Scott (squid314) wrote,

I don't believe in "believing in" things

[Epistemic Status | Polemical format. If I wanted to be scrupulously fair, I could admit that processing the statistics a different way - for example taking medians rather than means - makes things less *perfect*, though I still think the numbers support my point. Also, my reasons behind my answer to the final question are more complex than I admit, and may include "being dumb".]

RationalWiki has a discussion thread up: "Does Yvain believe in cryonics?" ("Yvain" is one of my Internet handles). Since no one bothered to ask me or even tell me about the thread I feel okay making anyone who wants the answer first read through a lengthy but important deconstruction of the question.

Many LWers are signed up for or considering cryonics, or think it's a good idea. Almost no one in the general population is signed up for or considering cryonics, and they mostly think it's stupid and crazy. This might lead some very silly people to think that reading Less Wrong convinces people to "believe in cryonics" more than the average member of the general population. This is absolutely wrong and I can prove it.

I'd like to be able to prove it by comparing LWers' degree of belief in cryonics with that of a matched control population. Unfortunately I don't have the data for that. I have great numbers on LWers' degree of belief - the Less Wrong Survey, which asked participants what percent chance they thought cryonics had of working, but I don't know of any survey on anyone else's belief in cryonics, let alone an identical question asked to a matched population.

But all is not lost! We can try to extract the data we want from the Less Wrong Survey (if you have a statistics package, you're welcome to download the data, check for yourself, and keep me honest). The trick will be to compare extreme newcomers - people who have just stumbled across the community and not been influenced by any of its ideas - to veterans who have been fully "indoctrinated". If Less Wrong convinces people to believe in cryonics, we should find that the first group has relatively low belief in cryonics, and the second group relatively high.

To represent the "low exposure to Less Wrong" population, I drew out of the survey results only those people who had been on the site less than a month and had 0 karma (karma is a measure of how much you contribute on Less Wrong). There were 47 of these people who had filled in cryonics information. To confirm that these people hadn't been influenced by Less Wrong much, I checked their exposures to the Sequences, the posts by Eliezer Yudkowsky that make up the core of the site. 64% of these low-exposure people had never read any of Eliezer's posts, and another 22% said they had read very few - less than a quarter of them. Only 14% of this population had read a quarter or more of Eliezer's posts.

Among this virgin population, no one was signed up for cryonics. Two people (4%) were in the process of trying to sign up. The mean probability given that cryonics could work was 15%

Now we set up their foil, the Less Wrong Veteran Group. These people have been in the community more than two years and have 1000 karma or above. 59 of this group had filled in cryonics information. Again to confirm that these people have been influenced by Less Wrong, I checked their Sequence exposure: over 70% of this group have read all of Eliezer's posts; only one of them had read less than 25%. There are a lot of these posts, so these are some serious site users.

Among this veteran population, 10 people - over a sixth of the total - were signed up for cryonics. When you add in the people who are in the process of trying to sign up, 53% - over half! - are would-be cryonauts. And the mean probability they give that cryonics could work is...12%.

So as people go from LW virgins to LW veterans, the probability that they are planning to get frozen increases by over thirteen times, but if you ask them whether cryonics will work, they are slightly but noticeably more skeptical. How could this be?

(dramatic question intended entirely for RationalWiki readers; this makes perfect sense for anyone from Less Wrong)

Less Wrong is a site that tries to teach people how to reason and make decisions. One of the first lessons it teaches is to think in probabilities. Let me give an example of this:

Consider a perfectly normal lottery with a jackpot of a million dollars and ten million tickets each of which costs $1. If you know how to do basic expected utility calculations it's very obvious that this lottery is a bad deal.

Now consider a second lottery where the lottery company makes some sort of horrible mistake. The jackpot is still a million dollars, but now there are only three tickets, each of which costs $1 and has a 33% chance of winning. If you know how to do basic expected utility calcuations it's very obvious that this lottery is a good deal. Even if you don't know the math, just eyeballing the chance to pay a dollar for a one in three chance of winning a million bucks seems like a good idea.

Now imagine you bought a ticket for the second lottery, and a friend - let's call her Rachel Nalwiki - comes up to you. "You bought a lottery ticket?!" she asks. "Do you really believe you'll win? I asked the lottery commissioner, and he said the odds of winning were only 33%! But you seem to believe you'll win anyway! How dumb can one person be?"

The flaw in Rachel's argument is that "believe you'll win" and "believe you'll lose" aren't really the right categories to use here. Since I know the probability of winning is 33%, if you asked me outright "Do you believe you will win the lottery", I would have to admit I do not - it's more likely that I'll lose than that I'll win. And yet I was perfectly justified to buy the ticket anyway.

Rachel retorts: "But everyone knows the lottery is a tax on people who are bad at math! Only stupid people play the lottery!" This is true. You definitely look stupid for playing, at least until the results come in. And there's a 2/3 chance you'll look stupid forever. But for a 1/3 shot at a million dollars, it's worth looking a little stupid.

Cryonics is much the same. Rachel tells the salesman: "Cryonics only has a 10% chance of success! That's less than 50%! That means it probably won't work! You are a bad person, to sell people this thing that probably won't work!"

You tell the salesman: "Wait, you mean I can buy a 10% shot at living forever for only $25 a month? That's less than I pay for car insurance! Where have you been my whole life?"

The difference between the newbie Less Wrongers who don't like cryonics and the veteran Less Wrongers who do like cryonics isn't that the veterans have a higher probability of it working. It's that they know what to do with probabilities once they have them, and they have a better estimate of the relative costs and benefits of looking stupid versus living forever.

To which Rachel might reasonably retort: "Well that just means that everyone, including the general population, has too high an estimate of cryonics. And it's only after people learn Less Wrong's techniques for thinking in probabilities that their ignorance has adverse consequences. You know what they say about a little learning being a dangerous thing..."

In fact, let's have that conversation:
Rachel: The general population is hopelessly biased in favor of cryonics. I am smarter than they are and I'm certain it won't work.

Scott: But we both know one hundred percent certainty is mathematically impossible. So exactly how sure are you that cryonics doesn't work, Rachel?

Rachel: Oh, at least ninety-nine percent certain.

Scott: And are you aware that when people say they're ninety-nine percent certain on a difficult question, they're wrong twenty to forty percent of the time?

Rachel: Uh, no?

Scott: It's a pretty universal problem. So unless you have personally studied techniques for estimating probability estimates and gone through some kind of laborious calibration training, I'm just going to mentally adjust every time you say "ninety-nine percent certain" on a difficult non-mathematical question to "approximately seventy percent certain".

Rachel: Um...

Scott: So, I hear you're only approximately seventy percent certain that cryonics doesn't work...

I guess if anyone from RationalWiki read all the way through here I might as well finally answer their original question of "whether I believe cryonics works".

I haven't really looked into it at all, so my default mode is to go with the opinions of smart people who have. I don't know of any neuroscientists who have given a precise numerical estimate, and even if they did I wouldn't expect them to know how to use probabilities and to avoid the "99% sure actually means 70% sure" problem. If some smart unbiased neuroscientists with calibration training have given some probability estimate on cryonics, I will happily accept whatever they say.

In the absence of that, I'll just go with that 12% number.

Am I signed up for cryonics? No, and I am not planning to. This isn't because I'm less crazy than the rest of Less Wrong, it's because I'm much, much crazier.
Tags: rationality
  • Post a new comment


    Anonymous comments are disabled in this journal

    default userpic

    Your IP address will be recorded  

← Ctrl← Alt
Ctrl →Alt →
← Ctrl← Alt
Ctrl →Alt →