Herbert Simon

Introduction

Herbert Alexander Simon (1916–2001) was one of the few people to win both a Nobel Prize and a Turing Award, and the reason he could is the same in both cases: he refused to believe that the human mind optimises.

Classical economics modelled decision-makers as agents who consider all options, weight all consequences, and choose the best. Simon observed actual executives, read actual organisational records, and concluded that nobody does this and nobody could. The mind has limited computational capacity and the world supplies incomplete information, so real decisions are made by searching until an option is found that is good enough and then stopping. He called the first condition bounded rationality and the second satisficing.

The same idea made him a founder of artificial intelligence. If rationality is a matter of procedure rather than of arriving at an optimum, then a procedure can be specified and run on a machine — which is what he and Allen Newell did, building programs that solved logic theorems and proved that they had.

Early Life and Education

Simon was born in Milwaukee, Wisconsin, on 15 June 1916. His father was an electrical engineer who had emigrated from Germany; his mother was a accomplished pianist, and music was a serious part of the household.

He entered the University of Chicago in 1933 intending to study the social sciences, and took his bachelor's degree in 1936. His interest was in how organisations actually make decisions, which at the time was not a subject — political science was concerned with formal structures and legal authority, not with what happened inside a municipal budget office.

He earned his doctorate in political science from Chicago in 1943, with a dissertation that became Administrative Behavior. Before finishing it he had worked at the International City Managers' Association and then ran a research group at the University of California, Berkeley, studying how administrative units actually operate. The empirical grounding came first; the theory was built to fit it.

That ordering — observation before formalism — is the single habit that explains an otherwise implausible career. Simon moved between political science, administration, economics, psychology, computer science and philosophy, and the moves look like restlessness only if one assumes a discipline is defined by its subject matter. On his own account he had one subject, decision-making, and used whatever formalism the instance required. Critics have sometimes read the same record as evidence of breadth purchased at the cost of depth in any one field; his defenders point out that the Nobel and the Turing are in different disciplines and were awarded for work that is continuous with itself.

Academic and Professional Career

In 1949 Simon moved to the Carnegie Institute of Technology in Pittsburgh to help build what became the Graduate School of Industrial Administration, and he remained at that institution — Carnegie Mellon University after 1967 — for the rest of his life.

This was an unusual base for someone who would win a Nobel Prize in economics: he never held an appointment in an economics department. He held professorships in administration, psychology, and computer science, which is an accurate description of what he actually did.

The 1950s were the decisive decade. Working with Allen Newell and J.C. Shaw, he built the Logic Theorist (1956), generally regarded as the first artificial intelligence program, which proved theorems from Whitehead and Russell's Principia Mathematica — and in one case found a shorter proof than the one in the book. The General Problem Solver (1957) followed, a program that attacked problems by means–ends analysis: compare the current state to the goal, find the largest difference, and select an operator that reduces it.

In 1975 he and Newell received the Turing Award. In 1978 he received the Nobel Memorial Prize in Economic Sciences for his research into the decision-making process within economic organisations. He received the National Medal of Science in 1986 and the American Psychological Association's Award for Distinguished Scientific Contributions in 1969.

Major Contributions to Psychology

Bounded rationality and satisficing

The argument, set out in Administrative Behavior (1947) and sharpened across the 1950s, has two parts.

The first is a capacity claim. An agent who optimises must enumerate alternatives, estimate the consequences of each, and compare them on a single scale. For any realistic decision the number of alternatives and consequences is very large, and no organism with finite computational resources can carry out the enumeration. Rationality is therefore bounded by the capacity of the agent and by the structure of the environment.

The second is a procedural claim. Given those bounds, real agents do not optimise; they satisfice. They set an aspiration level, search alternatives in some order, and stop at the first one that meets it. The aspiration level is itself adaptive — it rises when good options are easy to find and falls when they are not.

The important consequence is that satisficing is not a lesser form of optimising. It is a different procedure with different properties, and in environments where searching has costs it is frequently superior to optimising. An agent who stops when it finds something good enough spends the time it saved on other problems.

The ant on the beach

Simon's most economical argument is a simile. An ant walks across a beach, following a laboured, irregular path around pebbles and through patches of sand. The complexity of that path, he observed, is not mostly a property of the ant. It is a property of the beach.

The point is that apparently complex behaviour can be produced by a very simple mechanism interacting with a complicated environment. This is the founding insight of the study of ecological rationality and of the whole programme of explaining behaviour from the interaction of simple mechanisms with structured environments rather than from internal complexity. It also licenses the research strategy that made artificial intelligence possible: you do not need to model the ant in detail to predict its path if you can model the beach.

