he textbook decision-maker knows every option, every outcome and every probability, and picks the best. Nobody is like that. Bounded rationality is the study of how real minds decide when computation is expensive and information is short, and of when their shortcuts work.

the question

Herbert Simon put it as a challenge to economics (Simon, 1955): replace the all-knowing chooser with one that has the knowledge and computing power an organism actually has [paraphrase, not a quotation]. A year later Simon (1956) introduced satisficing: searching until you find an option that is good enough by some aspiration level, then stopping. He also insisted that rationality depends on the fit between a mind's methods and the structure of its environment.

the classic findings

Two research programmes grew from this. Tversky & Kahneman (1974) described three heuristics, representativeness, availability, and anchoring and adjustment, and catalogued the systematic errors they produce, including neglect of base rates. Kahneman & Tversky (1979) then offered prospect theory: people evaluate outcomes as gains and losses from a reference point, feel losses more strongly than equal gains, and weight probabilities non-linearly.

The second programme read the same shortcuts more kindly. Gigerenzer & Goldstein (1996) showed in simulations that a simple rule, "take the best" cue and ignore the rest, could match or beat multiple regression at predicting real-world quantities. On this view heuristics are not failures of rationality but tools adapted to particular environments, and the right question is where each one works.

what held up, and what did not

Loss aversion and reference dependence are among the most replicated results in behavioral economics, although the size of loss aversion varies a lot with how it is measured. Classic anchoring effects are generally robust. Some of the more striking "priming" demonstrations that were sometimes grouped with this literature did not replicate (see replication). The argument between the two programmes is less about the data than about the yardstick: compared with what is a heuristic an error?

in machines

Large language models make a strange new subject for this research. Binz & Schulz (2023) gave GPT-3 classic vignette problems from the judgment and decision literature. It did as well as or better than people on several of them and showed some human-like patterns, but small changes to a problem's wording could throw it, and it did poorly on tasks that needed causal reasoning [paraphrase, not a quotation]. Whether a model that has read about the Linda problem is "showing a bias" or reciting one is a real methodological difficulty.

live debates

Biases, or ecologically sensible heuristics? The exchange between Kahneman & Tversky (1996) and Gigerenzer (1996) is still the reference point. Gigerenzer & Brighton (2009) develop the "less can be more" side with the bias–variance trade-off: a simple heuristic that ignores information can predict better out of sample. Stanovich & West (2000) brought in individual differences, finding that normative responding tracks cognitive ability, which fits neither a pure-bias nor a pure-adaptation story neatly [paraphrase, not a quotation].

Rational, but with costs. Lieder & Griffiths (2020) recast bounded rationality as resource-rational analysis: behaviour as the optimal use of limited computation. Gal & Rucker (2018) meanwhile challenged one of the best-known biases, arguing that the evidence for general loss aversion is weaker than its fame.

methodological critiques

The flexibility problem. Bowers & Davis (2012) and Jones & Love (2011) argue that Bayesian models can fit almost any behaviour by choosing priors, likelihoods and cost functions after the fact; resource-rational models inherit the risk with the added freedom of choosing the costs. The heuristics programme has the mirror problem: ecological rationality depends on the environment the researcher chooses to simulate.

the open problem

Simon's own framing still has no full answer: a theory that predicts, for a given mind and environment, which shortcut will be used and how well it will do. The same theory would tell us what a resource-limited machine should do, which is why this old question is back.

Reading list

1. Simon (1955). A Behavioral Model of Rational Choice. The Quarterly Journal of Economics. doi:10.2307/1884852
2. Simon (1956). Rational choice and the structure of the environment. Psychological Review. doi:10.1037/h0042769
3. Tversky & Kahneman (1974). Judgment under Uncertainty: Heuristics and Biases. Science. doi:10.1126/science.185.4157.1124
4. Kahneman & Tversky (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica. doi:10.2307/1914185
5. Gigerenzer & Goldstein (1996). Reasoning the fast and frugal way: Models of bounded rationality. Psychological Review. doi:10.1037/0033-295X.103.4.650
6. Binz & Schulz (2023). Using cognitive psychology to understand GPT-3. Proceedings of the National Academy of Sciences. doi:10.1073/pnas.2218523120
7. Kahneman & Tversky (1996). On the reality of cognitive illusions. Psychological Review. doi:10.1037/0033-295X.103.3.582
8. Gigerenzer & Brighton (2009). Homo Heuristicus: Why Biased Minds Make Better Inferences. Topics in Cognitive Science. doi:10.1111/j.1756-8765.2008.01006.x
9. Stanovich & West (2000). Individual differences in reasoning: Implications for the rationality debate?. Behavioral and Brain Sciences. doi:10.1017/S0140525X00003435
10. Lieder & Griffiths (2020). Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources. Behavioral and Brain Sciences. doi:10.1017/S0140525X1900061X
11. Gal & Rucker (2018). The Loss of Loss Aversion: Will It Loom Larger Than Its Gain?. Journal of Consumer Psychology. doi:10.1002/jcpy.1047
12. Bowers & Davis (2012). Bayesian just-so stories in psychology and neuroscience. Psychological Bulletin. doi:10.1037/a0026450
13. Jones & Love (2011). Bayesian Fundamentalism or Enlightenment? On the explanatory status and theoretical contributions of Bayesian models of cognition. Behavioral and Brain Sciences. doi:10.1017/S0140525X10003134

Try it yourself. [playful] Small toys on the toy shelf that show this in action:   • the four cards — the selection task

see also

	Herbert Simon — satisficing, the Logic Theorist, and organizations
	glossary: satisficing — with heuristic and base rate
	Bayes, 1763 — the yardstick the heuristics are measured against
	calibration — confidence as a decision about yourself