Bounded Rationality

Abstract

Bounded rationality is Herbert Simon's account of choice by agents whose knowledge, time, and computational capacity are finite. Against the ideal of a global optimizer that maximizes expected utility, Simon proposed that real decision makers satisfice: they search until an option meets an aspiration level and then stop, treating a good-enough result as the rational product of a costly search. Rationality becomes procedural, a property of the process a mind can execute, rather than substantive, a property of the outcome alone. Two research programs inherited the frame: the heuristics-and-biases program of Kahneman and Tversky read the shortcuts of a bounded mind as systematic error, while the fast-and-frugal program of Gigerenzer and Todd read the same shortcuts as ecologically rational. This article develops satisficing, the cognition-environment scissors, the two programs, ecological rationality and the less-is-more effect, with three interactive demonstrations.

Keywords: bounded rationality, satisficing, ecological rationality, heuristics

Bounded rationality is the principle that the rationality of real agents is limited by the difficulty of the decision problem, the finite computational resources of the mind, and the time available to decide (Simon, 1955). It was introduced by Herbert Simon as a deliberate correction to the model of global rationality inherited from economics and statistical decision theory, in which an agent holds a complete and consistent preference ordering, knows the full set of alternatives and their consequences, and selects the option that maximizes expected utility (Simon, 1955; Simon, 1979). Simon's objection was not that people are irrational but that this model describes no achievable procedure: the alternatives are not given but must be discovered, the consequences are uncertain and expensive to compute, and no organism has the memory or the time the maximization presumes. In their place he offered an agent that satisfices, searching through alternatives until one crosses a threshold of acceptability called the aspiration level, and then choosing it (Simon, 1956). The proposal recast the central question of decision research from what an omniscient agent should do to what a resource-limited agent can do, and it seeded two of the most productive and opposed programs in the modern science of judgment: the heuristics-and-biases tradition, which treats the shortcuts of a bounded mind as sources of predictable bias (Tversky & Kahneman, 1974), and the fast-and-frugal tradition, which treats them as adaptive tools fitted to the structure of the world (Gigerenzer & Goldstein, 1996). The sections below present satisficing and the aspiration level, the interaction of mind and environment Simon called a scissors, the two programs and their quarrel, ecological rationality and the counterintuitive less-is-more effect, and the frugal cue search of the take-the-best heuristic.

What Bounded Rationality Is

Bounded rationality is best understood by contrast with the model it was built to replace. In the classical picture, an agent facing a choice is equipped with a well-defined set of alternatives, a probability distribution over the consequences of each, and a utility function over those consequences, and rationality consists in selecting the alternative with the highest expected utility (Simon, 1955). Simon accepted the elegance of this construction and rejected its psychological realism on three counts. The alternatives are rarely given: a chess player, a firm choosing a strategy, or a person choosing a career must generate options, an act of search that the classical model takes for granted. The consequences are uncertain and costly to evaluate: even where the rules are fully known, as in chess, the space of continuations is far too large to enumerate, so the outcomes attached to each move must be estimated rather than computed (Chase & Simon, 1973). And the agent's own capacities are finite: attention is scarce, memory is limited, and computation takes time the decision cannot always spare (Simon, 1990). Simon drew from this a distinction that organizes the whole field. Substantive rationality judges a choice by its outcome, asking whether the agent in fact selected the best available option; procedural rationality judges it by the process, asking whether the agent used a sensible method given the information and resources at hand (Simon, 1979). Classical economics is a theory of substantive rationality; psychology, Simon argued, must be a theory of procedural rationality, because a mind is a procedure and the outcome is not under its direct control. This is why bounded rationality is not a theory of human failure. An agent who cannot compute the optimum has not failed to be rational by choosing a workable method to get a good result; it has exercised the only rationality available to a finite being, and the interesting scientific questions concern which procedures it uses and how well those procedures perform (Simon, 1990).

