Book: Human Problem Solving

Overview
Human Problem Solving (Newell & Simon, 1972) presents a systematic, computationally grounded account of how people solve problems. It synthesizes decades of empirical observation and theoretical work to treat problem solving as a process of search through a space of possible states, guided by operators and heuristics. The book connects psychological data, formal models, and computer simulations to show how cognitive processes can be understood and modeled.

Core concepts
Central to the account is the notion of a problem space composed of initial states, goal states, and permissible operators that transform states. Problem solving is framed as navigating that space: selecting and applying operators to reduce the difference between current and goal states. Heuristics, rules of thumb that limit and guide search, are emphasized as essential for making search tractable, since exhaustive search is rarely feasible for realistic tasks.

Methods and evidence
A key methodological contribution is the systematic use of protocol analysis, especially verbal reports of thought processes, combined with controlled task experiments. These data were used to infer internal strategies and to validate computational models. The authors compare human protocols with the step-by-step behavior of programmatic models, using discrepancies to refine both theory and implementation.

Computational models
The book reports and evaluates early cognitive programs, notably the General Problem Solver (GPS) and predecessors such as the Logic Theorist. These systems implement symbolic representations of states and operators along with control structures that emulate human heuristics, like means-ends analysis. The models demonstrate how a small set of procedures and knowledge structures can reproduce many qualitative aspects of human problem-solving behavior, including backtracking, goal decomposition, and the selective use of subgoals.

Learning, expertise, and representation
Newell and Simon emphasize that problem representation critically influences solvability: the way a problem is cast can transform a difficult search into a manageable one. Expertise is portrayed as the accumulation of domain-specific operators and higher-level problem schemas that shape search. Learning mechanisms are discussed primarily in terms of the acquisition and refinement of productions and heuristics that improve search efficiency over practice.

Implications and influence
The work helped establish cognitive psychology's computational paradigm and influenced artificial intelligence research on search and knowledge representation. Its framing of problem solving as heuristic-guided search provided a foundation for later models of production systems, planning, and expertise. It also encouraged experimental techniques that link verbal protocols with formal models.

Limits and continuing relevance
The approach focuses on well-structured, symbolically representable problems and on internal, algorithm-like processes, so it underrepresents affective, social, and embodied dimensions of problem solving. Subsequent research has extended and revised its assumptions, but the core insights, problem spaces, operators, heuristics, and the interplay of representation and search, remain central. The synthesis of empirical data with explicit computational models continues to guide cognitive science and AI studies of human problem solving.