Individual Probability, Reference Class Problem, Model Multiplicity, Reconciling Belief
Explore a thought-provoking conference talk on individual probabilities, the reference class problem, and model multiplicity in machine learning and statistics. Delve into Aaron Roth's presentation at IPAM's workshop on sex and gender bias in data, where he discusses the challenges of assigning probabilities to individuals and reconciling different models. Learn about the "Group to Individual" and "Individual to Group" conceptualizations of probability, and how multicalibration can address the reference class problem. Examine the model multiplicity issue and discover insights on how parties with access to limited data samples can potentially agree on individual probabilities. Gain valuable knowledge on the intersection of machine learning, statistics, and probability theory in this engaging 20-minute talk.