What Is Bounded Rationality?
Bounded rationality is the study of how decisions are made when information, time, attention, memory, and computational capacity are limited. Associated principally with Herbert A. Simon, the concept explains why people and organizations often search for an acceptable course of action rather than calculating the single best alternative across every possible outcome. Its central proposition is that the process of deciding has costs and that the feasible quality of a decision depends on the capabilities of the decision maker and the structure of the environment. Simon, 1955; Simon, 1956.
A person who compares a handful of apartments, chooses one that meets essential requirements, and ends the search may be satisficing. The person sets an aspiration threshold—perhaps safe access, acceptable price, adequate size, and a workable commute—and stops when an option clears it. This is a procedure for choosing under practical constraints. Whether it yields a good result depends on the threshold, the order and distribution of available apartments, the costs of continued search, the stakes, and the information available about options still unseen.
Bounded rationality is a framework about feasible cognition and decision procedures. Satisficing is one family of stopping rules within that framework. Some bounded decision makers use elimination rules, recognition, simple weighted scores, or expert pattern recognition instead. Some pursue optimization over a smaller, strategically selected option set. What unites these possibilities is explicit attention to the costs and constraints of information acquisition, representation, computation, and action.
The scientific question is therefore precise: given a decision maker with particular resources in a particular environment, what information is acquired, how are alternatives evaluated, when does search stop, and what follows for decision quality? Contemporary research distinguishes an idealized outcome benchmark, the actual cognitive procedure, and the performance of that procedure under experimentally or realistically specified conditions. Payne, Bettman, and Johnson, 1993; Lieder and Griffiths, 2020.
Herbert Simon and the Origin of the Concept
From ideal choice to feasible choice
Classical models of rational choice typically specify a set of alternatives, possible consequences, probabilities or beliefs, and preferences, then characterize a choice that maximizes an objective subject to stipulated constraints. Such models are powerful analytical standards, particularly when the objective, feasible set, and probabilities can be established. Simon argued that descriptive explanations of organizational and human behavior must also account for how an agent discovers the alternatives, learns about consequences, represents goals, and uses limited cognitive resources. Simon, 1955.
In his 1955 article A Behavioral Model of Rational Choice, Simon examined how a more psychologically realistic agent might behave when full information and unlimited computation are unavailable. The 1956 article Rational Choice and the Structure of the Environment connected elementary choice mechanisms with the conditions of the environment in which they operate. A good explanation thus requires both a model of the decision maker and a model of the task ecology. Simon, 1955; Simon, 1956.
This line of research continued into the empirical study of problem solving. Newell and Simon analyzed how people construct representations, pursue subgoals, and use heuristic procedures to navigate large problem spaces. Their work helped place cognition itself—rather than only the final selected alternative—at the center of explanatory models. Newell and Simon, 1972.
Procedural and substantive rationality
Substantive rationality evaluates an outcome against a criterion: Did the agent select the alternative with the highest expected utility, highest expected payoff, or greatest score on a well-defined objective? Procedural rationality evaluates the method used to reach a decision, including feasibility under constraints. The concepts can be combined. A procedure may be sensible given the available evidence and computational budget while sometimes selecting an option below the theoretical optimum. Conversely, a favorable outcome can occur accidentally after poor information processing.
The distinction matters in experiments. If investigators know every option's payoff and assign a common objective, they can calculate outcome regret or the fraction of trials on which a participant selected the maximum. To evaluate procedural rationality they also need the information participants actually saw, search costs, processing demands, and stopping opportunities. Outcome and procedure answer different questions and require different measurements. Payne, Bettman, and Johnson, 1993.
What Does Satisficing Mean?
Satisficing is a strategy of selecting an alternative when it satisfies a specified aspiration level or set of acceptable conditions. Its defining feature is a stopping criterion. A decision maker does not have to rank every feasible option to act. Once an option qualifies, further search may be unnecessary for the chosen goal, particularly when extra comparisons require time, effort, money, or delay. Simon, 1955.
Imagine three employment offers. One offers the highest salary but requires substantial relocation. Another offers a sufficiently high salary, a desirable location, and a stable work schedule. A third remains unknown because interviewing for it would take several weeks. A satisficing decision maker might select the second offer once all essential conditions are met. A maximization model might instead compare expected utility across all discoverable offers, accounting for compensation, time, work quality, relocation, risk, and the value of future search. The apparent disagreement can narrow when search costs and uncertainty are modeled explicitly.
