Comment author: brianwang712 24 March 2018 10:24:57PM 1 point [-]

I wonder how much the "spend 1 year choosing and 4 years relentless pursuing a project" rule of thumb applies to having a high-impact career. Certain career paths might rely on building a lot of career capital before you can have high-impact, and career capital may not be easily transferable between domains. For example, if you first decide to relentlessly pursue a career in advancing clean meat technology for four years, and then re-evaluate and decide that influencing policymakers with regards to AI safety is the highest-value thing for you to do, it's probably going to be difficult to pivot. There's a sense in which you might be "locked in" to a career after you spend enough time in it. My sense is that, for career-building in the face of uncertainty, it might be best to prioritize keeping options open (e.g., by building transferable career capital) and/or spending more time on the choosing phase.

Comment author: Jeffhe  (EA Profile) 22 March 2018 01:32:28AM *  0 points [-]

Hey Brian,

I just wanted to note that another reason why you might not want to use the veil-of-ignorance approach to justify why we should save the greater number is that it would force you to conclude that, in a trade off situation where you can either save one person from an imminent excruciating pain (i.e. being burned alive) or another person from the same severe pain PLUS a third person from a very minor pain (e.g. a sore throat), we should save the second and third person and give 0 chance to the first person.

I think it was F. M. Kamm who first raised this objection to the veil-of-ignorance approach in his book Morality, Mortality Vol 1. (I haven't actually read the book). Interestingly, kbog - another person I've been talking with on this forum - accepts this result. But I wonder if others like yourself would. Imagine Bob, Amy and Susie were in a trade off situation of the kind I just described, and imagine that Bob never actually had a chance to be in Amy's or Susie's position. In such a situation, do you think you should just save Amy and Susie?

Comment author: brianwang712 23 March 2018 02:39:21PM 0 points [-]

Yes, I accept that result, and I think most EAs would (side note: I think most people in society at large would, too; if this is true, then your post is not so much an objection to the concept of EA as it is to common-sense morality as well). It's interesting that you and I have such intuitions about such a case – I see that as in the category of "being so obvious to me that I wouldn't even have to hesitate to choose." But obviously you have different intuitions here.

Part of what I'm confused about is what the positive case is for giving everyone an equal chance. I know what the positive case is for the approach of automatically saving two people vs. one: maximizing aggregate utility, which I see as the most rational, impartial way of doing good. But what's the case for giving everyone an equal chance? What's gained from that? Why prioritize "chances"? I mean, giving Bob a chance when most EAs would probably automatically save Amy and Susie might make Bob feel better in that particular situation, but that seems like a trivial point, and I'm guessing is not the main driver behind your reasoning.

One way of viewing "giving everyone an equal chance" is to give equal priority to different possible worlds. I'll use the original "Bob vs. a million people" example to illustrate. In this example, there's two possible worlds that the donor could create: in one possible world Bob is saved (world A), and in the other possible world a million people are saved (world B). World B is, of course, the world that an EA would create every time. As for world A, well: can we view this possible world as anything but a tragedy? If you flipped a coin and got this outcome, would you not feel that the world is worse off for it? Would you not instantly regret your decision to flip the coin? Or even forget flipping the coin, we can take donor choice out of it; wouldn't you feel that a world where a hurricane ravaged and destroyed an urban community where a million people lived is worse than a world where that same hurricane petered out unexpectedly and only destroyed the home of one unlucky person?

If so, then why give tragic world A any priority at all, when we can just create world B instead? I mean, if you were asked to choose between getting a delicious chocolate milkshake vs. a bee sting, you wouldn't say "I'll take a 50% chance of each, please!" You would just choose the better option. Giving any chance, no matter how small, to the bee sting would be too high. Similarly, giving any priority to tragic world A, even 1 in 10 million, but be too high.

