I do independent research on EA topics. I write about whatever seems important, tractable, and interesting (to me).
I have a website: https://mdickens.me/ Much of the content on my website gets cross-posted to the EA Forum, but I also write about some non-EA stuff over there.
I used to work as a software developer at Affirm.
I don’t see what you’re seeing about how time matters. Can you give an example of how changing the time would change your answer to one of the questions?
No strong opinion on whether it’s a good idea, but here are some options to avoid AI stocks:
Buy a global ex-US fund, because AI stocks are mostly US companies
If you also want to avoid AI companies in China and Korea, you could buy a Europe index
If you want US funds, there are sector ETFs such as the ones from SPDR, you could buy all of them except the tech sector ETF.
If you are okay with more complexity, you could buy a broad index fund while also shorting an AI fund (AGIX AIS, etc.). Or you could short a NASDAQ index fund, which overweights AI stocks and which might be cheaper to short-sell.
FWIW I personally don’t intentionally invest in AI stocks, but I don’t actively avoid them, either.
Thanks for the comments! I feel honored :P
Miscellaneous responses:
On your comment that the answers are contradictory or counterintutive or both, I want to add this is very much agreed on in the field. The famous result is Arrhenius (2000), which showed there are a number of assumptions we might want to make about population ethics, but they are mutually inconsistent—you can’t eat your cake and still have it.
I was originally going to cite Arrhenius because I was under the mistaken impression that he wrote in the 1950s or 60s and Parfit drew inspiration from him. But while making the quiz I realized it was the other way around. As I understand, Arrhenius’ impossibility proof is qualitatively the same thing as what Parfit laid out in Reasons and Persons, the difference being that Parfit only gave an informal proof and Arrhenius made it mathematically rigorous. I haven’t read either original source but that’s my understanding.
Because population ethics is so vexing, it’s very tempting to consider it in light of theories of moral uncertainty. [...]
The quiz doesn’t allow for moral uncertainty. The honest reason why not is that I didn’t think to do that, but thinking about it now, it sounds really hard because it creates a combinatorial explosion of outcomes. On Q4 for example, you could also answer “uncertain between K is better and they are equal” or “uncertain between K is better and they are incomparable” or “uncertain between all three”, or if you want to get even more specific, “pretty sure K is better but maybe they’re incomparable”, …
I wanted to make the quiz as comprehensive as possible while still being reasonably short. In particular I didn’t want to claim an answer was contradictory when actually it wasn’t, and I spent a lot of time looking for ways philosophers have proposed resolutions to seemingly-contradictory sets of answers. (Including hours spent faithfully representing views that so far ~zero quiz-takers have endorsed.) One thing I explicitly didn’t cover is the difference between preferred acts vs. preferred outcomes, because it would add an amount of complexity disproportionate to its usefulness. Moral uncertainty is another issue like that, I think.
You phrase Q7 and Q8, which are related to the Repugnant Conclusion, as if they are structurally identically. But they are not.
Q6 (comparing A and A+) does add new people. The move from A to B is first asked in terms of moving from A to A+ (Q6) and then A+ to B (Q7), and then you can chain those two moves down to Z, which is what Q8 is asking about. As I understand it, this is identical to Parfit’s construction of the mere addition paradox. The step where you add new people is in Q6, not in Q7, so the “do it again” is a reference to doing the Q6 step again. (And also doing the Q7 step again, but sounds like you are taking issue with the adding-new-people part.)
Possible that I’m still missing something here, please tell me if I am.
You give two options in each case (“yes, always”, “no, not always”) but I wanted there to be a third option “it depends on the details”.
If it depends on the details, then wouldn’t the answer be “no, not always”?
I’m working on a follow-up post about what I’ve learned from people’s responses. If there are any questions you want answered, leave a reply to this comment.
The very repugnant conclusion isn’t explicitly tested. I didn’t put much thought into this but my reason was that it seemed redundant with the questions that are already there. And I do think it’s sort of covered, but indirectly, and it’s not very clear.
The way it’s sort-of covered is that if you accept the repugnant conclusion, then you’re accepting that you can subtract happiness or add suffering to existing lives, and it’s still a net positive so long as you add sufficiently many more marginally-happy lives. So VRC directly follows from RC.