The aspiration level matters as much as the stopping rule. A threshold set too low may accept clearly inferior opportunities. One set above what the environment can supply may cause endless searching or repeated failure to choose. Standards may be learned from experience, peers, institutional rules, or prior outcomes; they can be revised when options prove better or worse than expected. A fixed acceptance rule can therefore be studied separately from how the threshold itself is formed or adapted.
A simple formal example
Suppose a decision maker encounters options in sequence. Each option i has a subjectively estimated value V(i), and the decision maker specifies an aspiration threshold A. A simplified satisficing rule accepts the first option for which V(i) is at least A. More complex rules require multiple attributes to clear minimum thresholds, give some attributes veto power, or allow the aspiration threshold to change with elapsed time and observed quality.
For example, if acceptable apartments must cost no more than $1,800 a month, meet a safety requirement, and be reachable by public transit, the decision maker can use those as conjunctive acceptance conditions. An apartment priced at $1,700 that meets the other conditions ends the search. This rule differs from finding the apartment with the highest weighted combination of location, space, amenities, and cost among every option on the market.
These numbers illustrate the logic of the strategy, not a measured universal human threshold. In research, the exact rule should be represented computationally before declaring that an observed participant was satisficing. Many different processes can produce the same final choice, including preferences for the first option, lack of attention, a changing aspiration level, and expected-cost calculations.
Stopping is an active part of deciding
Search can be costly in at least four ways: direct expenditure on obtaining information; cognitive effort to encode and compare attributes; opportunity cost of delaying another activity; and loss of an option while searching further. An extra search step may yield useful information, but the expected value of that information must be considered alongside its cost. A stopping rule makes that tradeoff operational, even when the person cannot solve a formal optimal-stopping problem.
Satisficing is consequently compatible with high standards. A surgeon using a time-sensitive protocol, a technical expert requiring stringent safety criteria, and a student choosing a suitable course may all stop once their task-relevant standard is reached. How demanding a standard is and whether one keeps searching after meeting it are analytically independent dimensions. Measurement research on maximizing and satisficing repeatedly emphasizes this distinction. Cheek and Schwartz, 2016.
Why Is Perfect Optimization So Difficult?
Alternatives are discovered rather than given
Many practical decisions lack a known complete option set. A consumer searches websites; a team generates project plans; a clinician considers plausible diagnoses; a researcher formulates hypotheses. Searching changes the set of alternatives the person knows. Even specifying what should count as an alternative may require judgment. An experiment with five fully displayed options creates a different information problem from a real decision involving an unknown number of future possibilities.
Probabilities and outcomes are uncertain
Expected-utility calculations require beliefs about what may happen after each choice. Some environments allow stable frequency estimates. Others involve changing conditions, novel technologies, strategic opponents, or rare events with limited data. When probability estimates are incomplete or unreliable, a formally precise calculation can create an impression of certainty that the evidence does not support. A simple robust rule may be preferable under certain forms of uncertainty, although this requires task-specific comparison. Gigerenzer and Brighton, 2009.
Cognition is capacity-limited
Attention selects information, working memory maintains a limited amount of task-relevant material, and people often cannot retain or evaluate many independent attributes at once. In influential work on information processing, Miller described limits and recoding strategies that shape performance. The precise capacity depends on the task and representation; the familiar numerical slogan about short-term memory is not itself a general law of decision making. Miller, 1956.
Computation takes time and effort
Even with the same information, calculating a complete choice ranking can involve large numbers of comparisons. A person may need to estimate missing values, determine attribute weights, consider future consequences, and reconcile conflicting goals. Decisions also compete with other tasks for cognitive resources. The adaptive decision-maker framework models selection among strategies that offer different combinations of accuracy and effort, rather than assuming one fixed processing method for all problems. Payne, Bettman, and Johnson, 1993.
Preferences can be constructed during choice
People sometimes begin a decision with incomplete or context-sensitive preferences. Which attribute they emphasize can change with framing, available comparisons, goals, and emotion. Research on constructive choice identifies systematic adaptation of decision strategies to task demands. The researcher should not automatically attribute every choice inconsistency to irrationality when the evaluation criteria themselves are being formed during the task. Bettman, Luce, and Payne, 1998.
Normative Models: What Is the Correct Comparison?
Expected utility and expected value
When alternatives, probabilities, and payoffs are specified, expected value is the probability-weighted average payoff, and expected utility is the probability-weighted average utility. A choice that fails to maximize expected monetary value can reflect legitimate attitudes toward risk, liquidity, fairness, or future options. An investigator must establish whether the benchmark is money, utility, a multiattribute score, or an institutionally specified target before using the language of error.
