Week 12: The blind spot

Nothing new. The account from eleven weeks, aimed at the person holding it, and the week 1 question in a form no card will find.

Part IV — The human case, and its limits10 May 2027Open the slidesWeek 12 seminarSchedule

Dunning-Kruger first, carefully: Krueger and Mueller (2002) showed the classic curve is largely reproducible from regression to the mean plus a better-than-average effect, and the exchange that followed is unresolved. The most famous claim about not knowing what you don’t know is itself not fully known.

Then the formal shape. Underdetermination via Duhem and Quine, with Laudan and Leplin against it because teaching the strong version unopposed would break this course’s own rules. Goodman’s riddle. Agrippa’s trilemma. And the No Free Lunch theorems, which make “there is no assumption-free learner” a proof rather than a sentiment.

I finish by insisting on the distinction the whole course depends on. Fallibilism is not relativism. Models are ranked by track record, Tetlock has twenty years of scored forecasts to show it, and anyone leaving here believing nothing can be known has the argument exactly backwards.