Problem Space
A problem space is the formal representation of a problem as a set of states, operators that transform states, an initial state (where the solver starts), a goal state (where the solver wants to arrive), and path constraints (the conditions that any valid solution must satisfy). Simon and Newell introduced the framework in Human Problem Solving (1972) as the proper unit of analysis for cognitive research: what the solver is solving, not merely what the solver is doing. The problem space for any non-trivial problem is too large to search exhaustively — chess has more positions than atoms in the observable universe, software architecture has more configurations than any mind can enumerate — so the solver must use heuristic search to navigate the space toward the goal. The quality of the navigation depends on the quality of the representation: well-represented problems are solvable by competent heuristics, while poorly represented problems produce energetic but misdirected search. AI has dramatically expanded the problem spaces that individual builders can explore, which makes representation discipline more consequential than ever — the builder who specifies the problem precisely before invoking the tool achieves outcomes the tool cannot produce for a builder who delegates representation to the tool's defaults.
In the AI Story
The problem-space framework was foundational to early AI. The Logic Theorist (1955), General Problem Solver (1957), and every symbolic AI system built through the 1980s used problem-space representations as their computational substrate. The framework shaped both AI research and cognitive psychology for decades, establishing the vocabulary through which structured problem-solving could be analyzed formally.
Simon and Newell's research revealed a counterintuitive pattern about expertise: experts spend more time on representation (understanding what the problem is) and less on search (finding solutions within the represented problem) than novices do. The novice dives into generating solutions immediately; the expert lingers at the representation stage, asking whether the goal is well-specified, whether the constraints are clear, whether the problem space is structured in a way that makes good solutions findable. The expert's investment in representation pays off during search, because a well-represented problem directs heuristics toward promising regions.
The framework has acquired new urgency in the AI age for a specific reason: the tool makes search so fast that the temptation to skip representation becomes nearly irresistible. When implementations take minutes, investing hours in understanding the problem feels like overhead. The result is the pattern Simon and Newell documented in novices — energetic but misdirected search, producing many artifacts without the judgment to assess whether any of them solve the problem that actually matters. The AI-augmented builder is at risk of becoming a permanent novice: expert-level at generation, novice-level at representation, because the discipline of specifying what the problem is has been crowded out by the seductive speed of generating answers.
Origin
Simon and Newell developed the framework through the mid-1950s as they built the Logic Theorist and General Problem Solver. The 1972 Human Problem Solving synthesized a decade of research into the comprehensive theoretical statement that established problem spaces as the dominant analytical framework for structured cognition.
The framework's influence extended well beyond AI. Cognitive psychology adopted it as a standard tool for analyzing reasoning and problem-solving. Operations research used it for formal optimization. Design theory extended it to wicked problems where the representation itself is unstable. The framework's durability reflects both its formal precision and its descriptive accuracy — real problem-solvers do navigate spaces structured along roughly the lines Simon and Newell described.
Key Ideas
Problems are spaces, not puzzles. Every problem can be formally represented as a structured space of states, operators, and constraints.
Representation precedes search. The first cognitive task in problem-solving is specifying what the problem is, not generating solutions to it.
Experts invest in representation. The distinguishing cognitive investment of expert problem-solvers is extended time spent understanding the problem before attempting solutions.
AI tempts representation neglect. Fast search makes representation investment feel costly, producing novice-style problem-solving behavior in builders whose generation capabilities appear expert.
Well-represented problems are solvable. The quality of the problem space — whether the goal is precise, constraints are clear, structure matches reality — determines whether subsequent search can produce good solutions.
Appears in the Orange Pill Cycle
Further reading
- Newell and Simon, Human Problem Solving (1972)
- Newell, Unified Theories of Cognition (1990)
- Anderson, How Can the Human Mind Occur in the Physical Universe? (2007)
- Gary Klein, Sources of Power (1998)