Comment author: Jeffhe  (EA Profile) 14 March 2018 08:24:46PM *  0 points [-]

It would be a mistake to conclude, from a lack of knowledge about one's position, that one has an equal chance of being in any one's position. Of course, if a person is behind the veil of ignorance and thus lacks relevant knowledge about his/her position, it might SEEM to him/her that he has an equal chance of being in any one's position, and he/she might thereby be led to make this mistake and consequently choose to save the greater number.

In any case, what I just said doesn't really matter because you go on to say,

"Note that it doesn't have to be actually true that Bob has an equal chance as Susie and Amy to have disease X vs. disease Y; maybe a third party, not behind the veil of ignorance, can see that Bob's genetics predispose him to disease X, and so he shouldn't sign the agreement. But Bob doesn't know that; all that is required for this argument to work is that Bob, Susie, and Amy all have the same subjective probability of ending up with disease X vs. disease Y, viewing from behind the veil of ignorance."

Let us then suppose that Bob, in fact, had no chance of being in either Amy's or Susie's position. Now imagine Bob asks you why you are choosing to save Amy and Susie and giving him no chance at all, and you reply, "Look, Bob, I wished I could help you too but I can't help all. And the reason I'm not giving you any chance is that if you, Amy and Susie were all behind the veil of ignorance and was led to assume that each of you had an equal chance of being in anyone else's position, then all of you (including you, Bob) would have agreed to the principle of saving the greater number in the kind of case you find yourself in now."

Don't you think Bob can reasonably reply, "But Brian, whether or not I make that assumption under the veil of ignorance is irrelevant. The fact of the matter is that I had no chance of being in Amy's or Susie's position. What you should do shouldn't be based on what I would agree to in a condition where I'm imagined as making a false assumption. What you should do should be based on my actual chance of being in Amy's or Susie's position. It should be based on the facts, and the fact is that I NEVER had a chance to be in any of their positions. Look, Brian, I'm really scared. I'm going to suffer a lot if you choose to save Amy and Susie - no less than any one of them would suffer. I can imagine that they must be very scared too, for each of them would suffer just as much as me were you to save me instead. In this case, seeing that we each have the same amount to suffer, shouldn't you give each of us an equal chance of being helped, or at least give me some chance and not 0?"

How would you reply? I personally think that Bob's reply shows the clear limits of this hypothetical contractual approach to determining what we should do in real life.

UPDATE (ADDED ON MAR 21): No need to read past this point since another person (kbog) made me realize that the paragraph below rests on a misunderstanding of the veil-of-ignorance approach.

Regarding the second point, I think what any person would agree to behind the veil of ignorance (even assuming the truth of the assumption that each has an equal chance of being in anybody's position) is highly dependent on their risk-adverseness to the severest potential pain. Towards the extreme ends that you described, people of varying risk-adverseness would perhaps be able to form a consensus. But it gets less clear as we consider "middle-of-the-road" cases. As you said people's intuitions here start to differ (which I would peg to varying degrees of risk-adverseness to the severest potential pain). But the question then is whether this hypothetical contractual approach can serve as a “reasonable way of adjudicating between the interests of different numbers of people suffering different amounts of pain” since your intuition might not be the same as the person whose fate might rest in your hands. Is it really reasonable to decide his fate using your intuition and not his?

Comment author: brianwang712 17 March 2018 07:00:58AM *  1 point [-]

Regarding the first point, signing hypothetical contracts behind the veil of ignorance is our best heuristic for determining how best to collectively make decisions such that we build the best overall society for all of us. Healthy, safe, and prosperous societies are built from lots of agents cooperating; unhappy and dangerous societies arise from agents defecting. And making decisions as if you were behind the veil of ignorance is a sign of cooperation; on the contrary, Bob's argument that you should give him a 1/3 chance of being helped even though he wouldn't have signed on to such a decision behind the veil of ignorance, simply because of the actual position he finds himself in, is a sign of defection. This is not to slight Bob here -- of course it's very understandable for him to be afraid and to want a chance of being helped given his position. Rather, it's simply a statement that if everybody argued as Bob did (not just regarding charity donations, but in general), we'd be living in a much unhappier society.

