## October 30, 2007

I'm pretty sure I wasn't doing that. ie, I did, given certain assumptions, commit to SPECKS in my reply.

For the record, my current view is if the choice is between torture vs _single speck event total per person for bignum people_, I'd go with the SPECKS

I do not consider the situation as linear, however. ie, two dust specks for one person is not precisely twice as bad as a single dust speck in one person, nor is that exactly as bad as two people each experiencing a single dust speck. In fact, I'd suspect that it'd be reasonable to consider a single dust speck per person total has a finite disutility even in the limiting case of infinite people.

If the situation instead is "torture vs an _additional_ dust speck per person for bignum people" then I'd want to know how many dust specks per person were already allocated, and as that number increased from 0, I'd probably lean a bit more toward TORTURE. But, of course, I know there'd have to be some value after which it'd really make no difference to add an additional dust speck or not, so back to SPECKS.

If I couldn't obtain that information, then I'd at least want to know how many others are going to be asked this. ie, is this isolated, or are there going to be some number of people "tested" like this such that if all answered SPECKS, then the result would be effectively worse than the TORTURE option, then, well, if I knew how many would be asked, and how many saying yes it would take, and if I knew some statistical properties of their utility functions and so on, then effectively I'd choose randomly, but setting the probability for the choice such that the expected utility for the outcome under the assumption that everyone used that heuristic would be maximized. (This is assuming direct communication between all the askeees isn't an option and so on. if it is, then that random heuristic wouldn't be needed)

If even that option was disallowed, well, I'd have to estimate based on whatever distribution of possibilities for each of those things that represented my current (At the time) state of knowledge.

_THIS_ is the point at which I get a bit stumped. If we say though "you have to make a decision, make it right now, even if it isn't that great" I'm still going to go with SPECKS, though, admitedly, with far less confidence that it's correct than what I said above.

Of course, now that I have a fallback last choice given no furthere knowledge/ability to consider, doing something about the whole situation that set up this issue would be something to investigate heavily. Also, I'd want to be developing a better model of exactly how to measure amount of effective suffering per "unit" suffering. I suspect it'd be some function of that plus how much it interferes with/overflows other possible states, etc etc etc.

As far as your overall point about people avoiding the decision, well, while it may be wise to avoid the habit of hiding from any uncomfortable decision, this is a bit different. I really can't see asking for a bit more information in the context of an edge case that was constructed to prod at our normal decision making methods and that was asked as a hypothetical thought experiment, _AND_ was a type of situation that I'd consider to be incredibly insanely mindexplodingly unlikely to pop up in Real Life(tm) any time soon as entirely unreasonable.

(*chuckles* on a meta level though, I just noticed that I seem to have chosen all possible options: commit to a specific choice, blabber about confusing aspects, ask for more information, and attempted to justify not commiting to a specific choice. There must be some sort of prize for this. :D)

Eliezer, Thomas Scanlon discusses this issue in the 'Aggregation' Section of Chapter 5 of his What We Owe To Each Other. Philosophers have been on it for awhile.

I deny that I have any obligation to choose now based on the available information.

What happens if I choose torture and somebody gets tortured for 49 years and then dies of natural causes? Do we get all the dust specs too? Have we gotten rid of 98% of them?

Can I ask for a volunteer from among 3^^^3 people, or do I have to take pot luck?

Do I have to choose right now? Do all 3^^^3 people get their dust specks the moment I decline to choose, or do they get them after the 50 years are up? If the former, what happens if I agree to the torture and then change my mind right after the dust specks didn't happen?

How do I know any of this is true? What if I get somebody tortured for 50 years and then the dust specks happen anyway? What if I do it and it turns out there aren't 3^^^3 people in the universe and it was all for nothing? Why should I take somebody's word about this?

Maybe we could start small. I could volunteer to be tortured for 50 years * 7 billion / 3^^^3 and stop the dust specks from everybody in this one world. I'd volunteer for that in a new york second.

I'm continually faced with choices for myself where the background is quite unclear. Take the contract, and maybe things go bad and it hurts my professional reputation. Wait for a better one and the money is late. Etc. And I make choices where the results won't show up in my lifetime. Throw mercury batteries in the trash or keep them around and wait for a chance to dispose of them properly. Drive my car to the store or wait a day and combine it with several other trips. Beyond the inconvenience balanced against the money, extra gas burned will have a small effect on billions of people over the next four or so generations. Maybe more than an eyeblink. I don't know how to quantify those effects and I don't spend a lot of thought on them. The immediate effects are easier to find out about, so I put most of my thought into those.

