Previous Post Revised

post by komponisto · 2009-12-14T06:56:50.528Z · score: 12 (13 votes) · LW · GW · Legacy · 26 comments

Followup to: The Amanda Knox Test: How an Hour on the Internet Beats a Year in the Courtroom

See also: The Importance of Saying "Oops"

I'm posting this to call attention to the fact that I've now reconsidered the highly confident probability estimates in my post from yesterday on the Knox/Sollecito case. I haven't retracted my arguments; I just now think the level of confidence in them that I specified was too high. I've added the following paragraph to the concluding section:

[EDIT: After reading comments on this post, I have done some updating of my own. I now think I failed to adequately consider the possibility of my own overconfidence. This was pretty stupid of me, since it meant that the focus was taken away from the actual arguments in this post, and basically toward the issue of whether 0.001 can possibly be a rational estimate for anything you read about on the Internet. The qualitative reasoning of this post, of course, stands. Also, the focus of my accusations of irrationality was not primarily the LW community as reflected in my previous post; I actually think we did a pretty good job of coming to the right conclusion given the information provided -- and as others have noted, the levelheadedness with which we did so was impressive.]

While object-level comments on the case and on my reasoning about it should probably continue to be confined to that thread, I'd be interested in hearing in comments here what people think about the following:

26 comments

Comments sorted by top scores.

comment by SilasBarta · 2009-12-14T16:50:46.426Z · score: 6 (6 votes) · LW(p) · GW(p)

I never bothered to ask: where are the sites presenting the pro-Guede innocence case?

comment by AnnaGilmour · 2009-12-15T23:52:11.524Z · score: 1 (1 votes) · LW(p) · GW(p)

The only way that Guede is innocent is if Meredith and he had consentual relational activity and if it could be also possible that the real perpetrator left no (or little/untested for) DNA while at the same time allowing for Guede to leave a lot of DNA. To leave it there is even highly suspect, as there may be further incriminating facts that I will think of after posting this comment. Also, his journal indicates that he was not outside the event - he did not call the police, for example. Also, at the far stretch of the possibility that someone else did it, he admits blatently in an unprovoked journal, that he left, letting her die.

comment by SilasBarta · 2009-12-16T04:13:45.385Z · score: 3 (3 votes) · LW(p) · GW(p)

Hold on: I was asking why there aren't any pro-Guede sites, not for reasons Guede could be innocent.

You don't need actual truth of a proposition to find people willing to be rabid defenders of it, remember? :-)

comment by AnnaGilmour · 2009-12-15T23:54:55.058Z · score: 0 (0 votes) · LW(p) · GW(p)

From there, after establishing that Guede is guilty, you have to establish that there was complicity between Rudy and Amanda and Raffaele, of which there is no evidence.

comment by wedrifid · 2009-12-14T09:11:07.141Z · score: 5 (5 votes) · LW(p) · GW(p)

How much of a distraction did you find my extremely confident probabilities to be from the substance of my arguments?

Very little. It isn't often that such estimates can be considered quantitative anyway. I usually interpret them as 'qualitative with pretty numbers'.

How much did those confident estimates make it seem like I was disagreeing, rather than agreeing, with the LW survey consensus? (It seemed to me that I had provoked people into trumpeting pro-guilt arguments more than they otherwise would have if I had initally given more "reasonable" numbers.)

The distinction I would make between the original and the edit is evaluating vs commentating. This was primed somewhat by an earlier comment and will naturally invoke more objections. It also brings into play the subjectivity and objectivity of probability estimates, as I discussed.

To what sorts of propositions, if any, do you yourself assign probabilities on the order of 0.999 or 0.001?

I usually only assign such probabilities when I am considering myself, literally, as a statistic. There are other matters of fact that I could assign probabilities of that order to but I don't particularly trust my own judgement at that level of certainty so don't usually bother with precise numbers. I certainly wouldn't assign them to any court cases over which there is any controversy. I would say "except if I was the alleged protagonist" but, as matt pointed out, I probably can't assign that probability even then except for the obvious rhetorical purposes.

comment by Blueberry · 2009-12-14T21:50:03.338Z · score: 3 (3 votes) · LW(p) · GW(p)

How much of a distraction did you find my extremely confident probabilities to be from the substance of my arguments?