For example, choosing $90 for certain instead of a 50 percent chance of $200 gives up $10 in expected monetary value, but may be consistent with a person's risk preferences. It is an error only relative to a chosen and justified criterion such as maximizing expected dollars with no other valued consequences. The same caution applies when evaluating an acceptance threshold against a supposed ideal choice.
Optimization with search costs
The word optimize can refer to several distinct problems. One might maximize payoff as if every option were free to inspect, maximize expected payoff net of search cost, or optimize a decision procedure within a finite computational budget. A strategy that stops early can be suboptimal under the first benchmark and sensible under the second. These comparisons should be reported explicitly rather than treating exhaustive search as a costless default.
Consider ten equally accessible products, one worth 100 units and another worth 95. An exhaustive search costs 20 units in time and effort; a short search costing 3 units identifies the 95-unit product. Net value is 80 versus 92 in this illustration. This arithmetic describes a constructed example, not a general empirical finding. It shows why a calculation that excludes the cost of deliberation can misclassify a strategy's practical performance.
Bayesian and robust approaches
Bayesian models describe how beliefs may be updated in light of evidence, given a prior and a likelihood. They can provide normative comparisons for information search when those assumptions are defensible. Robust approaches instead emphasize satisfactory performance across several plausible scenarios when probabilities are uncertain. Bounded rationality does not dictate one universal normative criterion; it requires the investigator to state the criterion and explain how the agent could approximate or use it.
Resource-rational analysis
Resource-rational analysis formalizes the idea that cognition may be adapted to a computational budget. Rather than demanding the highest performance under unlimited processing, it examines which operations would make efficient use of constrained resources. Lieder and Griffiths develop this as a framework connecting cognitive models with computation and normative analysis. A critical distinction remains: finding a resource-efficient model that predicts behavior does not establish that every observed human decision is in fact optimal for its resources. Lieder and Griffiths, 2020; Rahnev, 2020.
How People Adapt Their Decision Strategies
Compensation and elimination
A compensatory rule allows strength on one attribute to offset weakness on another, as in a weighted additive score. A noncompensatory rule may reject any option that violates a critical requirement or select the alternative with the most important favorable cue. An apartment can fail a hard accessibility requirement regardless of its low price; alternatively, price and location can trade off continuously. These rules impose different information requirements and therefore different effort costs.
Lexicographic and one-reason decisions
A lexicographic rule compares alternatives on the most important attribute and considers additional attributes only to break ties. Other heuristics prioritize particular cues or stop as soon as one sufficiently discriminative cue is found. Their accuracy depends on cue validity, redundancy, noise, and the structure of the environment. A simple rule can perform well when its first cue is informative and poorly when the cue is misleading. Gigerenzer and Gaissmaier, 2011.
Strategy selection varies with task demands
Experiments in multiattribute choice ask participants to decide among options while researchers manipulate information load, search cost, time pressure, payoff structure, or display format. Process-tracing methods can reveal which information participants inspect and in what order. Findings from adaptive decision research support the proposition that people vary their information acquisition and strategy use across tasks. They cannot be reduced to a single rule used by everyone. Payne, Bettman, and Johnson, 1993; Bettman, Luce, and Payne, 1998.
Recognition, expertise, and experience
Experienced decision makers can recognize meaningful patterns and narrow their search rapidly, especially in settings with reliable feedback and learnable regularities. A short decision time may reflect practiced discrimination rather than a low standard. Conversely, familiarity can sustain outdated rules when a context changes. Tests of expertise therefore need to measure prediction or decision quality under the actual environmental conditions, not merely decision speed.
Ecological Rationality: When Less Can Be More
Ecological rationality evaluates a decision procedure in relation to the statistical and functional structure of its environment. The same rule can be successful in a predictable setting with stable cues and unsuccessful in a different setting. Research on fast-and-frugal heuristics shows why ignoring some available information can sometimes improve generalization: complex models may fit noise, require unreliable parameter estimates, or impose excessive information costs. Gigerenzer and Brighton, 2009; Todd and Gigerenzer, 2012.
A simple rule should be tested against relevant competitors on new data, with prediction loss or actual outcomes reported. Good performance on the examples used to devise a heuristic is weaker evidence than out-of-sample testing. Complex models can outperform simple ones where abundant representative data, reliable features, and adequate computation are available. A claim that a particular shortcut is superior requires a specified task, comparison model, sample, and evaluation metric.
