# Yvain comments on The Amanda Knox Test: How an Hour on the Internet Beats a Year in the Courtroom - Less Wrong

42 13 December 2009 04:16AM

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

Sort By: Best

You are viewing a single comment's thread.

Comment author: 13 December 2009 02:36:02PM 9 points [-]

No different from the prior, which is dominated by the probability that someone in whatever reference class you would have put Amanda into on January 1, 2007 would commit murder within twelve months. Something on the order of 0.001 at most.

Out of one thousand criminal trials in which the Less Wrong conventional wisdom gave the defendant a 35% chance of being guilty, you would expect to be able to correctly determine guilt nine hundred ninety nine times?

Comment author: 13 December 2009 03:48:18PM 2 points [-]

Out of one thousand criminal trials in which the Less Wrong conventional wisdom gave the defendant a 35% chance of being guilty, you would expect to be able to correctly determine guilt nine hundred ninety nine times?

Maybe I'm missing something, but I think you read that wrong.

komponisto said the evidence should not cause anyone to change the prior probability much. Surely, for people in AK's reference class, the per-year probability of committing a 3-party sex killing is less than 0.001?

I think komponisto quite correctly described the effect of privileging the hypothesis, which might be what caused the LW community to be so much off from his estimate. Everybody seemed to be going backward from assuming AK's guilt at 50-50, whereas komponisto went forward from the background probability.

Comment author: 13 December 2009 05:56:35PM 1 point [-]

Everybody seemed to be going backward from assuming AK's guilt at 50-50, whereas komponisto went forward from the background probability.

I think I can see why. komponisto pretended to be a juror following the "innocent unless proven otherwise" mantra and updating on the evidence presented in the court. We, on the other hand, did what komponisto challenged us to do: figure out the answer to his riddle using the two websites he gave us. This being a riddle, not a court, we had no reason to favour one hypothesis over the other, hence the 50-50.

That being said, I did favour one hypothesis over the other (my stated priors were 75/75/25) because at the moment I paused to write down an approximation of my current beliefs, I had already updated on the evidence presented by komponisto himself in his post, namely, that there was a trial against AK and RS.

Maybe the reason why many of us gave so much importance to the fact that those particular individuals were on trial for murder was because it was our very first piece of information; and I don't think it's right for rationalists to do that.

Comment author: 13 December 2009 09:26:49PM *  0 points [-]

komponisto should not be going forward from the background probabilities because he isn't an experienced investigator with access to the crime scene. he's just a guy reading about evidence on the internet. a more reasonable prior for him to start with is, ''how often are people convicted of murder when they did not in fact commit a murder?'' (there are actual #s for this, too)

when juries sit around thinking, ''is this person guilty or not?'' they assume the investigators working on the case are competent. they assume, quite rightly, that there must be a damn good reason why reasonable investigators couldnt quickly dismiss a hypothesis with such an insanely low prior probability. lesswrong.com readers should do likewise.