Bayes theorem, the geometry of changing beliefs

This segment introduces Bayes' theorem through the classic 'librarian vs. farmer' problem studied by Kahneman and Tversky. It demonstrates how humans often ignore base rates (priors) in favor of representativeness (likelihoods). By constructing a representative sample grid, it calculates the posterior probability geometrically, showing that even strong evidence must be weighed against prior odds. Finally, it derives the algebraic formula from this geometric setup. This segment explores how geometric representations and natural frequencies can make Bayesian probability more intuitive. It begins by visualizing the joint probability P(H)P(E|H) as the area of a rectangle within a larger sample space. The discussion then shifts to cognitive psychology, contrasting abstract percentages with concrete 'out of 100' counts to explain why people often fall prey to the conjunction fallacy, as demonstrated by the classic Linda problem. By reframing probabilities as proportions of people or areas, the video demystifies Bayes' theorem, showing it simply asks for the fraction of evidence-supporting cases where the hypothesis is also true. Finally, it addresses criticisms regarding ambiguous priors in personality-based problems, concluding that while context influences initial beliefs, evidence should serve to update rather than rigidly determine our convictions.

Before you watch

• Basic concept of conditional probability
• Understanding of fractions and percentages
• Basic understanding of conditional probability notation $P(A|B)$
• Familiarity with set operations (intersection, union, subsets)