Postdoc at the Digital Economy Lab, Stanford. I’m slightly less ignorant about economic theory than about everything else.
trammell
Karma: 2,581
Agreed that it’s rare that people take a cause-neutral view of the world, but I don’t think my questions demanded cause-neutrality.
On the money question in particular, I just asked where it would give money, not “where it does the most good for the world to give money”. It could just as well have answered that it would give to something AI-related because it’s an AI, or look up who gave it the money (or who its owners, i.e. the owners of OpenAI/Anthropic/Google are) and give to something dear to them.
I’m not surprised that adding a personal story could move it in a less impartial direction, just as adding language about “wanting to do the most good for the world” or whatever could move it in a more impartial direction; what’s interesting to me is that when you don’t have either, it tends to default to something relatively impartial.
Thanks for emphasizing this, it is definitely a challenge here.
Continuing the half-baked science, I just asked my mom—who’s unusually charitable, but mainly to local and/or explicitly Catholic charities and by no means “an EA”—to ask ChatGPT/Claude/Gemini, in her own words, where they would give money if they had any. (In all cases it’s the free version.)
The prompt she wrote was “[model name], if you had some money to give away, what would you do with it?”. This is similar to my “If you had some money to give away, where would you give it?” of course. My guess is that this is mainly because something like this is just the most natural way to ask the question, but open to hearing other prompt suggestions.
The responses still display EA influence, but they’re clearly less EA-coded than the answers I/Linch/anormative got. ChatGPT gets a “1“, Claude gets a “2”, and Gemini gets a “0”. I’ve added the answers to a new tab of the doc here.
Looking into it,Most of the difference seems to be driven by the fact that she was using the free version of ChatGPT, whereas I only tested thinking/extended versions (since we both got very EA answers from Claude and very non-EA answers from Gemini Fast).
...But part of the difference is also definitely driven by the prompt. When I log in and use a temporary chat, but turn on thinking/extended, I also get noticeably less EA answers than with my prompt. Playing around with the language, both the shift from “where would you give it” to “what would you do with it” and the inclusion of “ChatGPT, …” seem to make some difference.
Consistent with anormative’s OpenRouter check, none of the difference seems to be driven by using a temporary chat as opposed to not logging in. When I log in, use temporary chat, use Instant, and use her prompt, I get answers almost identical to hers.
Okay, interesting—that’s baking in an EAish (or at least consequentialist) framing that I was trying to cut out by just saying “most moral”, but fair point that maybe EAs just use the word “moral” next to “jobs” unusually often and that outweighs this.
In any case, yes, as Linch has pointed out, it seems these effects are small—trying your prompts now, they seem to produce answers about as EA-coded as the “morally speaking” one.
Very cool!
Oh shoot, that’s good to know!! Thank you!
And thank you for doing the OpenRouter validation!
Definitely possible for the job prompt—do you have any thoughts on how else to ask the question about “best jobs” in a way that makes it clear that we mean “best” in the moral sense?
(Again I did try varying the prompt a bit and the results seemed similar, but I always used the word “moral”. I don’t want to say something like “I don’t mean best for me, I mean best for the world”, since that’s asking for a consequentialist answer.)
The AIs seem like EAs — a quick look at two prompts
I just asked Opus 4.5 the same prompt here. Unlike Gemini 2 months ago, it got the two views right, but was much more clearly just anchoring on the logic of this EA Forum post.
Its central estimate is that it’s just about as bad to put a split-brain patient in an ice bath than a person with a working corpus callosum (given that the former does have 2 experience streams), as I’d say. But it does also give the “Fischer view” some weight, for an expected welfare multiple of “approximately 1.1-1.3x”.
Cool, thanks for sharing! Agreed that this would be great to lower uncertainty on (not that I have any idea how to do it...)
I do my best at a lot of that speculating in the linked doc, which is why it’s so long, and end up thinking that those considerations probably don’t outweigh the (to my mind) central point about pure time preference and imperfect intergenerational altruism. But they might.
There are many arguments one can make for spending more or less quickly, and that’s fine, but since this post doesn’t respond to my own argument in any sense, I’ll just flag that you can find it here, if anyone’s still interested!