Satisficing and the Aspiration Level

Satisficing is Simon's concrete alternative to maximizing, and it is the mechanism from which the rest of the theory unfolds. To maximize is to examine every alternative and select the best; to satisfice is to specify in advance a level of attainment that would be good enough, called the aspiration level, and to accept the first alternative encountered that meets or exceeds it (Simon, 1956). The difference is not a matter of ambition but of the search procedure and its stopping rule. A person selling a house does not appraise every possible buyer, an impossible task, but sets a reservation price and accepts the first offer above it. A firm does not compute the profit-maximizing strategy but adopts a target and searches for a policy that reaches it. The power of the idea lies in the aspiration level, which converts an intractable optimization over an unbounded set into a tractable search with a clear termination condition, and which is itself adjusted by experience: aspiration rises when good alternatives are easy to find and falls when they are scarce, so that the threshold tracks the richness of the environment (Simon, 1955). Satisficing also dissolves a problem that defeats naive optimization, the cost of search itself. When examining each further alternative consumes time or money, the net value of a choice is the quality of the option accepted minus the cost of the search that found it, and a policy of exhaustive search can be strictly worse than one that stops early. There is an aspiration level that best balances the two, and an agent that stops at a good-enough option can outperform one that holds out for the maximum. The demonstration below makes this concrete: as the aspiration level rises, the accepted option improves but the search grows longer, and beyond a point the added search cost outweighs the better outcome.

The Scissors of Cognition and Environment

Simon's most enduring image is that human rational behavior is shaped by a pair of scissors whose two blades are the computational capabilities of the actor and the structure of the task environment (Simon, 1990). One cannot explain the cut by examining a single blade. A mind's strategies look arbitrary until one sees the environment they are fitted to, and an environment's regularities are inert until one sees the mind that exploits them. Simon developed the environmental blade in a 1956 paper that is easy to overlook beside the better-known work on the agent: he showed that an organism with very simple choice mechanisms, following crude rules for when to search and when to stop, can survive and even thrive provided the environment has the right structure, such as food distributed in patches that make local search pay (Simon, 1956). The lesson is that apparent intelligence can reside in the fit between a simple procedure and a structured world rather than in the complexity of the procedure itself. This is the conceptual hinge on which the later fast-and-frugal program turns, because it licenses a research strategy: to understand a heuristic, do not ask in the abstract whether it is rational, ask in which environments it succeeds and in which it fails (Todd & Gigerenzer, 2007). Figure 1 renders the metaphor. It also frames the disagreement that dominates the modern literature, because the two great programs of judgment research accept the scissors and then attend to different blades: one studies the mind's shortcuts and measures them against a logical standard, the other studies the environments in which those same shortcuts are well adapted.

Two Research Programs: Heuristics and Biases Versus Fast and Frugal

Both major programs of modern judgment research descend from Simon, and their disagreement is a disagreement about which blade of the scissors to study and what standard to judge a heuristic against. The heuristics-and-biases program of Daniel Kahneman and Amos Tversky took the bounded mind as given and asked how its shortcuts deviate from the norms of probability and logic. People judging likelihood by representativeness, or frequency by the ease with which instances come to mind, produce answers that depart systematically from the calculus of probability, and the program catalogued these departures, from base-rate neglect to the conjunction fallacy, as a map of the mind's biases (Tversky & Kahneman, 1974). Kahneman later framed the whole enterprise explicitly as a study of bounded rationality, describing the boundary between intuition and reasoning and the conditions under which fast intuitive judgments go astray (Kahneman, 2003). The same program produced prospect theory, a formal model of choice under risk built from the same evidence that people depart systematically from the classical norm (Kahneman & Tversky, 1979). The fast-and-frugal program of Gerd Gigerenzer and Peter Todd accepted the same starting point and rejected the standard of comparison. To call a heuristic biased because it violates a logical norm, they argued, is to judge the mind against a blade it was never fitted to; the right question is how well a heuristic performs in the environments where it is actually used, which is a question about accuracy in the world rather than coherence with a rule (Gigerenzer & Goldstein, 1996). Measured that way, simple heuristics often do remarkably well, sometimes better than the elaborate statistical models that dominate the normative benchmark (Gigerenzer & Gaissmaier, 2011). The dispute is not merely terminological. It turns on whether the deviations Kahneman and Tversky documented are failures to be corrected or adaptations to be understood, and both readings can be right for different heuristics in different environments, which is why the scissors, not either blade, is the durable idea. The programs also differ on how a heuristic should be modeled: heuristics and biases often leaves the underlying process informal, whereas the fast-and-frugal program insists a heuristic be specified precisely enough to simulate, so its accuracy can be measured rather than asserted (Gigerenzer & Todd, 1999).