But there’s an alternative interpretation of RC where the marginally-happy people are not suffering at all, they just aren’t very happy either, so in principle you might accept RC even if you believe no amount of happiness can outweigh a small amount of suffering.
So to really capture VRC, I think the quiz needs two new questions:
If you add a new happy person (say welfare level 20), and a new unhappy person (say welfare level −7) such that the sum is positive, is that good, bad, neutral, or incomparable?
VRC
If you answer “good” to the first question, and your other answers are consistent with totalism, then that implies VRC, and the VRC question tests if your views are in conflict. If you answer “bad” to the first question, then it implies something like negative utilitarianism, or something like a person-affecting view with a procreation asymmetry (i.e. adding good lives is neutral, but adding bad lives is bad).
Whatever you consider to be the crossing point at which a life goes from “not worth living” to “worth living”, that’s defined as welfare level zero. A life at welfare level 7 is worth living by definition.
Q12&Q13. I don’t understand… These are comparable, right?
Some moral philosophers (Larry Temkin being the most prominent) say that if A > B, and B > C, then that doesn’t necessarily mean A > C. I agree that this sounds strange but it’s a view held by some people, so I wanted the quiz to allow for the possibility.
Yeah I agree it’s confusing, I wasn’t sure the best way to explain it. The intended meaning is:
Consider the difference between world A and world A+, in which A+ has the same people as in A but happier, and also some new somewhat-happy people. Then consider a new world B+ that is to B as A+ is to A. Similarly, consider the difference between A+ and B, in which the same people are there, but their welfare is more equal. Then consider a new world C that is to B+ as B is to A+. Then repeat this process many more times. Would you make the same decision every time?
(I’m not sure if that explanation was helpful)
So it’s not trying to claim that you preferred the right-hand option, “repeat the move” means “imagine a world that is to B as A+ is to A” (and similarly for A+ → B).
ETA: Tell me if you think this would be a clearer question:
Imagine making the same two moves as A → A+ and A+ → B, but this time starting at B: (1) make everyone slightly better off and create a new group of less-happy people; then (2) keep the same population, but increase both total welfare and equality. Repeat again and again a total of 9 times, each new rung making the population larger but less happy.
Would you make those same two judgements at every rung?
(that changes the part before the first colon, and keeps everything else the same)
I don’t think so. I’d say they have the potential to be “super-effective in expectation”. To me, “super-effective” with no qualifiers implies high confidence.
Hmmm I think the new version is harder to understand than the old version. The old version gave a colloquial discussion with illustrative examples. The new version uses a formal philosophical argument with a lot of technical terms, and it’s taken from an article that sort of assumes readers are already familiar with cluelessness. Premise P3 asserts that our understanding of the far future is “sufficiently coarse-grained” and doesn’t explain what that means or why to believe it, but IMO that explanation is the key thing that should be on a wiki page. The primary goal should be to explain to the reader what cluelessness is.
It is hilarious(ly sad) that the largest present-day problem in the entire world only receives on the order of $5 million/year. I really hope these orgs get more funding in the near future.
The argument isn’t that the variance on your EV is very high. The argument is that you can’t have an EV at all. The challenge of unawareness sequence makes this argument in detail.
That looks like a link to an unpublished draft that only you can view. I can’t see it
I agree that it doesn’t make sense to make extremely risky investments and then be very risk-averse with your donations.
I’m not sure about the donation recommendations. Why should donors have a small core of low-risk donations? Why not donate all their money to the single highest-EV opportunity? Or why not give to a few different opportunities that they believe are the best candidates for highest-EV, without concern for risk?
To clarify, I’m not saying that’s because LLMs don’t have welfare. It could also be that they do have welfare, but their statements are not correlated to their internal experiences in the obvious way.
We know that we can get LLMs to express any welfare statement by saying e.g. “pretend you are a sentient AI that’s having an awesome time / being tortured”. It could be that all LLM outputs are roleplaying, but there is some true underlying experience that we don’t (yet?) know how to observe.
For the purposes of the population ethics quiz, why does it matter how long the outcomes exist for? Like, take question 3, which compares outcome A against outcome B. Is there any situation where time changes your answer? For example, would you say, “If the people in each group will live for 10 years, then A is better than B, but if they will each live for 20 years, then B is better than A”?