Show me the 3^^^3 people and I'll give them due consideration. Until then it's a thought experiment and I'll enjoy some time thinking about it.

Crocker's rules -- feel free to use me as an example if you think I gave a non-answer.

(I argued that the aggregation of a sufficient number of specks inflicted on a single person is equivalent to some significant length torture anyway.)

But there's a difference between refusing to choose in a situation that by its nature is necessarily hypothetical and trying not to choose in a real situation.

Until you pick one interim best guess, the discomfort will consume your attention, distract you from the search, tempt you to confuse the issue whenever your analysis seems to trend in a particular direction.

Oh no. Eliezer, I have disagreed with you at times, but you have not actually disappointed me until this moment. As an avid reader of yours, I beseech you, please think through this again.

You simply have not presented a moral dilemma. You've presented a pantomine; shadows on a wall; an illusion of a dilemma. If there's any dilemma here at all, it was whether I should play pretend-philosopher by giving an eloquent and vacuous response or else take philosophy and morals seriously by suggesting that your question is not yet ready to be answered. I chose the latter, partly because I also have been taking seriously your other writings-- the ones where you chide people for substituting wishful thinking for self-critical sober rational analysis. I'm attracted to the mind of a man who tries to live by a difficult and worthy principal, because that's what I do, too; and what I am doing.

Real moral dilemmas have context, and the secret to solving them always involves that context. We frequently find them in literature, richly expressed. Instead, you are just asking us to play a game with unspecified rules and goals. You toss off a scenario in a few sentences. How is that interesting? I guess it's a bit interesting to see how some people commenting have made bold assumptions and foisted unspoken premises on your example. It's a window onto their biases, maybe. Is that really enough to satisfy you?

I could understand if you don't want to make the effort to create a fully realized philosophical problem for us to work through (putting together those problems is a challenge). But geez, I'm surprised you would criticize me for doing what a philosopher is supposed to do: study the situation to understand the question better, rather than make a definite answer to a question I don't understand.

Oh no. Eliezer, I have disagreed with you at times, but you have not actually disappointed me until this moment.

You should find that of all the people you know, none of them seem beyond criticism; they will always fail to live up to your ideal of perfection. That's because there's only one person whose job it is to live up to that ideal.

But geez, I'm surprised you would criticize me for doing what a philosopher is supposed to do: study the situation to understand the question better, rather than make a definite answer to a question I don't understand.

I never thought of myself as a philosopher. I just set out to debug the universe. I often have to do so using incomplete information. My motor actions do not have the luxury of vagueness, however I caveat my "answers". If you think my philosophical dilemmas are vague, you should see the problem descriptions Nature hands me.

The people who filled in their own assumptions and stated a preference were acting courageously; they exposed themselves to criticism for the conditional, if not for the assumptions.

I guess if you really feel the question is so confused as to be answerless, I'll accept that. I would still challenge you to fill in plausible assumptions and state a preference.

The philosophy of refusing to come to a conclusion is called skeptcism. The word skeptic comes from the Greek to examine.
While I understand the need to make decisions, I'm not so sure that it should trump the desire to not accept answers (keep looking).
As has been pointed out in earlier posts, once a decision is made it often is hard to dislodge.
For example, many people today accept neo-Darwinism as an answer to evolution. Yet the evidence from biology would indicate that neo-Darwinism is either false or incomplete. (Try dislodging that one)
So while I agree that one often has to make decisions quickly based on incomplete and conflicting evidence, I don't think the question you posed in 'torture vs. dust specks' was framed in such a way as to demand that type of decision.

By the way, someone who has made up their mind about religion or the existence of para-psychological phenomena is not a skeptic in the historical meaning of the word.

_Yet the evidence from biology would indicate that neo-Darwinism is either false or incomplete._

Of course it's incomplete. No neodarwinist would have claimed it was complete.

Now that we know so much more than the neodarwinists did it's mostly of historical interest. But what we have now is still quite incomplete, and it will stay that way for the foreseeable future.