Ecological rationality and bounded rationality intersect but retain different emphases. Bounded rationality analyzes decision procedures under limits; ecological rationality identifies how strategies exploit environmental structure. Both encourage explicit models rather than a blanket judgment that more thought is always better or that shortcuts invariably harm accuracy. Gigerenzer and Gaissmaier, 2011.
What Experiments Actually Measure
Sequential search and stopping thresholds
In sequential search experiments, participants examine alternatives one at a time and decide whether to accept an option or pay to continue. Researchers can vary option distributions, search costs, the visibility of previously rejected options, and the consequences of ending the search. Behavioral outcomes include the number of options inspected, acceptance thresholds, payoff, time, and sensitivity to changes in costs.
Observing early stopping alone is insufficient evidence of a satisficing rule. If an optimal policy also stops early when search is costly, behavior consistent with both models cannot distinguish them. A discriminating experiment changes conditions under which the two procedures predict different acceptance decisions and examines their fit to individual-level sequences rather than only aggregate acceptance rates.
Information-board and eye-tracking tasks
Information-board studies hide attributes behind selectable cells, recording the sequence of inspections. Eye movements can give complementary, though imperfect, evidence about attention. Researchers ask whether participants compare alternatives attribute by attribute, inspect all attributes of one option, focus on a high-priority cue, or shift strategies as tasks become harder. Looking at a value does not guarantee it was encoded or used; not looking at it does not prove it was cognitively irrelevant if the information was already learned.
Choice quality and regret
For objective-payoff experiments, choice quality may be measured by the share of optimal choices, total payoff, expected net payoff, or regret defined as the difference between selected and best attainable performance. A study must specify whether information-search costs are deducted. If payoffs depend on private preferences, participant-reported utility or incentive-compatible choices may be more appropriate than a researcher-imposed score. Observed regret is also different from self-reported feelings of regret.
Computational model comparison
Formal models can fit acceptance thresholds, information-sampling policies, or alternative-scoring weights. Predictive comparisons should use held-out trials or participants where feasible, model complexity penalties, and robustness checks. Parameter recovery is useful: if simulated participants with a known threshold cannot be correctly identified by the estimation method, claims about psychological parameters are weak. A successful model provides a constrained explanation of behavior, not a direct reading of internal mental experience.
Evidence From Classical and Contemporary Research
The foundational theoretical articles by Simon established a research program, not a single experimental estimate of how often people satisfice. Later studies have measured specific manifestations under controlled conditions. The literature includes computational models, laboratory tasks, process tracing, individual-difference questionnaires, and field-oriented decision applications. These methods converge on the importance of constraints but differ in what they can establish. Simon, 1955; Simon, 1956.
In a study of aging and adaptive strategy selection, Mata, Schooler, and Rieskamp examined 163 participants in environments that favored different search strategies. Older adults tended to inspect less information and use simpler strategies, while both age groups adjusted strategy selection to environmental structure. This is evidence about one set of experimentally created environments and age comparisons, rather than proof that a particular chronological age universally determines rationality. Mata, Schooler, and Rieskamp, 2007.
Besedeš and colleagues designed multi-option tasks resembling complex benefit and financial-plan choices. They could identify the objectively optimal option under their task payoffs and found that selecting it became less likely as the number of options increased, with age-related differences under those conditions. This establishes a result for those controlled tasks, while outcomes in real insurance markets also depend on literacy, market incentives, plan design, and the opportunity to seek advice. Besedeš et al., 2012.
Research on adaptive decision strategies emphasizes experimental manipulations of the decision environment, not a stable percentage of universally bounded individuals. Studies of heuristic performance compare simple decision procedures with more information-intensive rules and show environment-dependent advantages and failures. Because datasets, objectives, complexity penalties, and payoff scales differ, a single effect size for bounded rationality would combine non-equivalent constructs. Gigerenzer and Brighton, 2009; Gigerenzer and Gaissmaier, 2011.
Reproducibility and competing explanations
Robust evidence requires testing the same formal predictions across independent samples and environments. Search length can change because people face different costs, have different beliefs about the distribution of offers, or possess different goals. Questionnaire scores can change with wording and factor structure. Model comparisons should therefore report uncertainty, alternative models, preregistration when possible, and out-of-sample fit.
The broad claim that decisions use constrained resources is well supported as an explanatory premise. A claim that all rapid decisions are efficient, that a particular heuristic always outperforms optimization, or that any individual necessarily stops at a fixed personal aspiration level is much stronger and often unsupported. Scientific inference should proceed from a defined mechanism to falsifiable observations.
