Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

alexflint comments on Parapsychology: the control group for science - Less Wrong

62 Post author: AllanCrossman 05 December 2009 10:50PM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (185)

You are viewing a single comment's thread. Show more comments above.

Comment author: alexflint 08 December 2009 10:16:40AM 15 points [-]

One in a bajilion? Guys, the numbers matter. 10^-9 is very different from 10^-12, which is very different from 10^-15. If we start talking about some arbitrarily low number like "one-in-a-bajillion" against which no amount of evidence could change our mind, then we're really just saying "zero" but not admitting to ourselves that we're doing so.

Other than that, I agree with Yvain and have found this to be perhaps the most belief-changing so far on LW!

Comment author: Eliezer_Yudkowsky 11 December 2009 02:13:27AM 14 points [-]

It takes only 332 pieces of evidence with likelihood ratios of 2:1 to promote to 1:1 odds a hypothesis with prior odds of 1:googol, that is 10^-100, which would be the appropriate prior odds of something you could describe with around 70 symbols from a 26-letter equiprobable alphabet.

"A bajillion to one" are odds that Bayesian updating can overcome surprisingly quickly - it isn't anything remotely like "no amount of evidence can change my mind". Now odds of one to a googolplex - that might as well be zero, relative to the amount of evidence you could acquire over a human lifetime. But the prior probability of any possibility you can describe over a human lifetime should be much higher than that.

Comment author: roystgnr 09 February 2012 06:52:49PM 7 points [-]

A nitpick: it takes 332 pieces of all mutually independent evidence to perform that level of update.

More confusing, for these purposes the independence level of the evidence depends on what hypotheses you're trying to distinguish with it. E.g. if you're trying to distinguish between "that subject has ESP powers" and "that experiment was random luck" then 332 repetitions of the same experiment will do. If you're trying to distinguish between "that subject has ESP powers" and "that experimenter's facial expressions differ based on what cards he was looking at", then you can't just repeat it; you've got to devise new and different experiments.