A big distraction. They made me strongly emotionally opposed to anything you had to say, and it was difficult to overcome my reaction.

How much did those confident estimates make it seem like I was disagreeing, rather than agreeing, with the LW survey consensus? (It seemed to me that I had provoked people into trumpeting pro-guilt arguments more than they otherwise would have if I had initally given more "reasonable" numbers.)

Very much. I definitely felt provoked into the pro-guilt position. But then again, I'm very stubborn and oppositional like that.

To what sorts of propositions, if any, do you yourself assign probabilities on the order of 0.999 or 0.001?

I would avoid assigning those kind of probabilities to political propositions, anything heavily polarized, or anything related to current events. In such cases it's very tempting to take one side or the other instead of looking at both objectively.

comment by AnnaGilmour · 2009-12-14T22:03:50.749Z · score: 2 (4 votes) · LW(p) · GW(p)

Blueberry said: "A big distraction. They made me strongly emotionally opposed to anything you had to say, and it was difficult to overcome my reaction."

Isn't that your own fault? You will lose out if you respond emotionally rather than rationally in such circumstances.

comment by Blueberry · 2009-12-14T22:18:09.638Z · score: 3 (3 votes) · LW(p) · GW(p)

Isn't that your own fault? You will lose out if you respond emotionally rather than rationally in such circumstances.

Yes. Which is why I tried to overcome my immediate impulse to be stubborn. But at the same time, I should still try not to trigger people's strong emotional responses when I'm trying to communicate. It's a two-way effort.

comment by LauraABJ · 2009-12-14T19:15:17.523Z · score: 3 (3 votes) · LW(p) · GW(p)

1) A big distraction. Since you were arguing that there had been a serious failure in rationality, giving an irrationally low estimate made me doubt your thought process.

2) Very much- it did seem like you were incredulous that even the less wrong community could possibly come up with such high estimates. Though to be fair, after reading your post, I downgraded my own estimate from 20% to 2%, so the 32% average estimate was still way too high.

3) The possibility that my friend's Rosacea is caused by excessive aspartame ingestion.

comment by wedrifid · 2009-12-14T20:15:56.027Z · score: 0 (0 votes) · LW(p) · GW(p)

3) The possibility that my friend's Rosacea is caused by excessive aspartame ingestion.

I'm assuming you mean the 0.001 option. What makes you so confident? The prior for reasonably active substances with no know relevance, specific trials or perhaps knowing she vigilantly avoids the stuff? I suppose that final category includes relevant legislative interventions, depending on her location. The vile stuff is not banned here yet.

comment by LauraABJ · 2009-12-14T21:35:23.575Z · score: 3 (3 votes) · LW(p) · GW(p)

1) Current meta-analysis of studies on the dangers of aspartame show NO dangerous side effects even at upper bounds of current consumption except in PKU patients, and there are many reasons a scientist would want to prove any danger given the ubiquity of the substance.

2) Aspartame = aspartic acid + phenylalanine + 10% methanol (this is the part that is usually focused on in studies, since it breaks down into formaldehyde). The first two are amino acids found in much higher levels in many foods. There are people with PKU who can't metabolize phenylalanine with resulting neurotoxicity since phenylalanine is normally converted to tyrosine which is converted to dopamine, and having excess phenyalanine competes with the normal pathways. Suffice it to say, she does not have PKU.

3) There is 50 times more methanol in a glass of wine than a can of diet soda, so my friend's obsession with the fact that all the diet soda she consumed in the 90s contained methanol is bogus- in terms of methanol, she should be more concerned about her connoisseurship of wine and her use of mouthwash than her consumption of diet soda, no matter how excessive it might seem to her.

4) Allergies exist to nearly everything-- so it's not impossible that she was allergic to aspartame, but she hasn't had it in 5 years and has never documented a reaction to diet soda, rather she thinks it somehow caused 'toxins' to accumulate in her which caused her rosacea. Also, if her rosacea is caused by an allergy, why focus on aspartame? It could be anything in her environment or diet. Aspartame is just one of millions of substances she comes into contact with.

5) She said herself that aspartame was the only 'sin' she ever indulged in and thus must be the cause of her biblical level deformity... Yes, something being sweet with no calories is so deliciously sinful... The whole theory is psychologically based on false assumptions.