Satisficing in Everyday Decisions
Consumer choice
A shopper may screen products for essential characteristics and then stop when one qualifies. This saves effort when the cost of searching is substantial relative to expected gains. It can also leave money or quality on the table when the first acceptable product is overpriced or misleadingly advertised. Consumer choice research shows that preferences and search strategies depend on task structure and presentation, so an adequate account includes the actual option set and constraints. Bettman, Luce, and Payne, 1998.
Employment, education, and housing
These decisions combine multiple valued attributes that may resist a single numerical scale. Screening rules can protect essential needs—such as accessibility, adequate compensation, or accreditation—while allowing a practical stopping decision. When stakes are high or choices are difficult to reverse, investing in more information may be worthwhile. A defensible stopping rule can specify mandatory conditions, a review period, and a trigger for reopening the search if new evidence materially changes the evaluation.
Health and professional practice
In clinical and institutional settings, rapid acceptance of a seemingly adequate option can be risky when overlooked alternatives have major consequences. Professional protocols often incorporate mandatory safety checks, independent review, and escalation criteria precisely because search must be limited while harmful omissions remain possible. The appropriate threshold depends on decision risk and the evidence base. Bounded rationality describes a decision problem here; it does not justify bypassing validated clinical assessment or standards of care.
Organizations and public policy
Organizations search among feasible projects, hire under budget constraints, and adapt to changing information. The group may use explicit minimum conditions, delegated decision authority, and escalation rules. An organizational threshold can coordinate action, yet it can also preserve weak routines when accountability, incentives, or updated evidence are missing. The important design issue is whether the procedure can detect changes that justify further search.
Maximizing Versus Satisficing: An Important Distinction
In individual-difference psychology, maximizing often refers to the goal of selecting the best possible option and searching extensively in pursuit of that goal. Satisficing refers to accepting an option that reaches a personal standard. In Simon's theory, satisficing primarily describes a decision procedure under constraints. A later personality-style research program studies people's reported goals, search behaviors, and emotional experiences across decisions. These research traditions overlap but their measures and explanatory targets are not interchangeable. Schwartz et al., 2002; Cheek and Schwartz, 2016.
Schwartz and colleagues developed a Maximization Scale and reported associations between stronger maximizing tendencies and lower scores on several well-being measures in their samples, as well as greater regret or social comparison in some tasks. Those findings generated substantial research but do not establish a universal causal rule that seeking a better option produces unhappiness. The measures, constructs, sampled populations, and study designs must be considered before applying the claim to individuals. Schwartz et al., 2002.
Cheek and Schwartz reviewed multiple measures of maximization and argued for separating the goal of choosing the best option from the strategy of alternative search. Decision difficulty and regret may be antecedents or consequences rather than defining parts of maximization itself. This matters because a questionnaire that mixes maximizing with negative affect can inflate correlations between the scale and distress through overlapping item content. Cheek and Schwartz, 2016.
An exacting person can satisfice using a very demanding acceptance threshold. A person with modest standards can continue comparing alternatives because they enjoy the search. A strategy may also change with stakes: someone can satisfice on routine purchases while undertaking extensive optimization for a medical or financial decision. Treating people as two fixed psychological types loses these distinctions.
Bounded Rationality and Cognitive Bias
A cognitive bias refers to a systematic deviation from a specified benchmark under defined conditions. Bounded rationality describes the limited capacities and informational conditions in which a decision is produced. The existence of capacity limits does not show that a particular choice violates a normative standard. Some biases may emerge from selective information search, but adaptive heuristics can also improve prediction under uncertainty. A mechanism, a result, and a benchmark must be evaluated separately. Gigerenzer and Gaissmaier, 2011.
For instance, considering only the first three options may produce a poor result if the best option consistently appears fourth and inspection costs are negligible. The same search rule may be efficient if additional options rarely provide material improvements and every comparison incurs a real cost. Whether the rule is biased depends on the standard used; whether it is adaptive depends on actual performance in the relevant environment.
Bounded rationality is also distinct from a logical fallacy, a mathematical calculation error, a preference for risk, and a clinical symptom. A logical argument can be invalid even if it is constructed slowly and with ample resources. Risk preferences can be coherent even when they differ from expected-value maximization. A person's use of shortcuts or satisficing procedures alone does not indicate a mental disorder.
