Means-end analysis is one of the most general and powerful heuristic strategies for problem solving, identified by Newell and Simon as central to human problem-solving behavior. The strategy works by: (1) comparing the current state to the goal state and identifying the most important difference, (2) finding an operator (action) that reduces this difference, and (3) if the operator cannot be applied directly, setting up a subgoal to enable its application. This recursive process of setting and achieving subgoals can solve complex problems by breaking them into manageable steps.
Key Structures
- Dorsolateral prefrontal cortex — A lateral prefrontal region critical for working memory, cognitive control, planning, and abstract reasoning.
- Anterior cingulate cortex — A medial frontal region involved in conflict monitoring, error detection, and the allocation of cognitive control.
- Parietal cortex — The cortical region between frontal and occipital lobes, integrating sensory information for spatial representation and attention.
- Problem Solving — The cognitive processes involved in finding solutions to novel, non-routine challenges — from well-defined puzzles to ill-defined real-world problems.
Key Functions
Solve problems by identifying the difference between current state and goal state, then selecting operators to reduce that difference, recursively creating subgoals.
The Algorithm
Means-end analysis follows a recursive procedure that can be stated precisely. Given a current state and a goal state: (1) if the current state matches the goal, the problem is solved; (2) otherwise, compute the difference between the current state and the goal state; (3) select an operator relevant to reducing the largest or most important difference; (4) if the operator's preconditions are met, apply it; (5) if its preconditions are not met, set a subgoal to satisfy those preconditions and recursively apply means-end analysis to this subproblem. This recursive structure allows the strategy to decompose complex problems into hierarchies of simpler subproblems.
2. Select operator O that reduces D
3. If preconditions of O are met → apply O
4. If preconditions are not met → set subgoal to satisfy preconditions
5. Recursively apply means-end analysis to subgoal
Repeat until current state = goal state
The General Problem Solver
Newell and Simon implemented means-end analysis in their General Problem Solver (Newell, Shaw, & Simon, 1959) — one of the earliest artificial intelligence programs. GPS could solve logical proofs, play chess endgames, and solve the Tower of Hanoi puzzle using the same general strategy. While GPS was too weak to handle complex real-world problems (which require extensive domain-specific knowledge), it demonstrated that a single problem-solving strategy could apply across diverse domains.
The Tower of Hanoi
The Tower of Hanoi is the paradigmatic means-end analysis problem. The goal is to move a stack of disks from one peg to another, moving only one disk at a time and never placing a larger disk on a smaller one. Means-end analysis solves it naturally: the largest difference between the current and goal state is the position of the largest disk. To move it, the smaller disks must first be cleared (a subgoal), which itself requires recursive application of the same strategy. The optimal solution for n disks requires 2^n - 1 moves.
Human performance on the Tower of Hanoi reveals several important aspects of problem solving. People have difficulty with "backward" moves — moves that temporarily increase the distance from the goal — because means-end analysis creates a bias toward immediate progress. Problems requiring more subgoal levels are harder, and errors increase when working memory is loaded, confirming that means-end analysis places heavy demands on the capacity to maintain and track subgoal hierarchies.
Means-end analysis is often confused with hill climbing, a simpler strategy that always selects the action producing the state most similar to the goal. Hill climbing cannot handle subgoals or detours — it gets stuck in local maxima. Means-end analysis is more powerful because it can recognize when an operator's preconditions are unmet and create subgoals to address them, enabling temporary movement away from the goal. However, human problem solvers frequently default to hill climbing when working memory is stressed, producing the characteristic errors seen in Tower of Hanoi and similar problems.
Working Memory Demands
Means-end analysis places substantial demands on working memory. The solver must simultaneously maintain a representation of the current state, the goal state, the differences between them, the operator being considered, any subgoals that have been set, and the relationships among all of these. Each level of subgoal recursion adds to the working memory load. Research by Kotovsky et al. (1985) showed that isomorphic problems (problems with the same logical structure but different surface features) vary enormously in difficulty depending on how much working memory their particular representation demands.
Comparison with Other Strategies
Means-end analysis is one of several general problem-solving heuristics identified in the problem-solving literature. Working backward starts from the goal and identifies what state would immediately precede it, then works recursively backward to the starting state — useful when the goal is well-defined but the starting conditions are complex. Analogical transfer involves recognizing structural similarities between a current problem and a previously solved problem. Generate-and-test involves producing candidate solutions and evaluating them. In practice, expert problem solvers flexibly combine these strategies, selecting the most appropriate approach based on the problem's structure, their domain knowledge, and the demands on working memory.
Real-World Applications
Means-end analysis is pervasive in everyday problem solving, from planning a trip (goal: arrive at destination; difference: not at destination; operator: book transportation; subgoal: have funds for ticket) to debugging software (goal: program produces correct output; difference: produces incorrect output; operator: fix the bug; subgoal: locate the bug). In artificial intelligence, means-end analysis influenced the development of STRIPS-style planning, GPS-based tutoring systems, and modern automated planning algorithms. In education, explicitly teaching students to identify the difference between their current understanding and the learning goal, and then selecting study strategies targeted at reducing that specific gap, applies means-end analysis to self-regulated learning.
Disorders
- Executive dysfunction in frontal lobe damage
- Impaired planning in schizophrenia