6) The only reason my p is as high as .001 is a single case study in which a patient had a hypersensitivity reaction to aspartame resulting in allergic dermatitis (which responded like normal allergy to steroids (friend's rosacea did not) and improved after she stopped use of aspartame). This is one case study. ONE.

comment by wedrifid · 2009-12-15T03:27:24.292Z · score: 2 (2 votes) · LW(p) · GW(p)

1) Current meta-analysis of studies on the dangers of aspartame show NO dangerous side effects even at upper bounds of current consumption except in PKU patients, and there are many reasons a scientist would want to prove any danger given the ubiquity of the substance.

Thankyou, I wasn't aware of such an analysis.

2) Aspartame = aspartic acid + phenylalanine + 10% methanol (this is the part that is usually focused on in studies, since it breaks down into formaldehyde). The first two are amino acids found in much higher levels in many foods. There are people with PKU who can't metabolize phenylalanine with resulting neurotoxicity since phenylalanine is normally converted to tyrosine which is converted to dopamine, and having excess phenyalanine competes with the normal pathways. Suffice it to say, she does not have PKU.

Ahh, I was aware of the cognitive effects often attributed to aspartame and that mechanism does explain exactly the kind of symptoms described. I believe I actually have some phenylalanine in my supplement stockpile so if I (blindly) tested myself on some of that I may well be able to establish whether aspartame would negatively affect me.

but she hasn't had it in 5 years and has never documented a reaction to diet soda, rather she thinks it somehow caused 'toxins' to accumulate in her which caused her rosacea. 5) She said herself that aspartame was the only 'sin' she ever indulged in and thus must be the cause of her biblical level deformity... Yes, something being sweet with no calories is so deliciously sinful... The whole theory is psychologically based on false assumptions.

That is bizarre reasoning but I suppose not too uncommon.

6) The only reason my p is as high as .001 is a single case study in which a patient had a hypersensitivity reaction to aspartame resulting in allergic dermatitis (which responded like normal allergy to steroids (friend's rosacea did not) and improved after she stopped use of aspartame). This is one case study. ONE.

0.001 was approximately my prior. With your information I think I updated down then back up slightly to 0.001 again.

comment by Jack · 2009-12-14T20:26:52.508Z · score: 1 (1 votes) · LW(p) · GW(p)
  • How much of a distraction did you find my extremely confident probabilities to be from the substance of my arguments?

It was somewhat a distraction as it lead to a comment basically suggesting that your probabilities were too high. On the other hand I didn't keep me from revising my estimated of guilt downwards after reading your arguments.

  • How much did those confident estimates make it seem like I was disagreeing, rather than agreeing, with the LW survey consensus? (It seemed to me that I had provoked people into trumpeting pro-guilt arguments more than they otherwise would have if I had initally given more "reasonable" numbers.)

A good deal. But you also quoted me as an example of bad reasoning about the case so I'm not sure you weren't disagreeing with me even though my estimation was in line with the LW consensus.

  • To what sorts of propositions, if any, do you yourself assign probabilities on the order of 0.999 or 0.001?

In the Less Wrong survey a while back I recall assigning probabilities this low and lower to the truth of the Bible, the existence of God and the existence of ontological fundamental mental entities. I'm pretty sure a lot of other people did the same.

comment by wedrifid · 2009-12-14T20:46:01.142Z · score: 0 (0 votes) · LW(p) · GW(p)

the existence of ontological fundamental mental entities

Does that mean 'soul'?

comment by Jack · 2009-12-14T21:50:52.032Z · score: 1 (1 votes) · LW(p) · GW(p)

It is a terrible turn of phrase, for sure. Souls are commonly conceived as one type of ontologically basic mental entity. God is another. Basically I take the claim "ontologically basic mental entities exist" to mean something like "there are things that think that do not have extension in space".

Edited: So that the words ontologically, basic and mental aren't repeated vertically three times right next to each other.

comment by Blueberry · 2009-12-14T21:59:43.663Z · score: 0 (0 votes) · LW(p) · GW(p)

So the number three doesn't exist? Freedom? Relationships? Holes?

comment by Jack · 2009-12-14T22:33:48.748Z · score: 1 (1 votes) · LW(p) · GW(p)

This doesn't strike me as a serious question but it also isn't funny enough to definitely be a joke. So I'm not sure what you're doing.