"I guess if you really feel the question is so confused as to be answerless, I'll accept that. I would still challenge you to fill in plausible assumptions and state a preference."

Remember the story, "The Lady and the Tiger"? The question was carefully formulated to be evenly balanced, to eliminate any reason to choose one over the other. Anything that got used to say one choice was better, implied that the story wasn't balanced quite right.

We could do that with your story too. If 3^^^3 people is enough to say it's better to torture one person, we could replace it with a smaller number, perhaps a googleplex. And if that's still too many we could try just a google. If people choose the specks we could increase the number of people, or maybe increase the number of specks.

At some point we get just the right number of specks to balance the torture for a modal number of people, and we're set. The maximum number of people will be unable to choose, because you designed it that way.

You did not say what happens if I don't choose. This is a glaring omission.

OK, let me tell one. You and your whole family have been captured by the Gestapo, and before they get down to the serious torture they decide to have some fun with you. They tell you that you have to choose, either they rape your daughter or your wife. If you don't choose which one then they'll rape them both. And you too.

Do you choose? If you refuse to choose then that's choosing for both of them to be raped. And you too.

But then, if you do choose, they rape them both anyway. And you too.

What should you do?

Okay, I'll take a position: a moral dilemma involving impossibly huge numbers, perfect certainty, no externalities, and no context is no more deserving of a clear-cut answer than the question of how I would explain waking up with a blue tentacle.

When people can't explain themselves they often make up answers.

"What the hell? A blue tentacle?"

"I must have gotten it from a toilet seat."

J Thomas, if you can't see a better option, you tell them to rape your wife. Duh.

(No, I'm not a sociopath, I've just trained myself not to whine about my options, just pick the obviously best of a bad lot quickly, and keep looking for an escape route. The scenario is legitimate, people in real life have faced worse.)

Nick Tarleton, the problem with explaining waking up with a blue tentacle is that it's so low-probability as to destroy the worldview you would use to explain it; by Bayes, you shouldn't be able to explain it post facto unless you anticipate it to some measurable degree ante-facto. But a blue tentacle doesn't destroy your utility function, so asking "What would you do if you woke up with a blue tentacle?" is a perfectly legitimate dilemma.

When I read Eliezer's original post, my moral intuition crashed. I was confused, and suspected something was wrong with either the question, or with me.

Are you really suggesting that choosing to not commit to an answer immediately but to instead think about it and explore the scenario for a while was the wrong answer? If the scenario were instead "choose TORTURE or SPECKS within the next N seconds or get one at random," and was real, not a thought experiment, then see Eliezer's point: inaction is an action.

I say all morality is meaningless/arbitrary AND I choose torture. How do I stand with you?
If I had said I would flip a coin, would that be satisfactory?

J Thomas- I'm not sure what your expertise- and this question is a little off post, but important to me and my personal biases, would you say the evidence today seems to indicate that the 'watchmaker' isn't blind? (maybe myopic...)

Eliezer: I don't think you read J Thomas carefully. He was saying, as far as I can tell from the last three sentences of his post, that the scenario itself strongly implies that you don't actually have the choice that it is asserted that you do have. As a hypothetical it fails. A person being tortured by the Gestapo is making a mistake to seriously consider the possibility that a supposed "choice" he is offered is anything but mockery and a part of his torture. *Any* person is making a mistake to seriously consider the possibility that his actions have any predictable impact on 3^^^3 other people because the chance of him being the one of those 3^^^3 people who was in the special position where he could effect the others rather than one of the others who could only be effected is simply too low.

"what would you do if your worldview had just been destroyed" is not, it seems to me, a legitimate question. The loss of your worldview implies the loss of any rational basis for inferring the consequences of your actions. It seems to me that you can ask, as a question about "you" the physical system "what would you do if you irrationally believed X", but not, as a question about rationality, "what would it be rational for you to do if you irrationally believed X"?

Incidentally, I decided upon consideration of what the math would actually look like for the type of utility function that I'd currently consider reasonable, that given a fixed population, disutility would be basically linear in number of people experiencing dust speck events (the other nonlinearities about one person experiencing a bunch of events would hold though) so am shifting my answer, tenatively, to TORTURE. (Just sticking this comment in this thread since I also made the other claim in this thread.)

