Herbert Simon
Herbert A. Simon (1916–2001) was an American social scientist who worked across political science, economics, psychology and computer science, and who is one of very few people to win both the Turing Award (1975, shared with Allen Newell) and the Nobel Memorial Prize in Economic Sciences (1978). Based for most of his career at Carnegie Mellon University, he started from a question about how real organizations decide things and ended up helping to found artificial intelligence as a way of building and testing theories of human thought.
Simon's economic reputation rests on bounded rationality, the claim that real decision-makers have limited information, limited time and limited computing power, and so cannot optimize in the way classical economic theory assumed. His first major book, Administrative Behavior (1947), grew out of his Chicago doctoral work on organizations and argued that people in institutions follow procedures and attention rules rather than calculate the best conceivable option. In Models of Man (1957) and the 1956 paper Rational choice and the structure of the environment he introduced satisficing, a term that blends satisfy and suffice: an agent sets an aspiration level and takes the first option that meets it, rather than searching for the maximum. The Nobel committee cited his research into decision-making processes within economic organizations.
The step from organizations to machines was direct: if human decisions are limited search under constraints, a program can model them as search guided by rules of thumb. At the Carnegie Institute of Technology's new Graduate School of Industrial Administration, Simon collaborated from the mid-1950s with Allen Newell and the RAND programmer Cliff Shaw. Together they produced the Logic Theorist (1955–56) and the General Problem Solver (late 1950s), and attended the 1956 Dartmouth workshop where the field took its name. Their book Human Problem Solving (1972) compared programs' traces with people's think-aloud protocols, treating the program as a theory of the person. Simon's view was that this made psychology testable by simulation, and he was as much a psychologist in these years as a computer scientist.
Simon was also a public forecaster, and his predictions are the usual source of his reputation for over-optimism. In 1957 he predicted that within roughly ten years a computer would be world chess champion and would discover and prove an important new mathematical theorem, and in 1965 he suggested machines would within twenty years be able to do any work a person can do. Chess took until Deep Blue's 1997 match against Garry Kasparov, and the broader prediction remains unmet. These dates are best read as an honest attempt to put numbers on a research program rather than as measured estimates, and they show the period's confidence in heuristic search more than any ignorance on his part.
A frequent misunderstanding is to treat bounded rationality as a claim that people are irrational. Simon's argument was the reverse: behavior is reasonable given the environment and the limits on computation, so the right model describes the procedures people use. This differs from the later behavioral-economics tradition of Kahneman and Tversky, which catalogs systematic deviations from rational choice, although the two are often lumped together. Satisficing is likewise not laziness; in a large search space it can be the best policy available once the cost of searching is counted.
The modern relevance is wide. Heuristic search under a compute budget is the practical core of game-playing and planning systems, and anytime algorithms and early stopping are satisficing in engineering form. In machine learning, resource-rational analysis and the study of bounded agents explicitly continue his line, and his insistence on testing theories of mind by building working programs foreshadowed modern cognitive modeling. His ideas about organization and hierarchy in The Sciences of the Artificial (1969) also anticipate how complex software systems are decomposed into nearly independent modules.
Read Herbert Simon as the scholar who treated the limits of the decision-maker as the thing to be modeled: bounded rationality in economics and heuristic search in AI are one idea, and a person, a firm or a program does well by choosing good enough answers within the resources it has.