Levy, Brown University
William N. McPhee, Columbia University
Irwin Pollack, Bolling Air Force Base
Roger N. Shepard, Harvard University
Gerald H. Shure, System Development Corporation
Donald W. Taylor, Yale University
Institute activities“The group spent part of its time examining closely some of the computer programs for the simulation of complex human processes that were already in existence.”
The institute was planned for a period of three weeks, June 30 to July 18, but this main session was preceded by a preliminary week, attended by about three quarters of the participants, for training in the elements of computer programming. (The presession was conducted by Fred Gruenberger of the RAND Corporation.) During the first half of the main session, the time was divided among three principal types of activity. The group spent part of its time examining closely some of the computer programs for the simulation of complex human processes that were already in existence. These examples included programs for simulating a subject’s behavior in a binary choice (partial reinforcement) experiment, in discovering proofs for theorems in symbolic logic, in selecting a chess move, and in elementary discrimination and memorizing tasks.
As a second major activity, the group studied the computer programming techniques developed by a RAND-Carnegie Institute of Technology research project for handling these kinds of nonnumerical programming problems. Members of the group learned to program in a so-called “interpretive code,” Information Processing Language IV (IPL-IV), that had been used to write the theorem-proving and chess programs.
A third activity was a survey of research in a number of centers that either employed computer simulation or afforded a promising prospect for such simulation. These sessions included discussions of the possibilities of simulation for studying concept formation, language translation, and learning—including self-organization of random nerve nets.
Out of these activities there grew a strong consensus in the group that computer simulation techniques do indeed have a promising future in psychology—both in the area of higher mental processes, where most of the work has been done to date, and in areas of greater traditional psychological activity, like concept formation, perception, and memorizing. This consensus was exhibited by what the members of the institute said, and even more by what they did. By the end of the first week, there was a virtually unanimous determination among the participants that before the institute ended they would learn how to write programs in Information Processing Language IV, and that they would actually produce programs having significant psychological content and run them on JOHNNIAC, the RAND computer. The members of the institute staff, although they were not quite willing to admit it aloud even to each other, secretly regarded these aspirations as somewhat unrealistic. Their experience with graduate instruction in this kind of computer programming suggested that a semester was a more reasonable learning period than three weeks. However, their estimates did not take account of the quantity and quality of effort the participants were prepared to pour into the task.
About the middle of the second week, the participants divided into some half-dozen working parties, each of which selected a programming task. Few formal sessions were scheduled thereafter for the whole group. Instead, members divided their time between working on practice exercises to develop their programming skills and working with their subgroups to formulate and carry through an appropriate larger programming task. The institute’s workrooms were open around the clock, and participants were to be found there well into the night. If we recollect correctly, about one third of them were still on hand at 1:00 a.m. one night of the final week when time was getting very short and most of the programs were almost, but not quite, ready to be tried out on JOHNNIAC. Only on the final two days was the group willing to take any considerable time out for general sessions—to report to each other on the status of their projects—and then only after we had arranged courier service to the computer room so that valuable “debugging” time would not be lost.
Not every working group reached its objective of running a program successfully on JOHNNIAC. The first team to do so, Ward Edwards and Edmund Klemmer, simulated a subject’s predictive behavior in a binary choice experiment. Another program, written by Lyle Jones and George Miller, successfully synthesized a grammatical English sentence, basing its procedures on Chomsky’s syntactical theories. Several other groups reached the checking-out stage before the institute ended, and went home with the promise that they could complete their debugging by correspondence. Programs were written by Robert Abelson, James Coleman, and William McPhee to simulate one aspect or another of the phenomena of cognitive dissonance. One of these programs was named the “Belief Dilemma Resolver.” A linguistic program related to that of Jones and Miller was developed by William Battig, Daniel Berlyne, and Irwin Pollack.
Three participants focused on problem-solving behavior. Jack Block investigated the possibilities of simulating behavior in the 9-dot problem situation. Robert Hamblin and Lawrence Alexander wrote three programs, two of them for simulating learning processes, that are to be incorporated in a larger General Problem Solver developed by Newell, Shaw, and Simon.“The remaining projects were aimed at simulating behavior in classical psychological experiments relating to perception and concept formation.”
The remaining projects were aimed at simulating behavior in classical psychological experiments relating to perception and concept formation. Roger Shepard wrote a program, which he dubbed “Epigram,” capable of learning to discriminate stimuli; another program for perceptual discrimination was written by Gilbert Krulee; and a third by a team comprised of Bert Green, Richard Atkinson, and Joseph Birch. In the development of these programs, it became increasingly obvious that the line between “stimulus discriminators” and “concept formers” was a cloudy one; hence, the last-mentioned program had a number of similarities with a program for a concept former that was written by Carl Hovland, Donald Taylor, Gerald Shure, and Nissim Levy.