My answer are exists, depends on what you mean, exists and exists.

Edit- with low confidence because ontology isn't something I'm that interested in and I have not thought a lot about these questions. The Holes piece is a classic one.

comment by Blueberry · 2009-12-14T22:59:45.457Z · score: 0 (0 votes) · LW(p) · GW(p)

Yeah, I love the Holes piece. I thought you were originally denying anything non-physical existed, but on re-reading it seems like you're just denying dualism. Sorry if I misread.

comment by Psychohistorian · 2009-12-23T19:32:23.579Z · score: 0 (0 votes) · LW(p) · GW(p)

Denying "the existence of ontological fundamental mental entities" is probably more efficiently expressed by, "I don't believe there are any minds without physical brains." This rules out God, ghosts, and pretty much all of the supernatural, since things like homeopathy are contingent on water being able to "remember" or "know" things.

comment by Paul Crowley (ciphergoth) · 2009-12-14T09:16:50.373Z · score: 1 (1 votes) · LW(p) · GW(p)

What general advice for yourself do you have as a result of this error? Presumably it's more than just "don't be so overconfident", since I'm guessing you asked yourself at the time whether you were being overconfident and concluded you weren't.

comment by komponisto · 2009-12-14T17:51:33.777Z · score: 1 (1 votes) · LW(p) · GW(p)

Probably something along the lines of: don't say anything more than you need to make your point.

My "gut feeling" is still somewhere in the region of what I originally said -- that there's roughly a snowball's chance in hell that Knox and Sollecito commited murder -- but my post is a stinging indictment of the verdict even if it only argues for a probability of 0.1.

comment by Eliezer Yudkowsky (Eliezer_Yudkowsky) · 2009-12-14T16:18:45.817Z · score: 0 (2 votes) · LW(p) · GW(p)

I noticed, but wasn't distracted.

It also seemed to me that assigning 0.001 to Knox's guilt was safer than assigning 0.999 to Guede's guilt, yet people who assigned extreme estimates wanted to assign equally extreme estimates to both. I'm not confident this is an error, though, because the case for Guede's guilt looks strong and it might be that in this state of evidence fewer than 1 in 1,000 people are innocent. On the other hand, I recall a strong tendency for people to say 0.10 and 0.90, 0.05 and 0.95, 0.01 and 0.99, or 0.001 and 0.999, and I don't see how that could happen naturally.

It looks to me like the clash between the concepts of "overconfidence and calibration" versus "privileging the hypothesis" may also be behind my horrible LHC inconsistency.

comment by komponisto · 2009-12-14T17:48:02.757Z · score: 0 (0 votes) · LW(p) · GW(p)

It also seemed to me that assigning 0.001 to Knox's guilt was safer than assigning 0.999 to Guede's guilt,

That's interesting. I'm wondering if you could elaborate on why you think that's so, since I would have guessed the opposite.

comment by Eliezer Yudkowsky (Eliezer_Yudkowsky) · 2009-12-14T18:08:56.243Z · score: 1 (1 votes) · LW(p) · GW(p)

It's a question of whether errors in the story you know make the probability more extreme or less extreme. Knox seems like a bystander, pretty much, so the "privileging the hypothesis" concept applies to her. Guede seems pretty definitely involved, but the probability of error or misunderstanding the story might not be so low as 1 in 1000, and errors in his story make the probability less extreme.

It's a question of how you try to apply compensation for overconfidence. With Guede, you apply compensation by lowering the probability of his guilt. But you can't just take everyone in the world and say that to compensate for overconfidence you're going to assign a non-extremely-low probability that they murdered Meredith.

comment by Paul Crowley (ciphergoth) · 2009-12-14T23:11:57.051Z · score: 3 (3 votes) · LW(p) · GW(p)

You're saying that sometimes compensating for overconfidence means moving a probability further away from 50%? That it somes means moving a probability estimate closer to some sort of "base rate"? Interesting and worth talking about more, I think. For one thing it gets you right into the "reference class tennis" you've talked about elsewhere - which in itself deserves further discussion.

comment by Eliezer Yudkowsky (Eliezer_Yudkowsky) · 2009-12-15T00:41:50.967Z · score: 1 (1 votes) · LW(p) · GW(p)

For one thing it gets you right into the "reference class tennis" you've talked about elsewhere

Yup.