Foundationsbehavioral
Bounded Rationality: Thinking With Limits
Overview
Before Kahneman, there was Simon: humans have limited computation, memory, and attention, so they satisfice — choosing 'good enough' — rather than optimize. Bounded rationality is the deep foundation beneath every bias in this hub: heuristics exist because optimization is impossible. This lesson covers Simon's original idea, satisficing vs. maximizing, the heuristics-and-biases program as its child, Gigerenzer's defense of heuristics, and what bounded rationality implies for institutions.
Learning objectives
- Define bounded rationality and satisficing
- Contrast substantive vs. procedural rationality
- Explain heuristics as adaptations to cognitive limits
- Evaluate Gigerenzer's 'fast and frugal' defense
- Apply bounded rationality to institutional design
Simon's insight
Herbert Simon (1950s, Nobel 1978): the rational agent of economics — with unlimited computation, perfect information, and infinite time — is a fiction. Real decision-makers face uncertainty, complexity, and cognitive limits, so they use simplified models of the world and satisfice: search until an option meets an aspiration level, then stop. A homebuyer doesn't evaluate every house (impossible) — she sets criteria and takes the first good-enough match. This isn't stupidity; it's the only feasible strategy. Simon's distinction: substantive rationality (the optimal choice) vs. procedural rationality (a reasonable decision procedure given limits).
Search until 'good enough' rather than optimal. Given real search costs and cognitive limits, satisficing is often the rational procedure — optimizing the decision procedure, not just the decision.
Heuristics: the bias program's parent
Kahneman and Tversky's heuristics-and-biases program is bounded rationality's empirical child: it cataloged the specific shortcuts (availability, representativeness, anchoring) and their systematic errors. The logic: heuristics are efficient given limits — judging frequency by availability usually works — but produce predictable biases in edge cases. This reframes 'biases' charitably: they're not design flaws but the cost of a system optimized for speed and frugality under constraints. Understanding bounded rationality first makes every bias in this hub intelligible as an adaptation pushed past its comfort zone.
Simon: human thought is one blade of scissors; the task environment is the other. Behavior comes from the fit between mind and environment — study both blades, not just the mind.
Gigerenzer's defense: fast and frugal
Gerd Gigerenzer argues heuristics aren't poor substitutes for optimization — they're often better. 'Fast and frugal' heuristics (take-the-best, recognition, tit-for-tat) exploit environmental structure and can outperform complex models, especially under uncertainty (the less-is-more effect: simple rules beat regressions on out-of-sample prediction). His ecological rationality: judge a heuristic by its fit to the environment, not by logic-textbook standards. The debate with Kahneman was real but partly terminological: both agree heuristics err systematically in some environments; they differ on whether to emphasize the errors (biases) or the successes (ecological fit).
Simple heuristics can out-predict complex models when data is noisy and environments uncertain — overfitting is optimization's curse. Frugality is a feature, not just a constraint.
Attention: the scarcest resource
Modern bounded rationality centers on limited attention: we can't process most available information, so what grabs attention drives choice. Applications: salience effects (vivid risks overweighted), inattention to shrouded attributes (add-on fees, overdraft charges — the Gabaix-Laibson account of exploitative pricing), and choice overload (too many options → worse decisions or paralysis — the famous jam experiment, debated but influential). Firms strategically manage attention: drip pricing, partitioned tariffs, and dark patterns all exploit inattention. Policy response: mandated salience (all-in pricing, standardized disclosures) — regulating what must grab attention.
When some price components are hidden (fees, add-ons), competition focuses on the visible base price and firms profit on the shrouded margins — inattention as a business model.
Designing institutions for bounded minds
If minds are bounded, institutions should be the prosthetics. Simplify: fewer, clearer options (Sweden's pension menu vs. America's 401(k) maze). Standardize: comparable disclosures, default rules, cooling-off periods. Automate: auto-enrollment, auto-escalation, robo-advice — outsourcing decisions to well-designed algorithms. Debias procedures, not people: checklists (surgery, aviation), structured interviews, premortems. Simon's ultimate lesson: don't lament cognitive limits — build environments where bounded minds decide well. That's choice architecture's deepest justification, and this hub's unifying theme.
Key points
- Bounded rationality (Simon): cognitive limits make optimization impossible; we satisfice with heuristics.
- Heuristics are efficient adaptations — biases are their predictable misfires outside their comfort zone.
- Gigerenzer's ecological rationality: judge heuristics by environment-fit; simple rules often beat complex models.
- Limited attention is the modern focus: salience, shrouded attributes, and choice overload shape markets.
- Institutional design — simplification, defaults, automation, checklists — is the constructive response.
Common misconceptions
✓ Truth: It means optimization is computationally impossible for finite minds. Satisficing and heuristics are often the best feasible procedures — smart given constraints.
✓ Truth: More information can worsen decisions (overload, overfitting). The binding constraint is processing capacity, not data — which is why simplification helps.
✓ Truth: Under uncertainty, fast-and-frugal heuristics often out-predict complex models (less-is-more). The right question is fit to the environment, not sophistication.
Practice questions
Q1. Herbert Simon's 'satisficing' means:
- Maximizing utility perfectly
- Choosing the first option that meets an aspiration level, given cognitive limits and search costs
- Acting randomly
- Copying others
Optimization is infeasible for finite minds; satisficing — search until good enough — is the procedurally rational strategy. It's the foundation of bounded rationality.
Q2. Simon's 'scissors' metaphor says behavior is explained by:
- Genetics alone
- The fit between the mind's capabilities and the task environment's structure — both blades
- Incentives alone
- Culture alone
One blade is cognition, the other the environment. Heuristics look smart or biased depending on the environment they cut — hence ecological rationality.
Q3. Gigerenzer's 'less-is-more' effect claims:
- Ignorance is bliss
- Simple heuristics can out-predict complex models under uncertainty by avoiding overfitting
- More data is always better
- Experts are useless
Frugal rules exploit environmental structure and dodge optimization's curse (overfitting). In noisy, uncertain domains, simplicity wins — a qualified defense of heuristics.
Click a question to reveal the answer and explanation.
Frequently asked questions
What is satisficing?
Simon's term for choosing the first option meeting an aspiration level, rather than searching for the optimum. Rational given search costs and cognitive limits — the feasible optimum of decision procedures.
How does bounded rationality differ from the biases program?
Bounded rationality is the general framework (limits → simplified procedures); heuristics-and-biases is its empirical research program cataloging specific shortcuts and errors. Parent and child.
What is ecological rationality?
Gigerenzer's criterion: a heuristic is rational if it fits its environment's structure. Take-the-best works where cues are skewed; it fails where they're not. Rationality is a mind-environment match.
Does AI solve bounded rationality?
It extends bounds but doesn't eliminate them — humans still choose among AI outputs with limited attention, and algorithms have their own biases. The design problem moves up a level.