If you're unmoved by this framing, consider this slightly different framing, illustrated by a thought experiment: Let's say that Bob successfully argues his case to the donor, who gives Bob a 1/2 chance of being helped. For the purpose of this experiment, it's best to not specify who in fact gets helped, but rather to just move forward with expected utilities. Assuming that his suffering was worth -1 utility point, consider that he netted 1/2 of an expected utility point from the donor's decision to give everyone an equal chance. (Also assume that all realized painful incidents hereon are worth -1 utility point, and realized positive incidents are worth +1 utility point.)

The next day, Bob gets into a car accident, putting both him and a separate individual (say, Carl) in the hospital. Unfortunately, the hospital is short on staff that day, so the doctors + nurses have to make a decision. They can either spend their time to help Bob and Carl with their car accident injuries, or they can spend their time helping one other indivdual with a separate yet equally painful affliction, but they cannot do both. They also cannot split their time between the two choices. They have read your blog post on the EA forum and decide to flip a coin. Bob once again gets a 1/2 expected utility point from this decision.

Unfortunately, Bob's hospital stay cost him all his savings. He and his brother Dan (who has also fallen on hard times) go to their mother Karen to ask for a loan to get them back on their feet. Karen, however, notes that her daughter (Bob and Dan's sister) Emily has also just asked for a loan for similar reasons. She cannot give a loan to Bob and Dan and still have enough left over for Emily, and vice versa. Bob and Dan note that if they were to get the loan, they could both split that loan and convert it into +1 utility point each, whereas Emily would require the whole loan to get +1 utility point (Emily was used to a more lavish lifestyle and requires more expensive consumption to become happier). Nevertheless, Karen has read your blog post on the EA forum and decides to flip a coin. Bob nets a 1/2 expected utility point from this decision.

What is the conclusion from this thought experiment? Well, if decisions were made to your decision rule, providing each individual an equal chance of being helped in each situation, then Bob nets 1/2 + 1/2 + 1/2 = 3/2 expected utility points. Following a more conventional decision rule to always help more people vs. less people if everyone is suffering similarly (a decision rule that would've been agreed upon behind a veil of ignorance), Bob would get 0 (no help from the original donor) + 1 (definite help from the doctors + nurses) + 1 (definite help from Karen) = 2 expected utility points. Under this particular set of circumstances, Bob would've benefitted more from the veil of ignorance approach.

You may reasonably ask whether this set of seemingly fantastical scenarios has been precisely constructed to make my point rather than yours. After all, couldn't Bob have found himself in more situations like the donor case rather than the hospital or loan cases, which would shift the math towards favoring your decision rule? Yes, this is certainly possible, but unlikely. Why? For the simple reason that any given individual is more likely to find themselves in a situation that affects more people than a situation that affects few. In the donor case, Bob had a condition where he was in the minority; more often in his life, however, he will find himself in cases where he is in the majority (e.g., hospital case, loan case). And so over a whole lifetime of decisions to be made, Bob is much more likely to benefit from the veil-of-ignorance-type approach.

Based on your post, it seems you are hesitant to aggregate utility over multiple individuals; for the sake of argument here, that's fine. But the thought scenario above doesn't require that at all; just aggregating utility over Bob's own life, you can see how the veil-of-ignorance approach is expected to benefit him more. So if we rewind the tape of Bob's life all the way back to the original donor scenario, where the donor is mulling over whether they want to donate to help Bob or to help Amy + Susie, the donor should consider that in all likelihood Bob's future will be one in which the veil-of-ignorance approach will work out in his favor moreso than the everyone-gets-an-equal-chance approach. So if this donor and other donors in similar situations are to commit to one of these two decision rules, they should commit to the veil of ignorance approach; it would help Bob (and Amy, and Susie, and all other beneficiaries of donations) the most in terms of expected well-being.

