Story of Psychology - Morton Hunt [337]
In 1948 the idea that the computer might in some ways function like a mind—at the time this seemed more like science fiction than science—was first broached by von Neumann and the neurophysiologist Warren McCulloch at a California Institute of Technology conference, “Cerebral Mechanisms in Behavior.”
That notion captivated Herbert Simon, then a young professor of political science at the Carnegie Institute (now Carnegie-Mellon University).5“Professor of political science” hardly describes him, however. Simon, the son of an electrical engineer, was so bright that he was skipped in school and was considerably younger than his friends and classmates. Add to that his being unathletic and growing up in Wisconsin keenly aware of his Jewishness, and it is not surprising that he solaced himself by becoming an exceptional student. In college he liked to think of himself as an intellectual, but in fact his interests were freakishly wide-ranging; although he became a political scientist, he was interested and self-taught in mathematics, economics (for which he was awarded a Nobel Prize in 1978), administration, logic, psychology, and computer science.
In 1954, Simon and a brilliant young graduate student of his, Allen Newell, discovered that they shared passionate interests in computers and thinking (both men later earned degrees in psychology), and in creating a computer program that would think. For a first attempt, they chose a very limited kind of thinking, namely, proving theorems in formal logic, an entirely symbolic and almost algebraic process. Simon’s task was to work out proofs of theorems while “dissecting as minutely as possible, not only the proof steps, but the cues that led me to each one.” Then the two men together tried to incorporate this information in a flow diagram that they could turn into a computer program.
After a year and a half of work, Simon and Newell electrified the audience at a 1956 symposium on information theory at MIT with a description of their intellectual offspring, Logic Theorist. Running on JOHNNIAC, a gigantic, primitive, vacuum-tube computer, it was able to prove a number of theorems in formal logic in anywhere from under a minute to fifteen minutes per proof.6 (On a modern computer it would do the same thing in virtually the blink of an eye.) Logic Theorist, the first artificial intelligence program, wasn’t very intelligent; it could prove only logic theorems—at about the same speed as an average college student—and only if they were presented in algebra-like symbols. Still, as the first computer program that did something like thinking, it was a breathtaking achievement. (George Miller was at the presentation; he regards that day as the birthday of cognitive science, even though it took him another four years to declare his apostasy from behaviorism.7)
By the end of the following year, 1957, Newell, Simon, and a colleague, Clifford Shaw, had created a much cleverer program, General Problem Solver (GPS), which incorporated a number of broad principles common to many intellectual tasks, including proving theorems in geometry, solving cryptarithmetic problems, and playing chess. GPS would make a first move or probe to begin determining the “problem space” (the area containing all possible moves between its initial state and the desired goal), look at the result to see whether the move had brought it closer to the goal, concoct possible next moves and test them to see which one would advance it toward the goal, back up to the last decision point if the train of reasoning veered off course, and start again in another direction. A simple problem that GPS solved easily early in its career went as follows (the problem was presented not in these words, which GPS could not understand, but in mathematical symbols):
A heavy father and two young sons have to cross a swift river in a deep wood. They find an abandoned