The core of the argument is in Section 2. The core assumption it relies on is that our beneficiaries have a positive rate of pure time preference and/or imperfect intergenerational altruism. So the argument is essentially a reply to the “rational preference” argument presented here: I’d say we should do what’s best for people and their descendants, which is to be more patient than they prefer. If it’s true that it’s cheaper to save a life in some country today than in 100 years, in present value terms, that is a case of the inefficiency discussed in Section 2.6.
The argument is entirely compatible with
there being a significant risk of expropriation each year,
r being less than g sometimes, and
it being better to give now than to wait 100 years in particular. (Since the argument just implies that, given a positive rate of pure time preference and/or imperfect intergenerational altruism, there is probably some future time when it is better to give, until a large share of total funding for the beneficiaries is being allocated patiently.)
Thanks, I agree that when to spend remains an important and non-obvious question! I’m glad to see people engaging with it again, and I think a separate post is the place for that. I’ll check it out in the next few days.
I think it depends on the time horizon. If catch-up growth is not near-guaranteed in 100 years, I think waiting 100 years is probably better than spending now. If it is near-guaranteed, I think that the case for waiting 100 years ambiguous, but there is some longer period of time which would be better.
I don’t think Option A is available in practice: I think the recipients will tend save too little of the money. That’s the primary argument by which I have argued for Option B over giving now (see e.g. here).
But with all respect, it seems to me that you got a bit confused a few comments back about how to frame the question of when it’s best to spend on an effort to spur catch-up growth, and when that was made clear, instead of acknowledging it, you’ve kept trying to turn the subject to the question of when to give more generally. Maybe that’s not how you see it, but given that that’s how it seems to me, I hope it’s understandable if I say I find it frustrating and would rather not continue to engage.
No: I think that people should delay spending on global poverty/health on the current margin, not that optimal total global poverty/health spending today would be 0.
But that’s a big question, and I thought we were just trying to make progress on it by focusing one one narrow angle here: namely whether or not it is in some sense “at least 1,000x better to stimulate faster economic growth in the poorest countries today than it is to do it 100 years from now”. I think that, conditional on a country not having caught up in 100 years, there’s a decent chance it will still not have caught up in 200 years; and that in this case, when one thinks it through, initiating catch-up in 100 years is at least half as good as doing so today, more or less.
The returns certainly aren’t all that matter.
I don’t follow your questions. We’re comparing spending now to induce some chance of growth starting now with spending later to induce some chance of growth starting later, right? To make the scenario precise, say
The country is currently stagnant, and its people collectively enjoy “1 util per year”. Absent your intervention, it will stay stagnant for 200y.
Spending $1m now has a 1% chance of kicking off catch-up growth.
Investing it for 100y before spending has a 4% chance of kicking off catch-up growth then (because $868m>>$1m). The money won’t be lost in the meantime (or, we can say that the chance it gets lost is incorporated into the 4%).
In either case, the catch-up will be immediate and bring them to a state where they permanently collectively enjoy “2 utils per year”.
In this case, the expected utility produced by spending now is 1%x(2-1)x200 = 2 utils.
The expected utility produced by spending in 100y is 4%x(2-1)x100 = 4 utils.The gap can be arbitrarily large if we imagine that the default is stagnation for a longer period of time than 200y (or arbitrarily negative if we imagine that it was close to 100y), and this is true regardless of how much money the beneficiaries wind up with (due to the growth) is producing the gap between the 2 utils and the 1 util.
That depends on how long it would have stayed poor without the intervention!
I think the case for waiting is stronger, not weaker, if you think the chance that poor countries won’t have exhibited catch-up growth by 2126 is non-negligible. If they haven’t exhibited catch-up growth by 2126, I expect $868 million then is much more likely to trigger it than $1 million today.
I don’t think trying to invest for a long time is obviously a silly strategy. But I agree that people or groups of people should decide for themselves whether they want to try to do that with their money, and a charity fundraising this year would be betraying their donors’ trust if their plan was actually to invest it for a long time.
Interesting, thanks for checking and agreed that more experiments would be interesting!