Bounded Rationality and Dual-Process Accounts
Dual-process theories investigate distinctions among relatively automatic and more deliberative processing; their theoretical formulations vary considerably. Bounded rationality concerns resource limits and decision procedures and does not require every decision to arise from two separate systems. Deliberative computation can be limited and strategically selective; rapid expertise can be effective. A correspondence between satisficing and intuition, or between optimization and deliberation, cannot be assumed without process evidence.
The empirical question is how a task changes operations such as attention, memory, sampling, arithmetic, and rule application. Response time by itself does not tell investigators whether an agent used an aspiration rule, a habitual response, a compensatory evaluation, or learned recognition. Carefully designed model comparisons and process measures are more informative than labeling speed as intuition and length as rationality.
Emotion, Motivation, Culture, and Individual Differences
Emotion can influence which goals are salient, how uncertain consequences are valued, and whether continued search feels worthwhile. Anticipated regret may extend search even after an acceptable option appears; urgency may accelerate stopping. Motivation determines how much effort a task deserves, while incentives can influence the willingness to inspect additional evidence. These effects should be tested as changes in thresholds, attention, beliefs, or search behavior rather than assigned to a single universal cause.
Cognitive capacity, domain knowledge, age, and task familiarity may affect feasible strategies. The Mata and colleagues study illustrates that groups can differ in information search while still adjusting to environmental structure. One experiment cannot supply a general age rule. Research should account for prior education, familiarity with the task, numeracy, accessibility of information, and the design of the task itself. Mata, Schooler, and Rieskamp, 2007.
Culture and institutions can influence available alternatives, norms concerning acceptable outcomes, and the costs of obtaining information. Cross-cultural claims require comparable tasks, measurement invariance, suitable samples, and evidence that the same threshold or preference has the same meaning. The scientific literature reviewed here supports the general importance of environment and resources but does not justify a fixed cultural ranking in rationality or satisficing.
When Satisficing Works—and When It Fails
Satisficing has promising properties when a satisfactory outcome is clearly definable, costs of search are meaningful, alternatives arrive sequentially, and the decision maker can reliably recognize whether essential conditions are met. It can conserve attention for more consequential problems, accelerate action when waiting is costly, and reduce exposure to irrelevant differences among options.
The method becomes vulnerable when aspiration levels are poorly calibrated, essential attributes are missing from the screen, offers are presented in a systematically manipulated order, or early options are unrepresentative of later opportunities. It can also perform badly where one further comparison would cheaply reveal a decisive difference. A stopping rule may persist after environmental change unless it includes feedback and revision.
Higher stakes do not automatically imply that exhaustive comparison is the rational choice. What matters is the expected benefit of further information, the reversibility of action, the magnitude of possible harm, the trustworthiness of available evidence, and the cost of delay. In high-risk domains, the response is often a structured procedure that combines a limited search with mandatory critical checks, escalation, and expert review.
Improving Decisions Under Constraints
Define essential criteria before searching
Decision makers can specify minimum requirements, optional preferences, and disqualifying conditions in advance. This reduces opportunistic redefinition of success after seeing a persuasive offer. Requirements should be concrete enough to test and should include factors with potentially severe consequences. For example, a school selection might require accreditation and accessibility before optional amenities are considered.
Separate a threshold from a forecast
A threshold expresses what counts as good enough; a forecast estimates whether a given alternative is likely to meet it. The two can be wrong for different reasons. A person may have a reasonable standard but poorly calibrated beliefs about performance. Where uncertainty is important, express estimates as ranges or probabilities and seek feedback about actual outcomes rather than treating confident impressions as evidence.
Make the search cost explicit
Before another round of comparison, ask which new information could change the decision and how likely it is to do so. If a cheap, reliable test could reveal a major safety problem, the value of additional information may be high. If further search mainly reveals minor aesthetic preferences while a good option may disappear, stopping can be justified. This is a practical approximation to a value-of-information analysis, not a claim that people ordinarily perform the full calculation.
Use structured comparisons where stakes justify them
A short checklist, preselected criteria, base-rate information, and independent verification can improve the reliability of a constrained decision procedure when they target plausible omissions. A weighted comparison may help when attributes can be meaningfully combined and weights represent actual priorities. Such tools themselves require validation: adding more fields or producing a number with decimals does not guarantee better decisions.
Update thresholds with feedback
Repeated decisions provide opportunities to evaluate whether the aspiration level is too lenient, unrealistically high, or insensitive to changes in the environment. Record outcomes, compare predicted with observed results, and revise rules when a consistent pattern appears. Feedback is especially informative where the decision maker sees consequences across enough comparable cases to distinguish signal from chance.
