Reading focus
Syllabus: How Do We Reason and Solve Problems?
Compare the model's information-gain objective with the logical rule-checking interpretation of the selection task. Trace the role of assumed property frequencies.
Why you're reading it. Evaluate a probabilistic explanation against both its data fit and its assumptions about the task.
A question to consider. What observation would distinguish people responding to expected information from people simply choosing cards that repeat the words in the rule?
Why this reading
The same card choice can be evaluated as a test of a logical rule or as a way to learn about an uncertain environment. Oaksford and Chater make that difference explicit through a probabilistic model. The result is an opportunity to inspect how a proposed task objective changes an explanation of performance.
Before you read
In the selection task, a conditional rule takes the form 'if p, then q'; p and q stand for properties shown on cards. A p case paired with not-q violates that rule. Bayesian inference updates relative support for hypotheses using evidence. Expected information gain evaluates possible observations before their outcomes are known. A rational analysis specifies an objective and environmental assumptions; it need not describe the steps a person consciously performs.
Read for
- State the objective and competing hypotheses in the model. Keep deciding which observation is informative separate from checking the validity of a deduction.
- Examine how assumptions about the rarity of p and q change which cards provide information. Use the model's figures to connect parameter changes with predictions.
- Compare the model's account of existing data with evidence that would independently establish participants' assumptions. Ask whether a good fit identifies a mental procedure or only a useful computational description.
Consider afterwards
- Imagine two otherwise identical card tasks in which the relevant properties are rare in one population and common in the other. What change in selection would support an information-gain account, and what should you check about participants' understanding first?
Editorial context
This influential rational analysis is a conditional explanation, not proof that all card choices are optimal or that logical standards are irrelevant. Its predictions depend on the hypothesized sampling situation, goals, and probabilities. The paper itself distinguishes normative and adaptive rationality. Comparing it with other accounts requires specifying the same task interpretation; changing the objective can otherwise disguise a disagreement about what participants are doing.