Where does bounded rationality collide?

The idea in its original economic form: because a chooser has limited information, limited computation, and limited time, they cannot solve the optimization problem that classical theory assumes, and instead work within a drastically simplified model of the world. Simon's 1955 A behavioral model of rational choice replaced the maximizing agent with one who searches until an option clears an aspiration level and then stops, a rule he later called satisficing, and distinguished substantive rationality, judged by the outcome, from procedural rationality, judged by the process of choosing. The referent is the choosing agent's actual, resource-limited procedure, not an ideal that always computes the best act.[1]

Unit

A chooser's resource-limited decision procedure within a simplified model

Recovery means

Not applicable

Measured by

Studied by tracing search, stopping rules, and aspiration levels rather than by fit to a maximizing solution

Bounded rationality as psychologists have filled it in: accounts of the actual mental procedures a limited mind uses to decide. Two traditions read Simon's programme in opposite keys. Kahneman's 2003 Maps of bounded rationality treats the heuristics people substitute for hard judgements as a source of systematic bias, mapping where intuition departs from the normative model. Gigerenzer and Goldstein's 1996 Reasoning the fast and frugal way treats simple heuristics such as take-the-best as ecologically rational, often matching or beating information-hungry models when data are scarce. Both fill the empty slot Simon left, the second blade of his scissors of cognition and environment, but one counts the bounds as flaws and the other as fit.[2][3]

Unit

A mind's actual heuristic procedure for reaching a judgement

Recovery means

Not applicable

Measured by

Tested in the lab against a normative benchmark, or against real-world accuracy under limited information

The same premise carried up from the individual to the firm: if managers cannot optimize, the organizations they build are structures for coping with cognitive limits. Cyert and March's A behavioral theory of the firm, first published in 1963, made bounded rationality a foundation stone, describing firms that satisfice against aspiration levels, search only when performance falls short, and lean on standard operating procedures and organizational routines to economize on attention. The referent moves from a single mind to the division of decision labour across a coalition, where structure, rules, and slack stand in for the omniscience the firm does not have.[4]

Unit

A firm's routines and rules for deciding under limited attention

Recovery means

Not applicable

Measured by

Inferred from decision rules, standard procedures, aspiration-triggered search, and slack in the organization

The idea recast as an engineering standard for machines. Classical AI took perfect rationality, always choosing the best act, as the goal, but a real agent runs on finite hardware in real time and cannot compute it. Russell and Subramanian's 1995 Provably bounded-optimal agents proposed bounded optimality as the replacement: an agent program is bounded-optimal if it is the best program its machine can run in its task environment, so the optimization is over programs subject to computational constraints rather than over acts assuming unlimited computation. Here bounded rationality stops being an empirical description of people and becomes a formal criterion an artificial agent can be proven to meet.[5]

Unit

An agent program that is optimal for its machine and task environment

Recovery means

Not applicable

Measured by

Established by proof or construction that no better program runs on the given architecture in real time