Another way to put this is that, even if you don't buy that Bob should put himself behind a veil of ignorance because he knows he doesn't have an equal chance of being in Amy's and Susie's situation, and so shouldn't decide to sign a cooperative agreement with Amy and Susie, you should buy that Bob is in effect behind a veil of ignorance regarding his own future, and therefore should sign the contract with Amy and Susie because this would be cooperative with respect to his future selves. And the donor should act in accord with this hypothetical contract.

I would respond to the second point, but this post is already long enough, and I think what I just laid out is more central.

I will also be bowing out of the discussion at this point – not because of anything you said or did, but simply since it took me much more time to write up my thoughts than I would have liked. I did enjoy the discussion and found it useful to lay out my beliefs in a thorough and hopefully clear manner, as well as to read your thoughtful replies. I do hope you decide that EA is not fatally flawed and to stick around the community :)

Comment author: Jeffhe  (EA Profile) 13 March 2018 10:03:42PM *  1 point [-]

Hi Brian,

Thanks for your comment and for reading my post!

Here's my response:

Bob, Susie and Amy would sign the agreement to save the greater number if they assumed that they each had an equal chance of being in any of their positions. But, is this assumption true? For example, is it actually the case that Bob had an equal chance to be in Amy's or Susie's position? If it is the case, then saving the greater number would in effect give each of them a 2/3 chance of being saved (the best chance as you rightly noted). But if it isn't, then why should an agreement based on a false assumption have any force? Suppose Bob, in actuality, had no chance of being in Amy's or Susie's position, then is it really in accordance with reason and empathy to save Amy and Susie and give Bob zero chance?

Intuitively, for Bob to have had an equal chance of being in Amy's position or Susie's position or his actual position, he must have had an equal chance of living Amy's life or Susie's life or his actual life. That's how I intuitively understand a position: as a life position. To occupy someone's position is to be in their life circumstances - to have their life. So understood, what would it take for Bob to have had an equal chance of being in Amy's position or Susie's position or his own? Presumably, it would have had to be the case that Bob was just as likely to have been born to Amy's parents or Susie's parents or his actual parents. But this seems very unlikely because the particular “subject-of-experience” or “self” that each of us are is probably biologically linked to our ACTUAL parents' cells. Thus another parent could not give birth to us, even though they might give birth to a subjective-of-experience that is qualitatively very similar to us (i.e. same personality, same skin complexion, etc).

Of course, being in someone's position need not be understood in this demanding (though intuitive) way. For example, maybe to be in Amy's position just requires being in her actual location with her actual disease, but not e.g. being of the same sex as her or having her personality. But insofar as we are biologically linked to our actual parents, and parents are spread all over the world, the odds of Bob having had an equal chance of being in his actual position (i.e. a certain location with a certain disease) or in Amy's position (i.e. a different location with an equally painful disease) is highly unlikely. Think also about all the biological/personality traits that make a person more or less likely to be in a given position. I, for example, certainly had zero chance of being in an NBA position, given my height. Of course, as we change in various ways, our chances to be in certain positions change too, but even so, it is extremely unlikely that any given person, at any given point in time, had an equal chance of being in any of the positions of a trade off situation that he is later to be involved in.

UPDATE (ADDED ON MAR 18): I have added the above two paragraphs to help first-time readers better understand how I understand "being in someone's position" and why I think it is most unlikely that Bob actually had an equal chance of being in Amy's or Susie's position. These two paragraphs have replaced a much briefer paragraph, which you can find at the end of this reply. UPDATE (ADDED ON MAR 21): Also, no need to read past this point since someone (kbog) made me realize that the question I ask in the paragraph below rests on a misunderstanding of the veil-of-ignorance approach.

Also, what would the implications of this objection be for cases where the pains involved in a choice situation are unequal? Presumably, EA favors saving a billion people each from a fairly painful disease than a single person from the excruciating pain of being burned alive. But is it clear that someone behind the veil of ignorance would accept this?


Original paragraph that was replaced: "Similarly, is it actually the case that each of us had an equal chance of being in any one of our positions? I think the answer is probably no because the particular “subject-of-experience” or “self” that each of us are is probably linked to our parents' cells."