Expertise and the ten-year rule

With William Chase, Simon studied chess expertise and established that the master's advantage is not general memory or general computing speed but a very large store of perceptual chunks — configurations of pieces recognised as units. They estimated that acquiring such a store takes roughly ten years of concentrated practice. This finding is the origin of the deliberate-practice research programme that K. Anders Ericsson later developed, and of "the ten-year rule" as it entered popular discussion.

The physical symbol system hypothesis

With Newell, Simon proposed that a physical symbol system — anything that manipulates symbols according to rules — has the necessary and sufficient means for general intelligent action. It was the strongest possible statement of the classical AI programme, and its failure to scale is as instructive as its early successes.

Major Publications

Administrative Behavior (1947; 4th ed. 1997) — where bounded rationality begins
"A Behavioral Model of Rational Choice" (1955, Quarterly Journal of Economics)
"Rational Choice and the Structure of the Environment" (1956, Psychological Review)
Organizations (1958, with James March and Harold Guetzkow)
Human Problem Solving (1972, with Allen Newell) — the culmination of the programme
The Sciences of the Artificial (1969; 3rd ed. 1996) — the ant, and the case for design as a science
Models of My Life (1991) — his autobiography

Impact and Legacy

Behavioural economics exists because Simon established that the rational agent is an empirical question and not an axiom. Every model of choice that incorporates search costs, limited attention, or procedural rules descends from his objection. Daniel Kahneman's Nobel citation rests on a programme whose premise — that human judgment departs systematically from the optimising model — Simon had stated thirty years earlier.

In artificial intelligence, means–ends analysis, heuristic search, and the whole architecture of problem-solving-as-search come directly from Newell and Simon. In organisation theory, Administrative Behavior remains a founding text. In cognitive psychology, the shift from studying association to studying the processing of symbol structures is substantially his doing.

Criticism. The most substantive challenge comes from Gerd Gigerenzer and the ecological rationality programme, and it is a friendly amendment that has teeth. Gigerenzer's objection is that Simon conceded too much: by defining rationality as bounded, Simon abandoned optimisation as the normative standard without supplying a replacement standard by which a heuristic could be judged good or bad. On Gigerenzer's account, heuristics should be evaluated against the structure of the environment they operate in — a fast-and-frugal heuristic can be more accurate than a computationally expensive optimising model when the environment has the right structure. That is a different and stronger claim than "people satisfice."

Satisficing has a technical weakness too: it does not specify where aspiration levels come from or how they adjust, which makes the model difficult to test and difficult to use predictively. It describes the stopping behaviour well; it is much less clear about the search that precedes it or the threshold that terminates it. "Bounded rationality" has also proved so elastic that it is now used to describe almost any departure from the optimising model, which gives it enormous reach in the literature and correspondingly little specific content.

The physical symbol system hypothesis has been the most decisively rejected part of his programme. Connectionist, embodied and situated approaches argued — and largely persuaded the field — that intelligent behaviour need not rest on the manipulation of explicit symbolic representations, and the systems that eventually solved the problems Simon set out to solve (continuous perception, language, games of imperfect information) did so without symbol structures of the kind he and Newell specified. The hypothesis was stated as a necessary and sufficient condition; the sufficiency claim is what failed, and it failed for fifty years before the field found a different route.

The ten-year rule has had a similar career. Ericsson's later deliberate-practice programme both depends on and qualifies it: the estimate was a rough figure from a small number of chess players, and subsequent meta-analytic work finds that accumulated practice explains far less of the variance in expert performance than the popular version implies — and less still at the highest levels. The chunks are real; the timetable is not a law.

Simon was also wrong on a bet worth recording. In 1957 he predicted that a computer would be world chess champion within ten years. Deep Blue beat Kasparov in 1997 — forty years late. The underlying claim, that machines would master tasks requiring what looked like judgment, has been vindicated; the confidence about timescales was characteristic of the field in that decade and has not been one of its better-calibrated features.

Learn More

Daniel Kahneman — whose work on judgment made Simon's premise into an empirical research programme
Amos Tversky — who established the systematic departures from rationality that satisficing predicts
K. Anders Ericsson — who turned Simon and Chase's chess finding into the deliberate-practice programme, and then into a dispute
Thinking and Reasoning — the cognitive processes Simon modelled as search
History of Psychology — the cognitive revolution he helped start

References

Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2), 129-138.