Rationally Speaking Podcast Rationally Speaking #143 - Scott Aaronson on "The theorem that proves rationalists can't disagree"
12 snips
Sep 20, 2015 
Scott Aaronson, MIT computer science professor and quantum computing author, joins to unpack Aumann's Agreement Theorem. He explains why Bayesian agents should converge after sharing information. They probe practical loopholes, when priors differ, what counts as an epistemic peer, and a case study on disagreement over quantum computing.
AI Snips
Chapters
Books
Transcript
Episode notes
Aumann's Theorem Means No Rational Persistent Disagreement
- Aumann's theorem: if two Bayesians share common priors and their opinions become common knowledge, they must have identical probabilities on the question discussed.
- Scott Aaronson explains this makes persistent, rational disagreement impossible under strong assumptions, summarized as "cannot agree to disagree."
Bayesian Conversations Follow A Martingale Pattern
- Ideal Bayesian conversation has the martingale property: you immediately update on another's message rather than bargaining, so you can't predict which direction they'll move you next.
- Scott notes that if you could predict the next move, you should've already updated to that position.
Mathematicians Rapidly Switch Sides In Arguments
- Scott recalls mathematicians arguing by a blackboard who switch sides within minutes, showing rapid opinion reversals typical of ideal updating.
- He contrasts this with politics where such quick side-switching is virtually unheard of.
