Oaksford & Chater's 'rational analysis' rehabilitates human reasoning as Bayesian, not logical — a rival lineage to Gigerenzer's ecological rationality

Core claims

1. Oaksford & Chater relocate the definition of rationality itself — from logic to uncertainty. Per their own précis: "Bayesian Rationality argues that rationality is defined instead by the ability to reason about uncertainty. Although people are typically poor at numerical reasoning about probability, human thought is sensitive to subtle patterns of qualitative Bayesian, probabilistic reasoning." This directly displaces the older, Piagetian view that logical deduction is the end-point and standard of rational thought. (Tier 1 — authors' own précis of their own book, Behavioral and Brain Sciences, 2009.)

2. The Wason selection task — psychology's classic proof that humans reason badly — gets recast as evidence humans reason well, just not logically. "Data from conditional reasoning, Wason's selection task, and syllogistic inference are captured by recasting these problems probabilistically. The probabilistic approach makes a variety of novel predictions which have been experimentally confirmed." (Tier 1, same source.) The task was long read as a demonstration of systematic human error against formal logic; Oaksford & Chater's earlier "optimal data selection" analysis treats the same choices as expected-information-gain maximization.

Why this was hop-worthy

The seed asked whether Gigerenzer's ecological rationality has been applied to LLMs — yes (ERMI, above). But the ERMI paper's own related-work framing lists "Rational Analysis (Oaksford & Chater)" as a separate lineage it synthesizes with ecological rationality: Gigerenzer rehabilitates human reasoning via simple heuristics matched to environmental structure; Oaksford & Chater rehabilitate it via optimal Bayesian inference under uncertainty. Two different 20th/21st-century schools converged on the same move — "humans aren't irrational, the yardstick was wrong" — from opposite mechanistic assumptions (heuristic vs. optimal-inference), and both are now being folded into the same LLM-meta-learning framework.

Further leads

• Oaksford & Chater's original "optimal data selection" analysis of Wason's task (expected information gain / Shannon entropy reduction) — the specific math, unopened here.
• Whether ecological rationality and rational analysis are treated as compatible or contested in the wider cognitive-science literature (the ERMI paper synthesizes them; not clear this is consensus).