Comment author: brianwang712 14 March 2018 05:22:03AM *  3 points [-]

I do think Bob has an equal chance to be in Amy's or Susie's position, at least from his point of view behind the veil of ignorance. Behind the veil of ignorance, Bob, Susie, and Amy don't know any of their personal characteristics. They might know some general things about the world, like that there is this painful disease X that some people get, and there is this other equally painful disease Y that the same number of people get, and that a $10 donation to a charity can in general cure two people with disease Y or one person with disease X. But they don't know anything about their own propensities to get disease X or disease Y. Given this state of knowledge, Bob, Susie, and Amy all have the same chance as each other of getting disease X vs. disease Y, and so signing the agreement is rational. Note that it doesn't have to be actually true that Bob has an equal chance as Susie and Amy to have disease X vs. disease Y; maybe a third party, not behind the veil of ignorance, can see that Bob's genetics predispose him to disease X, and so he shouldn't sign the agreement. But Bob doesn't know that; all that is required for this argument to work is that Bob, Susie, and Amy all have the same subjective probability of ending up with disease X vs. disease Y, viewing from behind the veil of ignorance.

Regarding your second point, I don't think EA's are necessarily committed to saving a billion people each from a fairly painful disease vs. a single person being burned alive. That would of course depend on how painful the disease is, vs. how painful being burned alive is. To take the extreme cases, if the painful disease were like being burned alive, except just with 1% less suffering, then I think everybody would sign the contract to save the billion people suffering from the painful disease; if the disease were rather just like getting a dust speck in your eye once in your life, then probably everyone would sign the contract to save the one person being burned alive. People's intuitions would start to differ with more middle-of-the-road painful diseases, but I think EA is a big enough tent to accommodate all those intuitions. You don't have to think interpersonal welfare aggregation is exactly the same as intrapersonal welfare aggregation to be an EA, as long as you think there is some reasonable way of adjudicating between the interests of different numbers of people suffering different amounts of pain.

Comment author: brianwang712 13 March 2018 02:27:53PM 10 points [-]

One additional objection that one might have is that if Bob, Susie, and Amy all knew beforehand that you would end up in a situation where you could donate $10 to alleviate either two of them suffering or one of them suffering, but they didn't know beforehand which two people would be pitted against which one person (e.g., it could just as easily be alleviating Bob + Susie's suffering vs. alleviating Amy's suffering, or Bob + Amy's suffering vs. Susie's suffering, etc.), then they would all sign an agreement directing you to send a donation such that you would alleviate two people's suffering rather than one, since this would give each of them the best chance of having their suffering alleviated. This is related to Rawls' veil of ignorance argument.

And if Bob, Susie, Amy, and a million others were to sign an agreement directing your choice to donate $X to alleviate one person's suffering or a million peoples' suffering, again all of them behind a veil of ignorance, none of them would hesitate for a second to sign an agreement that said, "Please donate such that you would alleviate a million people's suffering, and please oh please don't just flip a coin."

More broadly speaking, given that we live in a world where people have competing interests, we have to find a way to effectively cooperate such that we don't constantly end up in the defect-defect corner of the Prisoner's Dilemma. In the real world, such cooperation is hard; but in an ideal world, such cooperation would essentially look like people coming together to sign agreements behind a veil of ignorance (not necessarily literally, but at least people acting as if they had done so). And the upshot of such signed agreements is generally to make the interpersonal-welfare-aggregative judgments of the type "alleviating two people's suffering is better than one", even if everyone agrees with the theoretical arguments that the suffering of two people on opposite sides don't literally cancel out, and that who's suffering matters.

Bob, Susie, Amy, and the rest of us all want to live in a world where we cooperate, and therefore we'd all want to live in a world where we make these kinds of interpersonal welfare aggregations, at the very least during the kinds of donation decisions in your thought experiments.

