Viewing page 239 of 313

This transcription has been completed. Contact us with corrections.

Do Machines Operate The Way People Do?

It IS no longer debatable whether computers can simulate Human thinking, Professor Herbert A. Simon, Associate Dean of the Graduate School of Industrial Administration at the Carnegie Institute of Technology, contended in the third of the M.I.T. Centennial Series of lectures on computers and management in the future.
  “A dozen or more computer programs have been written and tested that perform some of the interesting symbol-manipulating, problem-solving tasks that humans can perform and do so in a manner which simulates, at least some general respects, the way in which humans do these tasks.” he said. “Computer programs now play chess and checkers, find proofs for theorems in geometry and logic, compose music, design electric motors and generators, memorize nonsense syllables, form concepts, learn to read . . .
  “Computer programs can be written that use non-numerical symbol-manipulating processes to perform tasks which, in humans, require thinking and learning. These programs can be regarded as theories, in a completely literal sense, of the corresponding human processes. These theories are testable in a number of ways: among them, by comparing the symbolic behavior of a computer so programmed with a symbolic behavior of a human subject when both are performing the same problem-solving or thinking tasks.”
  Professor Simon, who is now on leave of absence with the RAND Corporation, is now noted both for work in the social sciences and with computers. He was formally on the faculty of the Illinois Institute of Technology, and has been the author or co-author of many books, including Models of Man: Organizations (1957) and The New Science of Management Decision (1960).
  Professor Simon’s lecture dealt particularly with a program which is called the General Problem Solver because it can employ a system of methods, believed to be those commonly used by college students, to
[[column 2]] 
work on any problem that can be put in a specified but fairly general form. This system, he argued, is basically similar to the way a child learns to speak, the way a mathematician solves problems, and the way a businessman makes decisions.
  “The child acquires perceptual auditory symbols corresponding to words he has heard and associated with visual symbols,” Professor Simon explained. “He tries, on a trial-and-error basis, to produce words, hears his productions, and compares these auditory symbols with those already stored. When he detects differences, he varies the motor symbols to try to remove them. As he learns, he detects that changes in certain components of the motor symbols alter only certain components of the auditory symbols. Thus he is able to factor the correction process and thereby accelerate it greatly.”
 
[[image]]
A discussion of “What Computers Can Do Better” (on May 22) will conclude the Centennial Lecture series reported on these pages. Shown here with a machine’s arm is Professor Claude E. Shannon, ‘40.

[[column 3]] 
  This, in effect, he went on, is the way that living organisms survive. They associate appropriate perceptual symbols and motor symbols. The former describes the world as it is and as it is desired, and the latter describe possible actions. Each set of symbols is a kind of internal language, and how hard a problem is depends on the simplicity or complexity of the rules that define the correspondence between the symbols that constitute these languages.
  As an example of a relatively simple correspondence between symbols, Professor Simon cited the relation between the decimal and octal representations of integers. There is a simple and direct algorithm that solves all problems in translating decimal representations into octal representations. But in other cases the correspondence between the vocabularies of two different languages maybe purely conventional or arbitrary. Rote learning is the only means of building up a translation dictionary then, and immense amounts of trial-and-error searching may be required.
  “The aspects of the environment with which we, as organisms, deal effectively reach neither of these two extremes,” Professor Simon said. “The translation between the ‘state’ language that describes our perceptions of the world and the ‘process’ language that describes our actions on the world is reducible to no simple rule, but it is not arbitrary. Most of our skill in dealing with the environment is embodied in elaborate heuristics, or rules of thumb, that allow us to factor (approximately) the complex perceived world into highly simple components and to find (approximately and reasonably reliably) the correspondences that allow us to act on the world predictably.”
  The General Problem Solver program, he concluded, describes the core of these heuristics. Hence, in such a program, he said, we now have a first approximation to what Walter Pitts of M.I.T. has described as “the hierarchy of final causes traditionally called the mind.”
  Professor Sidney S. Alexander of the School of Industrial Management presided on the occasion of this lecture, and the commentators were Professor George A. Miller of Harvard and Marvin L. Minsky of M.I.T.
[[bottom]]
20                  THE TECHNOLOGY REVIEW