The teams all modestly refused to claim that their programs incorporated revolutionary new psychological theories, but they were struck by the ease and directness with which known and significant psychological processes could be translated into programs and simulated. There was general agreement that the information processing languages are natural languages for stating psychological theories in precise and testable terms, and that digital computers are obvious devices for tracing and testing the implications of theories so stated. On the basis of the experience in the institute, the amount of effort required of research workers in order to use the programming techniques appears to be relatively modest. Most of the participants thought that, given some help in the form of adequate manuals and perhaps a little assistance from the personnel of their local computer centers, they would be able to continue to use the techniques on their own campuses.
Disseminating simulation techniques
The Council’s committee met several times while the institute was in session, and held one general session with the participants. The committee reached the following conclusions as to next steps:
1. Simulation techniques represent a significant new tool for psychological and other social science research. The next important objective is to make this tool available, as a normal part of graduate training opportunities, at the major graduate training centers in the social sciences. For a university to provide this tool to its graduate students and research faculty, it needs the following: (a) a modern computer facility equipped at least with an IBM 650, a Datatron, or an equivalent or larger digital computer; (b) one or more persons associated with the computer facility who are familiar with the information processing languages; (c) manuals for programming in information processing languages; (d) one or more behavioral scientists who are using these techniques in their own research and who are sufficiently skilled in them to inject them into their graduate teaching.
2. Computing facilities of at least minimum adequacy are now to be found on most campuses with large graduate programs—some 70 campuses in all. Helping to secure such facilities, or computing time for research projects, for other campuses lies outside the capability and province of the committee, although members of the committee may be able to assist in identifying the best sources of information and know-how relating to these matters.
3. Some steps have already been taken to acquaint computer center staffs with the information processing languages. Messrs. Newell and Simon have been active in summer training programs for computer personnel, and have developed channels for getting manuals describing information processing languages into the hands of computer center directors.
4. A manual is now being prepared for using the information processing languages with the IBM 650 and IBM 704. The manual, which is being prepared by the RAND-Carnegie group, will be distributed to computer centers and to behavioral scientists—probably early in 1959.
5. The participants in the 1958 institute will constitute a nucleus of faculty familiar with these techniques on some dozen campuses, and several of the universities that were represented in the institute should be able to develop an adequate program of activities with little or no additional outside help. The committee invites proposals, however, for the support of activities to make these techniques available on more campuses, and on all campuses sooner. For example, it is prepared to provide support for individuals who would like to spend a half-year or year studying computer simulation, at one of the existing centers active in this field, with an intention of returning to their own campuses to develop similar activities there. It is prepared also to entertain proposals for research planning conferences of small groups of social scientists who wish to use simulation techniques in their work.
In a field as novel and challenging as this one, it has not been easy for the committee to plan an elaborate, specific, long-range program of action. We think we see what are some of the important next steps, but we would welcome both general advice and specific proposals for activities and projects to whose support the committee might fruitfully contribute.
Herbert A. Simon (1916–2001) was a renowned economist and political scientist who primarily focused on decision-making and is well-known for his theories of “bounded rationality” and “satisficing.” From 1949 until 2001, he taught at Carnegie Mellon University, as part of various departments, including the Tepper School of Business, the School of Computer Science, and the Dietrich College’s Psychology Department. During his tenure at Carnegie Mellon, Simon played a role in developing numerous fields, including cognitive science and computer science. Simon also played an important role in the SSRC’s history, serving as a member-at-large of the Council’s board of directors (1958–1971) and as board chairman (1961–1965). He also participated in various SSRC committees, including the Committee on Business Enterprise Research (1953–1957) and the Committee on the Simulation of Cognitive Processes (1959–1962). In 1978, Herbert Simon was awarded the Nobel Memorial Prize in Economics.
Allen Newell (1927–1992) is recognized as one of the founders of the field of artificial intelligence (AI) and cognitive science, starting his work with the RAND Corporation (1950–1961). There, he helped develop the Information Processing Language (IPL) (1956), as well as two of the earliest AI programs, the Logic Theory Machine (1956) and the General Problem Solver (1957, alongside Hebert A. Simon). He would go on to teach at Carnegie Mellon University. The Association of Computing Machinery would award Newell and Herbert Simon the A.M. Turing Award in 1975 for their accomplishments in the field of AI.
This essay originally appeared in Items Vol. 12, No. 4 in December of 1958. Visit our archives to view the original as it first appeared in the print editions of Items.