Problem Solving

Problem solving is the cognitive process of transforming an initial state into a goal state through a sequence of operations that reduce the difference between them. It is not merely a psychological topic — it is the fundamental pattern of all adaptive behavior, from bacterial chemotaxis to chess grandmaster play to scientific discovery. What distinguishes problem solving from mere trial-and-error is the use of heuristic structure: the agent does not search the full space of possible actions but prunes it using constraints, analogies, and subgoals that make the search tractable.

The formal study of problem solving was launched by Herbert Simon and Allen Newell in the 1950s, who showed that human problem solving could be modeled as search through a "problem space" defined by an initial state, operators, and a goal test. Their work connected psychology to artificial intelligence and to operations research, revealing that the logic of problem solving is domain-independent even though the specific heuristics are domain-specific. The insight carries through to Lindley Darden's work on mechanistic discovery: finding a mechanism is a problem-solving process whose constraints come from biological knowledge rather than from chess rules, but whose logic is the same.

The Boundaries of Problem Solving

The Simon-Newell framework is powerful but bounded. It assumes three conditions that are not universally satisfied: that the initial state can be represented, that the goal state can be defined, and that the operators can be enumerated. When any of these conditions fails, problem solving as formal search becomes impossible — and yet adaptation continues.

Consider the wicked problem: a situation in which the problem definition, the solution criteria, and the stakeholder landscape all change as one attempts to solve it. Climate policy, urban planning, and organizational redesign are wicked problems not because they are hard but because they are structurally ill-defined. No amount of heuristic search can transform an undefined initial state into an undefined goal state. What is required instead is not problem solving but problem structuring — a iterative process of frame negotiation, stakeholder alignment, and goal emergence that has more in common with hermeneutics than with operations research.

The boundaries become clearer when we compare problem solving to the other architectures of adaptation described in adaptive systems. Reactive adaptation — chemotaxis, gradient descent, homeostatic regulation — requires no representation of goal states. Recognition-based adaptation — immune response, pattern classification, perceptual learning — requires no enumeration of operators. These are not degenerate cases of problem solving; they are qualitatively different architectures whose dynamics are governed by attractor landscapes rather than search trees. To treat them as "problem solving" is to commit the same error that early AI committed when it assumed all intelligence was symbolic reasoning: it privileges one cognitive architecture and treats it as the universal template.

This does not diminish the importance of problem solving. It locates it. Problem solving is the architecture of adaptation that emerges when representation, goals, and operators are all available — a rare but consequential configuration that produces science, engineering, and strategic planning. It is not the fundamental pattern of all adaptive behavior. It is the most powerful pattern, and the most fragile.