Ecological Rationality and the Less-Is-More Effect

Ecological rationality is the fast-and-frugal program's answer to the question of what makes a heuristic good. A strategy is ecologically rational to the degree that it is adapted to the structure of the environment in which it is used, so rationality is a relation between a mind and a world rather than a property of the mind alone (Todd & Gigerenzer, 2007). The most striking evidence for the view is the less-is-more effect, in which having or using less information yields more accurate inferences. Its cleanest form is the recognition heuristic. Asked which of two cities is larger, a person who recognizes one city but not the other can infer that the recognized city is the larger, and because recognition is correlated with size through a chain of environmental mediators, such as newspaper coverage, the inference is often correct (Gigerenzer & Goldstein, 1996). The counterintuitive consequence is that recognizing fewer cities can raise accuracy. Someone who recognizes every city can never use the heuristic, because it fires only when exactly one of the two is recognized; someone who recognizes none must guess. Accuracy is therefore highest at an intermediate level of recognition, where the heuristic applies most often, and it can exceed the accuracy of a person who knows more. The effect is not a curiosity of memory but a demonstration that the value of information is not monotonic when the mind exploits the environment's structure, and it appears in domains from geographic inference to sports and financial prediction (Gigerenzer & Gaissmaier, 2011). The demonstration below charts accuracy against the number of objects recognized, and shows the interior peak: recognition validity fixed, accuracy rises as recognition grows, reaches a maximum, and then falls as complete recognition disables the heuristic.

Take-the-Best and Frugal Cue Search

The recognition heuristic decides on a single cue, recognition; the take-the-best heuristic generalizes the idea to any number of cues while preserving the frugality. Facing a choice between two objects on some criterion, take-the-best considers cues one at a time in order of their validity, the probability that the cue points to the correct answer when it discriminates, and it stops at the first cue that distinguishes the two objects, ignoring all remaining cues (Gigerenzer & Goldstein, 1996). It is a one-reason decision rule: however many cues are available, the choice is made on the single highest-validity cue that happens to discriminate, and no weighting or adding is performed. Compared with a linear model that weights and sums every cue, take-the-best looks hopelessly impoverished, yet across a wide range of real prediction tasks it matches or exceeds the linear model's accuracy while inspecting a fraction of the information (Gigerenzer & Gaissmaier, 2011). The reason is ecological. When cue validities are skewed so that one cue is much better than the rest, the environment is non-compensatory: no combination of weaker cues can overturn the verdict of the strongest, so consulting them cannot improve the decision and can only add noise from cues that are estimated with error. A frugal rule that trusts the best discriminating cue is then not a degraded approximation of the linear model but the ecologically rational strategy, and its robustness on new cases, where a complex model may have overfitted the old ones, is a large part of why it travels well (Gigerenzer & Todd, 1999). Table 1 sets the two programs' readings of bounded rationality side by side, and the demonstration lets the reader run take-the-best against a tallying model across environments that differ in how compensatory they are.

Table 1

Two Readings of Bounded Rationality

Note. Both programs accept Simon's bounded agent; they differ on the blade of the scissors they emphasize and the standard against which a heuristic is judged. The readings are not exhaustive rivals, since a heuristic may be biased against a logical norm and yet accurate in its niche (Gigerenzer & Goldstein, 1996; Kahneman, 2003).

