Somehow I feel like we’ve hashed over this before, but should “Bayesian” be normative? The word seems to have a rather idiosyncratic usage around these parts, and its choice as the name for Doing Rationality Right seems to come from a silly historical accident (Eliezer happened to have a silly conversation with someone about probability, and thus associated wrong thinking with “frequentist”).
The use may be somewhat idiosyncratic, but the point stands. Bayes’ rule is correct, provable from basic axioms of probability.
The “naive” scientific method (advocated in the clip) doesn’t account for probabilistic evidence. Even the slightly more sophisticated “statistical significance”/hypothesis testing method doesn’t do it right.
“Bayesian” is normative in the sense that if you think that plausibility assessments should follow Cox’s assumptions, then they ought to be isomorphic to probabilities and updated using Bayes’ Rule.
(Note that Cox’s argument requires an unintuitive mathematical assumption (as discussed by Halpern, references at the Wikipedia link). Frank J. Tipler of Omega Point infamy has a paper on the arXiv claiming to avoid this assumption.)
Somehow I feel like we’ve hashed over this before, but should “Bayesian” be normative? The word seems to have a rather idiosyncratic usage around these parts, and its choice as the name for Doing Rationality Right seems to come from a silly historical accident (Eliezer happened to have a silly conversation with someone about probability, and thus associated wrong thinking with “frequentist”).
The use may be somewhat idiosyncratic, but the point stands. Bayes’ rule is correct, provable from basic axioms of probability.
The “naive” scientific method (advocated in the clip) doesn’t account for probabilistic evidence. Even the slightly more sophisticated “statistical significance”/hypothesis testing method doesn’t do it right.
“Bayesian” is normative in the sense that if you think that plausibility assessments should follow Cox’s assumptions, then they ought to be isomorphic to probabilities and updated using Bayes’ Rule.
(Note that Cox’s argument requires an unintuitive mathematical assumption (as discussed by Halpern, references at the Wikipedia link). Frank J. Tipler of Omega Point infamy has a paper on the arXiv claiming to avoid this assumption.)