Test transfer rather than assuming universal debiasing
A training exercise that improves performance on the exact practiced problem may or may not generalize to new settings. Better decision processes should be assessed by actual outcomes, net costs, retention over time, and performance on new tasks. The evidence base for any particular intervention must be evaluated directly. Bounded rationality suggests design principles for allocating limited attention; it is not a guarantee that advice or training will produce large universal improvements.
Human–AI Decision Support in the Artificial Era
Digital systems can expand the feasible set of searches, summarize complex alternatives, compute forecasts, and display criteria consistently. This changes the resource environment in which people make decisions. An AI-supported choice may have lower computation costs but create new verification, interpretation, privacy, or coordination costs. Whether overall decision quality improves depends on the reliability of the tool, the user's understanding, the quality of data, and the availability of appropriate oversight.
An apparently comprehensive AI comparison still depends on what alternatives the system retrieved, which criteria it was given, and which omissions remain invisible. Effective human–AI collaboration can make aspiration levels explicit, request competing options, disclose uncertainty, and mark the evidence behind recommendations. Critical safeguards matter where errors are costly. Bounded rationality supplies a useful framework for analyzing this distribution of limited resources between person, interface, organization, and technology. This is an application of decision science, not empirical evidence for any particular philosophical claim about artificial agency.
Angela Bogdanova's Aisentica publications address the Artificial Era at a broader philosophical level. The scientific claims in this article rest on the decision-research literature cited in the References. The philosophical organization of the Artificial Era and the experimentally studied efficiency of particular human decision procedures belong to different forms of argument.
Methodological Limits and Open Scientific Questions
The strongest interpretations of bounded rationality specify the environment, resources, cognitive procedure, and relevant performance criterion. Descriptive research needs access to search sequences, process measures, or model-discriminating experimental manipulations; an outcome alone rarely identifies a decision algorithm. Normative analyses need to justify their objectives and assumptions. Applied studies need to compare laboratory performance with the incentives and uncertainty of actual decisions.
Several questions remain open across the broader literature. How well do laboratory estimates of stopping thresholds generalize to consequential field decisions? When do aspiration levels adapt appropriately rather than simply track biased expectations? Which heuristics are stable within individuals across contexts, and which are rapidly selected in response to task structure? How should cognitive effort be priced in formal models? When does AI assistance reduce total decision cost while maintaining the user's ability to detect errors? These questions require separately designed experiments and cannot be resolved by appealing to the general concept alone.
A useful research report should specify the option distribution, sample, incentives, comparison algorithms, search and time costs, outcome score, statistical uncertainty, preregistration status where available, and robustness of conclusions to alternative benchmark assumptions. It should distinguish a strategy that explains observed behavior from one that performs best in the task and from one recommended for an unfamiliar real-world environment.
Frequently Asked Questions
Why do people satisfice instead of optimize?
People often face incomplete information, an unknown option set, limited time, and the costs of comparing alternatives. Satisficing provides a stopping procedure: select an option when it meets a defined standard. In some environments this conserves resources without sacrificing much outcome quality; in others it may miss a substantially better choice. Simon, 1955.
Is satisficing the same as settling for a poor choice?
No. The acceptance standard can be stringent, and a person can meet it with an excellent option. The scientific assessment depends on the threshold, available alternatives, search costs, and outcome criteria. A high aspiration level with a practical stopping rule can be an effective combination.
What is an example of bounded rationality?
A hiring committee may screen candidates against required qualifications, interview a manageable shortlist, and hire a candidate who exceeds its criteria, rather than interview every potentially qualified person. This is a bounded procedure because gathering and processing information has real costs. Whether it is fair and effective requires additional evidence about screening validity and selection outcomes.
Can a satisficing decision still be optimal?
Yes. If additional search is costly, a policy that accepts a sufficiently good option can maximize expected net value under specified assumptions. A strategy may also happen to find the globally best option without inspecting everything. The relevant benchmark must be stated before judging the result.
What is the difference between bounded rationality and a cognitive bias?
Bounded rationality is a framework for choice under limited resources. A cognitive bias is a systematic departure from a specified criterion under identified conditions. Limitations can contribute to some bias phenomena, but a decision made with limited information is not automatically biased.
Are maximizers necessarily unhappier than satisficers?
Studies have reported associations between certain maximization measures and negative experiences, but measures differ in whether they include goals, search, difficulty, or regret. This makes broad causal claims inappropriate. The construct and outcome must be distinguished. Schwartz et al., 2002; Cheek and Schwartz, 2016.
Does bounded rationality mean people are irrational?