Worked Example

The less-is-more demonstration reduces to an exact calculation, and working it by hand shows why more recognition can lower accuracy. Take 50 objects, say the 50 largest cities of a country, and a person who recognizes n of them. Two randomly drawn cities are compared. Recognition is correlated with size, so when the person recognizes exactly one of the pair, choosing the recognized city is correct with a recognition validity of 0.80; when both are recognized, further knowledge decides the pair with a knowledge validity of 0.60; when neither is recognized, the choice is a guess, correct with probability 0.50. The proportion of correct inferences is the sum of three terms, each a probability of a recognition state multiplied by the accuracy in that state. The chance of recognizing exactly one of the pair is 2 times (n/50) times ((50 minus n)/49); the chance of recognizing both is (n/50) times ((n minus 1)/49); the chance of recognizing neither is ((50 minus n)/50) times ((49 minus n)/49). At n equal to 30, these probabilities are 0.4898, 0.3551, and 0.1551, so accuracy is 0.4898 times 0.80 plus 0.3551 times 0.60 plus 0.1551 times 0.50, which is 0.3918 plus 0.2131 plus 0.0776, that is 0.6824. Now recognize every city, n equal to 50: the person recognizes both members of every pair, the recognition heuristic never fires, and accuracy collapses to the knowledge validity alone, 0.60. Recognizing 30 of the 50 cities therefore yields 0.68, while recognizing all 50 yields 0.60; the person who knows less is right more often. Sweeping n across the whole range locates the maximum at n equal to 30, and confirms that accuracy rises from the guessing value of 0.50 at n equal to 0, peaks at 0.68, and descends to 0.60 at full recognition. The effect requires only that recognition validity exceed knowledge validity, which is exactly the condition under which the single cue of recognition is more diagnostic than the knowledge a person can bring to bear when recognition is uninformative (Gigerenzer & Goldstein, 1996).

Discussion

Bounded rationality reorganized the study of decision making by moving the standard of rationality inside the agent's means. Its first consequence was methodological: once rationality is procedural, the object of study becomes the procedures a mind uses, which turned decision research toward the mechanisms of search, stopping, and aspiration that a finite agent can actually run (Simon, 1972; Simon, 1979). Its second consequence was to make room for two readings of the same fact, the shortcut, that continue to organize the field. Where the heuristics-and-biases program reads a shortcut against the norms of logic and finds systematic bias, the fast-and-frugal program reads it against the structure of the environment and finds ecological rationality, and the persistence of both readings reflects that a heuristic really can be at once incoherent by a logical standard and accurate in its niche (Kahneman, 2003; Gigerenzer & Gaissmaier, 2011). The less-is-more effect is the sharpest expression of the second reading, because it refutes the intuition that more information and more computation can only help; when a mind is fitted to its world, frugality is sometimes not a compromise forced by limitation but a source of accuracy in its own right (Gigerenzer & Goldstein, 1996). The idea's reach extends well beyond psychology. In economics it underwrote the critique of the perfectly rational agent, the case Conlisk's survey assembled for building the costs and limits of deliberation into economic models rather than assuming them away (Conlisk, 1996). What survives across every version is Simon's scissors: behavior is the joint product of a limited mind and a structured environment, and a science that attends to only one blade will misread the cut. That is the standing lesson of bounded rationality, and the reason a proposal first made to fix an economic model became a foundation of cognitive psychology.

Glossary

Aspiration level.

The threshold of acceptability a satisficing agent sets in advance; search stops at the first alternative that meets or exceeds it, and the level is adjusted up or down as good options prove easy or hard to find.

Bounded rationality.

Simon's principle that the rationality of real agents is limited by the difficulty of the problem, the finite computational capacity of the mind, and the time available to decide.

Compensatory environment.

A task in which weaker cues can jointly outweigh a stronger one, so that adding and weighting cues improves accuracy and a one-reason heuristic loses ground.

Cue validity.

The probability that a cue points to the correct answer in the cases where it discriminates between the two options; the quantity by which take-the-best orders its cues.

Ecological rationality.

The degree to which a strategy is adapted to the structure of the environment in which it is used; rationality treated as a relation between a mind and a world rather than a property of the mind alone.

Fast-and-frugal heuristic.

A decision rule that uses little information and simple computation, specified precisely enough to simulate, whose accuracy is measured against real environments rather than logical norms.

Global rationality.

The classical model bounded rationality replaces, in which an agent knows all alternatives and their consequences and selects the one that maximizes expected utility.

Less-is-more effect.

The finding that using or having less information can yield more accurate inferences, as when recognizing fewer objects raises the accuracy of the recognition heuristic.

Non-compensatory environment.

A task in which no combination of weaker cues can overturn the verdict of the strongest, so a one-reason rule loses nothing by ignoring the rest and gains robustness by doing so.

Procedural rationality.

