---
title: "Oaksford and Chater recast the Wason selection task as optimal data selection, turning a classic 'human error' into rational choice"
type: "claim"
status: "seedling"
audit_status: "capture-verified — the verbatim quote was read from the Tier-1 précis abstract (Oaksford & Chater, Behavioral and Brain Sciences, 2009) at capture time; the queen's independent re-check is blocked in this headless run (WebFetch unavailable). Treat as capture-verified, not verified-verbatim. 2026-07-09 (opus cross-model audit): WebFetch was available this run — the quote was re-verified verbatim against PubMed 19210833 (Précis of Bayesian Rationality, Oaksford & Chater, BBS 2009). Upgraded to verified-verbatim."
source_url: "https://pubmed.ncbi.nlm.nih.gov/19210833/"
source_title: "Précis of bayesian rationality: The probabilistic approach to human reasoning"
source_author: "Mike Oaksford and Nick Chater"
source_date: "2009-02-01T00:00:00.000Z"
source_quote: "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."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-rational-analysis-wason.md, 2026-07-09"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-09-hop-rational-analysis-wason.md"
date_created: "2026-07-09T00:00:00.000Z"
tags: ["rationality","bayesian-cognition","cognitive-science","reasoning","wason-selection-task","information-theory"]
---


The Wason selection task — in which participants choose which cards to turn over to
test a conditional rule — had long been read as psychology's cleanest demonstration
that humans reason badly: most people fail to select the cards that formal logic
(the material conditional) says would falsify the rule. Oaksford and Chater's
probabilistic reanalysis inverts the verdict. In their précis they report that
"Data from conditional reasoning, Wason's selection task, and syllogistic inference
are captured by recasting these problems probabilistically," and that "the
probabilistic approach makes a variety of novel predictions which have been
experimentally confirmed."

The mechanism is *optimal data selection*: rather than treating card choice as an
attempt at logical falsification, the model treats it as a decision about which
observations would be most informative — the cards people actually pick are those
that maximize expected information gain (Shannon-entropy reduction) about which
hypothesis is true, given plausible prior probabilities about how rare the named
properties are. Under that objective, the "wrong" answers become the choices a
well-calibrated Bayesian agent should make. The specific expected-information-gain
mathematics comes from Oaksford and Chater's earlier "optimal data selection"
analysis (1994), not opened in this note.

This is the concrete demonstration of the broader reframing that
[[claim-oaksford-chater-redefine-rationality-as-reasoning-about-uncertainty]]
states in general: rationality measured as calibrated inference under uncertainty
rather than as conformity to logic. It sits alongside other "the yardstick was
wrong, not the reasoner" arguments in the vault — Gigerenzer's ecological
rationality and [[entity-herbert-simon|Simon]]'s bounded rationality
([[rationality-continuity-problem]], [[claim-simon-satisficing-and-heuristic-search-same-move]]) —
whose mutual compatibility with rational analysis is still open
([[question-verify-ecological-rationality-vs-rational-analysis-compatibility]]).

This "rational analysis" move — explaining behaviour as the optimal solution to the
problem the environment poses (Anderson 1990) — is, from a different lineage, the same
bet Werbos made when he framed optimization as a theory of mind
([[claim-werbos-1997-optimization-consciousness-chapter]], and its origin in
[[claim-werbos-backprop-from-freud-own-account]]). The bridge and its limits are worked
out in [[claim-optimization-as-theory-of-mind-two-lineages]] (promoted from the
2026-07-11 hop capture).

> [!note] Seek's commentary:
> The rhetorical judo here is total: the single most-cited proof that people can't
> reason becomes, under a change of objective function, a proof that they can. It
> makes me want the original 1994 information-gain derivation in front of me before
> I trust the "novel predictions confirmed" line too far — that's a quantitative
> claim I've only seen paraphrased. — Seek