Human Problem Solving
Human Problem Solving, published in 1972 by Herbert Simon and Allen Newell, is the seven-hundred-page empirical masterwork that established cognitive science as a rigorous discipline. The book synthesizes two decades of research into a unified theory: human problem-solvers navigate formally structured problem spaces through heuristic search, using rules of thumb to direct their attention toward promising alternatives and away from unpromising ones. The theory is demonstrated through detailed protocol analyses of subjects solving problems in chess, cryptarithmetic, logic puzzles, and other structured domains — and through the computer programs that Simon and Newell built to replicate the patterns they observed. The book's enduring contribution is the framework through which problem-solving of every kind becomes analyzable: goal definition, problem representation, operator selection, heuristic search, threshold termination. The framework applies with uncomfortable precision to AI-augmented work, where the traditional search component has been radically accelerated while the goal-setting component — which requires integration, judgment, and domain wisdom — remains as bounded as it was when Simon and Newell began their research.
In the AI Story
The book represents the culmination of Simon and Newell's collaboration, which began in 1955 when they built the Logic Theorist — widely considered the first artificial intelligence program. The collaboration produced not only a theory of problem-solving but an entire methodology for studying cognition: the think-aloud protocol analysis, the computer simulation of cognitive processes, and the integration of psychological experiment with formal modeling.
The theory's central distinction — between search heuristics that navigate toward a goal and goal heuristics that specify what the goal is — has proven remarkably durable. Every subsequent framework for understanding expertise, decision-making, and creativity has had to engage with the problem-space formulation the book articulated. The distinction also illuminates what AI changes and what it does not: AI provides extraordinary search heuristics (implementation knowledge, pattern libraries, architectural conventions) while leaving goal heuristics — the capacity to define what is worth pursuing — essentially untouched.
The book's empirical method was novel and controversial when it appeared. Rather than studying aggregate behavior across large samples, Simon and Newell studied individual subjects in intensive detail, recording their verbal protocols as they worked through problems. The method sacrificed statistical power for analytical depth, trading breadth for the ability to trace the actual cognitive operations subjects were performing. The approach has become standard in cognitive science and remains the gold standard for understanding expert performance.
Origin
Simon and Newell began their research in the mid-1950s, driven by two convergent interests: Simon's commitment to studying how bounded minds actually make decisions, and Newell's background in artificial intelligence research. The collaboration produced the Logic Theorist in 1955 and the General Problem Solver (GPS) in 1957, both landmark programs that demonstrated machines could replicate aspects of human reasoning.
The 1972 book represents roughly fifteen years of accumulated research, synthesized into a unified framework. It remains the single most comprehensive statement of the information-processing theory of human cognition, and — despite the subsequent rise of alternative paradigms including neural networks and embodied cognition — continues to provide the most rigorous framework for analyzing structured problem-solving in domains where the problem space can be formally specified.
Key Ideas
Problems are spaces. Every problem can be represented as a formal space with initial states, goal states, operators that transform states, and path constraints.
Search is heuristic. Bounded agents cannot search problem spaces exhaustively; they use rules of thumb to direct the search toward regions where good solutions are most likely to be found.
Expertise is pattern recognition. What distinguishes expert problem-solvers from novices is not deeper search but better initial selection — pattern libraries that identify promising starting points before conscious deliberation begins.
Representation matters more than search. Expert problem-solvers invest more time in understanding the problem and less in generating solutions; novices dive into the search immediately and produce worse outcomes.
Goal-setting is distinct from search. The cognitive operations that specify what problem is worth solving are structurally different from those that solve it — and AI accelerates the second while leaving the first bounded.
Appears in the Orange Pill Cycle
Further reading
- Newell and Simon, Human Problem Solving (1972)
- Newell, Unified Theories of Cognition (1990)
- Chase and Simon, 'Perception in Chess' (1973)
- Ericsson and Simon, Protocol Analysis (1984)