Week 7 — Resource-Rational Minds: Is Cognition Just Sampling?

Week 7 of 8. If the mind is Bayesian, why is it so bad at probability?
This is the standing embarrassment of the field: rational models fit aggregate data beautifully, while the heuristics-and-biases literature documents people failing basic probability judgments constantly. Resource-rational analysis proposes a reconciliation. Suppose the brain approximates the correct posterior by drawing a small number of samples — sometimes just one. Then anchoring, base-rate neglect, and the variance in people's individual judgments stop being bugs and start being the predicted signature of a good algorithm running on a tight budget.
Agenda:

1. Recap and intros (10 min)
2. The rational-vs-biased tension, laid out honestly (15 min)
3. Sampling as approximate inference: MCMC, importance sampling, and "one and done" (30 min)
4. Vul et al.'s optimal-sample-size argument: why sampling twice is often worse than once (20 min)
5. Resource rationality as a research program, and its critics (25 min)
   Reference: Vul, Goodman, Griffiths & Tenenbaum (2014), "One and Done?"
   This is an online event. The video link appears on this page once you RSVP.