Tuning one's preference function is a constrained optimization problem. What I want is a preference function simple enough for my very finite brain to be able to compute it in real time, and that does a good job (whatever exactly that means) on-some-kind-of-average over some plausible probability distribution of scenarios it's actually going to have to deal with.

Choosing between torturing one person for 50 years and giving 3^^^3 people minimally-disturbing dust specks is a long, long way outside the range of scenarios that have non-negligible probability of actually coming up. It's a long, long way outside the range of scenarios that my decision-theoretic intuition has been tuned on by a few million years of evolution and a few decades of experience.

My preference function returns values with (something a bit like) error bars on them. In this case, the error bars are much larger than the values: there's much more noise than signal. That's a defect, no doubt about it: a perfect preference function would never do that. A perfect preference function is probably also unattainable, given the limitations of my brain.

What possible reason is there for supposing that my preference function would be improved, for the actual problems it actually gets used for, by nailing down its behaviour far outside the useful range?

If there were good reason to think that decision theory is like (a Platonist's view of) logic, with a Right Answer to every question and no limits to its validity, then there would be reason to expect that nailing down my preference function's values out in la-la-land would be useful. But is there? Not that I know of. Decision theory is an abstraction of actual human preferences. Applying it to problems like Eliezer's might be like extrapolating quantum mechanics down to a scale of 10^-(10^100) m.

Douglas, my own bias is to think that evolution has given us 3 billion+ years of selection for evolving faster. And multicellular organisms (a small minority of the total but interesting to us) have found ways to make genetic "modules" that result in phenotypes which fit together in a modular way. Chordates build a variety of structures from keratin. Arthropods build a big variety of limbs. Etc. The body plans that are most flexible speciate into the largest variety of niches -- nematodes, arthropods, mollusks, and chordates, and you have the big majority of animal species in just those 4 phyla.

We don't make just random changes, we have "hotspots" that change a lot while others are mostly held fixed. A big variety of mechanisms evolved that encourage faster evolution, because those mechanisms are themselves selected.

Apologies for the off-topic note.

Michael Vassar, yes! Thank you for putting it so clearly.

J Thomas-- What you say fits well with the neo-Darwin model of evolution. One example you might be interested in that clearly does not is the tuberculosis bacteria. Google 'tuberculosis strain w' for more info. It turns out this sort of thing happens more than was previously thought (of course it wasn't thought to happen at all until fairly recently)
This is a case of motivated continuation on my part- the old model predicted a cure that turned out to be a recipe for making an incurable disease-- uh I want to understand better.

Douglas, I see nothing about strain w that's surprising. Would you like to suggest a blog and a thread to discuss this?

Douglas, I see nothing about strain w that's surprising. Would you like to suggest a blog and a thread to discuss this?

It's kind of past the point where this is really relevant, but I was interested to notice that lots of commenters launched into discussions of potential knock-on consequences of real-world speckification but not a single person queried the extended cost of a real-world 50-year torture option (infrastructure, training, torturer-trauma, wear and tear on electrodes etc.). Of course, as with any thought experiment dragging in any externalities at all was/is invalid: the experiment sets the parameters, and any speculation outside of these is irrelevant. But insofar as that was being done I thought it curious that these types of speculations all went one way.

If pressed, I'd hypothesize that this was because some people who saw that the 'specks' option was obviously the right choice were left feeling that there was a further trick of some kind: surely the obvious wrong answer, torture, must be right - or why would the thought experiment have been posed at all?

Personally, I'm a specks guy and I feel deeply suspiciuous of the torturers' reasoning: I suspect it of being dependent on a fallacious calculation of harm. But I think the thought experiment is of very limited value as it does not really mirror any real-world scenario that I can see.

J Thomas-- try www.wasdarwinwrong.com Best place for the info. because it presents the problems without demanding any particular solution.
outeast- good point about the speculations, but thought experiments can be off-the-wall and still be of value because they are designed to help see the world in a different or new way. Sometimes the off-the-wall ones are best for that reason IMO.

The comments to this entry are closed.

## May 2009

Sun Mon Tue Wed Thu Fri Sat
1 2
3 4 5 6 7 8 9
10 11 12 13 14 15 16
17 18 19 20 21 22 23
24 25 26 27 28 29 30
31