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

Eliezer_Yudkowsky comments on You Be the Jury: Survey on a Current Event - Less Wrong

31 Post author: komponisto 09 December 2009 04:25AM

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

Comments (260)

You are viewing a single comment's thread.

Comment author: Eliezer_Yudkowsky 09 December 2009 08:04:30PM *  18 points [-]

On a cursory reading of Wikipedia the obvious interpretation is that Knox and Sollecito are innocent and Guede is guilty. I didn't go through all the sites so I don't know if this would qualify as a litmus test, and assigning probabilities in this state of knowledge would be extra work.

EDIT: Read comments and am surprised at how many estimates of "Knox and Sollecito probably didn't do it" have probabilities in the range of 40% attached that they did. If it were a binary judgment or a confidence interval, then yes we should avoid extreme probabilities and widen intervals to compensate for known overconfidence biases. But in this case the hypothesis space of equally plausible possibilities is large, and some low-probability symbols were used to write the message (multi-person rape-murder vs. single-person rape-murder, female rape-murder vs. male rape-murder). It may not always be easy to unravel crime scenes (though this one sounds pretty straightforward) but to focus on Knox or Sollecito seems like privileging the hypothesis.

Unless there's major prosecutorial evidence not in Wikipedia, then this seems like a case of paying too much attention to other people's opinions (the jury, in a case where we have further information that the verdict gained media attention as possibly inaccurate), and I would suggest that anyone who gave a probability higher than .15 be more arrogant in the future.

If, of course, I just haven't done enough reading, then ignore the above.

Comment author: Eliezer_Yudkowsky 20 October 2011 07:42:09PM 9 points [-]

I would suggest that anyone who gave a probability higher than .15 be more arrogant in the future.

This does not mean my assigned probability was 15%. It means, "Even after accounting for fudge factors and people using different numbers to express similar emotional degrees of certainty, if you gave a number higher than FIFTEEN PERCENT it means you've got a MAJOR problem with paying WAY too much attention to really lousy evidence, what other people think, and the authority of idiots."

Comment author: lessdazed 21 October 2011 03:05:10AM 1 point [-]

The most important lesson here has nothing to do with innocence and guilt, but with people's confidently paraphrasing the opinions of others.

How much the more so when those opinions aren't saved as text, both to give the interpreters another chance to parse properly and to introduce the possibility that they could be embarrassed if later shown wrong.

Comment author: imaxwell 10 December 2009 07:32:11PM *  2 points [-]

The main reason my estimate is so high is because

  • I know my information came from two heavily biased sites, and

  • I found the "innocent" site a lot easier to follow and therefore paid more attention to it, so I know my information is particularly biased in that direction.

That said, I did consider a more-arrogant probability of 0.25 or so. My caution in this case isn't on general principle, but because I have something of an old history of embracing cause celebre cases like this only to decide on further reading that the person I'm defending is guilty as hell.

Comment author: saliency 10 December 2009 08:40:06PM 1 point [-]

I agree with Eliezer but like Maxwell's point about assigning extra probability to Knox and Sollecito because the guilty argument was so poorly formated. "She was convicted but I don't get why, perhaps I don't understand this."

That said I think 15% or less more then accounts for this uncertainty. I gave Knox a 6% probability.

Side note, I am surprised that more people are not assigning probability to the chance that none of them did it.