Discussion article for the meetup : Virtual Evidence

WHEN: 05 February 2014 07:00:00PM (-0800)

WHERE: 10850 West Pico Blvd, Los Angeles, CA 90064, USA

How to get in: Go to the Westside Tavern in the upstairs Wine Bar (all ages welcome), located inside the Westside Pavillion on the second floor, right by the movie theaters. The entrance sign says "Lounge". Parking is free for 3 hours.

Discussion: If we ask an expert to judge, purely based on their observations and not on prior expectations, whether a rug is a genuine antique, and we get the answer "Based only on what I could observe, I'd say 5 to 1 it's genuine", how should we revise our beliefs? We could assume that 5/6 is the posterior probability which the expert arrives at, but this is in bad faith with respect to our request to ignore the prior and infer based on observation alone. In "Probabilistic Reasoning in Intelligent Systems", Judea Pearl discusses a solution to this problem called "virtual evidence". I'll approach this issue from a perspective of how we should treat opinions offered to us in everyday conversation, and relate it to the general idea of approximate Bayesian updates. The discussion may be more toward the mathematical side, or more toward the practical side, depending on the audience response.

No prior knowledge of or exposure to Less Wrong is necessary; this will be generally accessible.

Discussion article for the meetup : Virtual Evidence