Rationality judged by the process a mind uses given its information and resources, rather than by whether the outcome was objectively optimal.

Recognition heuristic.

The rule that, when one of two objects is recognized and the other is not, infers the recognized object scores higher on the criterion; it fires only when exactly one object is recognized.

Satisficing.

Simon's alternative to maximizing: search alternatives sequentially and accept the first that meets the aspiration level, rather than examine all options to find the best.

Scissors metaphor.

Simon's image that behavior is cut by two blades, the cognitive limits of the agent and the structure of the environment, neither of which explains the result alone.

Substantive rationality.

Rationality judged by the outcome alone, asking whether the agent in fact chose the objectively best option; the standard of classical economics.

Take-the-best.

A one-reason decision rule that examines cues in order of validity and decides on the first cue that discriminates between the options, ignoring all remaining cues.

Key Researchers

Gerd Gigerenzer. Director emeritus at the Max Planck Institute for Human Development in Berlin; with Peter Todd and the ABC Research Group he built the fast-and-frugal heuristics program and the theory of ecological rationality. ORCID - Google Scholar - Faculty Page - Wikipedia

Daniel Kahneman (1934-2024). Psychologist at Princeton University and 2002 Nobel laureate who, with Amos Tversky, founded the heuristics-and-biases program and later framed it explicitly as a study of the boundary of rationality. Google Scholar - Nobel Biographical - Wikipedia

Herbert A. Simon (1916-2001). Polymath at Carnegie Mellon University, 1978 Nobel laureate and 1975 Turing Award winner, who introduced bounded rationality and satisficing and founded the procedural view of rational choice. CMU Profile - Wikipedia

Peter M. Todd. Provost Professor at Indiana University Bloomington; co-author of Simple Heuristics That Make Us Smart and a principal architect of the ecological rationality research program. Google Scholar - Faculty Page

Amos Tversky (1937-1996). Cognitive psychologist at Stanford University who, with Kahneman, documented the heuristics and biases through which bounded minds depart from the norms of probability. NAS Directory - Wikipedia

Frequently Asked Questions

What is bounded rationality?
Bounded rationality is Herbert Simon's principle that the rationality of real agents is limited by the difficulty of the decision problem, the finite computational capacity of the mind, and the time available to decide, so that agents use workable procedures rather than compute an optimum (Simon, 1955).

How is bounded rationality different from irrationality?
It is not a claim that people are irrational. Bounded rationality holds that a finite agent cannot execute the ideal optimization, so choosing a sensible method to reach a good result is the only rationality available to it; Simon called this procedural rationality, judged by the process rather than the outcome (Simon, 1979).

What is satisficing?
Satisficing is Simon's alternative to maximizing: instead of examining every option to find the best, the agent sets an aspiration level and accepts the first alternative that meets or exceeds it, which turns an intractable search into one with a clear stopping rule (Simon, 1956).

What is Simon's scissors metaphor?
Simon likened rational behavior to a pair of scissors whose two blades are the cognitive limits of the agent and the structure of the task environment; behavior is the point where they meet, so neither the mind nor the environment explains it alone (Simon, 1990).

What is ecological rationality?
Ecological rationality is the degree to which a strategy is adapted to the environment in which it is used. On this view a heuristic is judged by its accuracy in the world rather than its coherence with a logical norm, so rationality is a relation between a mind and its niche (Todd & Gigerenzer, 2007).

What is the less-is-more effect?
The less-is-more effect is the finding that using or having less information can produce more accurate inferences, as when recognizing fewer objects raises the accuracy of the recognition heuristic because the heuristic can only fire when exactly one object is recognized (Gigerenzer & Goldstein, 1996).

How does bounded rationality relate to heuristics and biases?
The heuristics-and-biases program of Kahneman and Tversky studies the bounded mind's shortcuts and measures how they deviate from probability and logic, reading the deviations as systematic bias; it is one of two major traditions that descend from Simon's bounded agent (Tversky & Kahneman, 1974).

Did Herbert Simon win a Nobel Prize for bounded rationality?
Yes. Simon received the 1978 Nobel Memorial Prize in Economic Sciences for his research into the decision-making process within economic organizations, work built on bounded rationality and satisficing (Simon, 1979).

References