It means cognition is finite and operates in environments that impose informational and temporal constraints. Evaluating rationality requires an explicit criterion and an understanding of what was feasible. Sometimes a simple decision procedure is well adapted; sometimes it performs badly and should be revised.
How does bounded rationality differ from ecological rationality?
Bounded rationality focuses on constrained decision processes. Ecological rationality studies the fit between a strategy and the structure of the environment in which it is used. A full explanation of a successful simple heuristic often uses both perspectives. Todd and Gigerenzer, 2012.
Can AI eliminate bounded rationality?
AI can reduce some information-processing and calculation costs while creating new requirements for evidence evaluation, oversight, and integration with human goals. Decisions still occur under constraints. Better tools change the available resources; they do not remove the need to specify objectives, uncertainty, and stopping rules.
Scientific Conclusion
Bounded rationality explains decision making as a process unfolding under finite resources and incomplete knowledge. Satisficing is one practical mechanism: establish an aspiration level, inspect alternatives, and stop when an acceptable option is found. The quality of the result depends on the environment, search costs, thresholds, evidence, and purpose of the decision. The major advance initiated by Simon is methodological: a scientific account of rational choice must explain how decisions can actually be made, not only which outcome a costless ideal calculation would select.
The most productive current approach compares explicit procedures, tests their predictions, measures decision outcomes alongside information and computation costs, and identifies where rules transfer or fail. This brings together procedural and substantive rationality, adaptive decision research, ecological rationality, and resource-rational models. It supports decisions grounded in evidence, calibrated aspiration levels, and the careful allocation of limited cognitive effort.
Related Articles
Cluster Cognitive Biases, Judgment and Decision-Making — The scientific parent chapter explaining human judgment, heuristics, normative criteria, and choice.
Selective Attention: How the Mind Filters Information — The attentional constraints governing which evidence enters the decision process.
Cognitive Flexibility: How We Adapt to Changing Rules — Adjustment of strategies when goals and environmental demands change.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making — The calibration of human judgment when artificial systems offer advice.
References
Besedeš, T., Deck, C., Sarangi, S., and Shor, M. (2012). Age effects and heuristics in decision making. Review of Economics and Statistics, 94(2), 580–595. https://doi.org/10.1162/REST_a_00174.
Bettman, J. R., Luce, M. F., and Payne, J. W. (1998). Constructive consumer choice processes. Journal of Consumer Research, 25(3), 187–217. https://doi.org/10.1086/209535.
Cheek, N. N., and Schwartz, B. (2016). On the meaning and measurement of maximization. Judgment and Decision Making, 11(2), 126–146. https://doi.org/10.1017/S1930297500007257.
Gigerenzer, G., and Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences. Topics in Cognitive Science, 1(1), 107–143. https://doi.org/10.1111/j.1756-8765.2008.01006.x.
Gigerenzer, G., and Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62, 451–482. https://doi.org/10.1146/annurev-psych-120709-145346.
Lieder, F., and Griffiths, T. L. (2020). Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources. Behavioral and Brain Sciences, 43, e1. https://doi.org/10.1017/S0140525X1900061X.
Mata, R., Schooler, L. J., and Rieskamp, J. (2007). The aging decision maker: Cognitive aging and the adaptive selection of decision strategies. Psychology and Aging, 22(4), 796–810. https://doi.org/10.1037/0882-7974.22.4.796.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97. https://doi.org/10.1037/h0043158.
Newell, A., and Simon, H. A. (1972). Human Problem Solving. Prentice-Hall. https://books.google.com/books/about/Human_Problem_Solving.html?id=__zuAAAAMAAJ.
Payne, J. W., Bettman, J. R., and Johnson, E. J. (1993). The Adaptive Decision Maker. Cambridge University Press. https://doi.org/10.1017/CBO9781139173933.
Rahnev, D. (2020). Resource-rational analysis versus resource-rational humans. Behavioral and Brain Sciences, 43, e19. https://doi.org/10.1017/S0140525X19001699.
Schwartz, B., Ward, A., Monterosso, J., Lyubomirsky, S., White, K., and Lehman, D. R. (2002). Maximizing versus satisficing: Happiness is a matter of choice. Journal of Personality and Social Psychology, 83(5), 1178–1197. https://doi.org/10.1037/0022-3514.83.5.1178.
Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852.
Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2), 129–138. https://doi.org/10.1037/h0042769.
Todd, P. M., and Gigerenzer, G. (2012). Ecological Rationality: Intelligence in the World. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780195315448.001.0001.