(For a much longer explanation of this line of reasoning, see this Scott Alexander post:

Comment author: brianwang712 12 March 2018 04:34:30AM *  3 points [-]

To add onto the "platforms matter" point, you could tell a story similar to Bostrom's (build up credibility first, then have impact later) with Max Tegmark's career. He explicitly advocates this strategy to EAs in 25:48 to 29:00 of this video:

Comment author: brianwang712 04 March 2018 04:58:46PM 7 points [-]

I'd like to hear more about your estimate that another non-human civilization may appear on Earth on the order of 100 million years from now; is this mostly based on the fact that our civilization took ~100 million years to spring up from the first primates?

If there is a high probability of another non-human species with moral value reaching our level of technological capacity on Earth in ~100 million years conditional on our own extinction, then this could lessen the expected "badness" of x-risks in general, and could also have implications for the prioritization of the reduction of some x-risks over others (e.g., risks from superintelligent AI vs. risks from pandemics). The magnitudes of these implications remain unclear to me, though.

Comment author: brianwang712 20 July 2017 01:56:28AM 6 points [-]

I think one important reason for optimism that you didn't explicitly mention is the expanding circle of moral concern, a la Peter Singer. Sure, people's behaviors are strongly influenced by laziness/convenience/self-interest, but they are also influenced by their own ethical principles, which in a society-wide sense have generally grown better and more sophisticated over time. For the two examples that you give, factory farming and slavery, your view seems to be that (and correct me if I'm wrong) in the future, people will look for more efficient ways to extract food/labor, and those more efficient ways will happen to involve less suffering; therefore, suffering will decrease in the future. In my head it's the other way around: people are first motivated by their moral concerns, which may then spur them to find efficient technological solutions to these problems. For example, I don't think the cultured meat movement has its roots in trying to find a more cost-effective way to make meat; I think it started off with people genuinely concerned about the suffering of factory-farmed animals. Same with the abolitionist movement to abolish slavery in the US; I don't think industrialization had as much to do with it as people's changing views on ethics.

We reach the same conclusion – that the future is likely to be good – but I think for slightly different reasons.

Comment author: Owen_Cotton-Barratt 30 October 2016 05:19:29PM 9 points [-]

Although there are dangers of having norms that dedicated EAs are less likely to pledge, because then not-pledging might become higher status in the community.

Comment author: brianwang712 30 October 2016 06:42:36PM *  1 point [-]

This is a good point; however, I would also like to point out that it could be the case that a majority of "dedicated donors" don't end up taking the pledge, without this becoming a norm. The norm instead could be "each individual should think through themselves, giving their own unique situations, whether or not taking the pledge is likely to be valuable," which could lead to a situation where "dedicated donors" tend not to take the pledge, but not necessarily to a situation where, if you are a "dedicated donor," you are expected not to take the pledge.

(I am highly uncertain as to whether or not this is how norms work; that is to say, whether norms connecting a group of people and a certain action could refrain from developing even though a majority of that group of people take that action.)

Comment author: Carl_Shulman 06 October 2016 05:03:53AM 0 points [-]

I don't think I'm following your argument. Are you saying that we should care about the absolute size of the difference in effort in the two areas rather than proportions?

Research has diminishing returns because of low-hanging fruit. Going from $1MM to $10 MM makes a much bigger difference than going from $10,001 MM to $10,010 MM.

Comment author: brianwang712 06 October 2016 06:52:12AM 0 points [-]

I guess the argument is that, if it takes (say) the same amount of effort/resources to speed up AI safety research by 1000% and to slow down general AI research by 1% via spreading norms of safety/caution, then plausibly the latter is more valuable due to the sheer volume of general AI research being done (with the assumption that slowing down general AI research is a good thing, which as you pointed out in your original point (1) may not be the case). The tradeoff might be more like going from $1 million to $10 million in safety research, vs. going from $10 billion to $9.9 billion in general research.

This does seem to assume that absolute size in difference is more important than proportions. I'm not sure how to think about whether or not this is the case.

View more: Next