I guess the obvious question is: Where do you draw the line regarding who to include in “the community”, why, and how does it affect the results?
An explicit swap is very clean and the counterfactual is known. Could they make a non-official, un-enforcable deal (I am very much not an expert on this at all and much less a lawyer!)?
Thinking about how ones situation impacts effective opportunities is also reasonable [1], but who is to say that they have not already assessed their specific viable options and are still convinced that their preferred choice is the best? People want to use their donations to “vote” and express an opinion, and that’s probably valuable?
[1] I suspect this is relatively common among people donating to underfunded areas in Animal welfare.
simon
While this sounds intuitively right, I think in the simplest utility maximising setting (iid additive errors with mean zero) your first claim does not seem true? The best looking noisy option is still most likely to be the best?
(I need to think more about the maths, but at least you need some kind of shrinkage to a prior that can change the ranking, which you’re unlikely to get, and if you’re maximising utility the solution is always fully concentrated?)
To be clear, it need not be deliberate and they need not benefit personally!
I was more referring to the diversification as implied by “don’t focus the vast majority of efforts on one cause”, which to me meant more “if you’re a decision maker over some amount of resources, you should diversify the allocation across cause areas”. Which I agree with, but it’s quite hard to really justify.
Yes, via nonlinearity you can get to diversification, but this means making additional assumptions beyond sampling error/publication bias/optimiser curse type effects.
The nonlinearity you’re describing matters on a movement level, but not on a individual decision makers level.
”Impact risk aversion” is another mechanism to get diversification which I think can be reasonable in cases where eg low impact reduces the probability of future donations or similar.
One channel I think is under explored and might work well as a justification for diversification in practice is something like this (I haven’t thought about this rigorously though): if I predictably optimise and my objective function is known to others, they will (in the worst case, possibly thru misaligned incentives) feed me biased information to influence my decision, and optimisation is very sensitive to noise, therefore I subject myself to adverse selection. So basically by not optimising but diversifying across good options you reduce the negative impact of this type of adverse selection. In this case, the “errors” are not iid. Hard to say how much diversification that yields.
I like diversification as a reaction to this type of uncertainty, but it does not trivially follow? I might be missing something—do you have a favourite minimal set of assumptions that rigorously yield diversification as a function of this?
Or perhaps the conclusion could just be “if you want to try very hard, don’t forget to have good epistemics and course correct if needed”?
Cause prioritisation is a function of the marginal impact on the outcome per marginal dollar/hour spent or similar.
This adds another layer of complexity because you can’t just eg “integrate out” the existential risk first and then reason about the impact, you kind of need to do it jointly.This means you also need to include sources of impact uncertainty.
To me, this ~ removes many longtermist cause areas because it drags most of the “impact mass” too close to zero—but people have different opinions on this of course.
Point 1 is interesting—you can do better if you have and incorporate a prior, but the question is where that comes from. I think often it’s easier to have priors over intervention success than existential outcomes per se.
The signalling aspect is probably overwhelmingly important for career success and ultimately alignment. A high GPA will open up roles that you could otherwise not access. Unfortunately hiring processes often filter based on it. Top 100 is likely not good enough to merely rely on brand recognition.
(I don’t work in this field but in a competitive & technical area and I’m sure that the recent graduate CVs I see are already pre-filtered by HR based on these kinds of blunt metrics.)
An important (and to me fairly open) related question is to what extent this ends up being action-guiding?
Eg if I lower my probability estimate of this materialising by 10 percentage points, how much will it affect resources I spend on helping to prepare for the possible outcome? Perhaps people here have thoughts on that—my impression is that working on improving the future opportunity set for such donations is relatively robustly good right now?
Yes, I agree, it is non-trivial to assess (approximately, probably) bounded downside and this needs to be done on a case-by-case basis, including in global health.
However, if you find that there are fat tails on a portfolio level but none on the individual intervention level, then the individual intervention was probably not assessed comprehensively? Interaction effects and second order effects matter, of course, but at least a portion of them should arguably be credited back to individual iterventions as you assess them.
This is a super valuable article, thank you for sharing!
Based on this, would it be sensible or interesting to fund a third-party M&E organisation, whose costs are considered somewhat separately, that provides independent evaluation at a relatively early stage and that builds more knowledge around how to do this well over time? Are there funders that might be interested in this?
The exchange rate & inflation fluctuation argument is also interesting. Clearly, EA organisations won’t plausibly be able to predict this (otherwise they should be trading currencies!).
However, the volatility can be modelled reasonably well, and this is a source of uncertainty in CE estimates that should be considered. However, it is usually not communicated. RCTs’ uncertainty, implementation biases and publication bias are somewhat similar in nature. They are tricky to model, but even a rough model would be great.
I would love to see more CE estimates whose output is an entire distribution and not just a point estimate (looking at you, GiveWell ;) ). This can be done, even very approximately, and it would be an improvement over what is commonly shared. I feel a reasonable version of this could be vibecoded fairly quickly?
When the downside is limited and the upside is not, variance is your friend. When the downside is possibly very negative, it is not.
There does indeed seem to be a split in the community, but I’m not sure it’s great to work towards that rather than against it.
I kind of try to speak to folks in EA 2.0 occasionally despite being pretty squarely in EA 1.0 and that’s probably net positive, e.g. to avoid a complete echo chamber?
Yes, funding eg “research that also produces forecasts” seems in a completely different category to me than eg prediction markets or platform building.
I feel the original article perhaps conflates different types of “forecast” funding a bit too much, although I tend to agree with its overall sentiment.
Yes, exactly. My point is that people are pretty aware and claiming otherwise is a bit of a straw man type fallacy—but I might be wrong, perhaps I interact with different people :D
I think this is essentially a straw man?
Everyone I know who doesn’t like donating to AI safety basically thinks it’s because p(influencing the outcome positively) is too low.
Prediction markets seem to be a great business (mostly gambling with all the problems associated with it) so “funding” in the sense of investing in them could be sensible while “funding” in the donation sense not. (And then later donation to AMF or similar).
In general, I’m hesitant to donate to stuff that’s plausibly just a really good business in its own right.
Note that in the context of trading/investing, the two terms are often used differently. There, “mean reversion” often means negative autocorrelation of returns, which can either be ~causal or driven by price level noise (which in turn is more like a “regression to the mean” idea). If you invest in a mean reversion strategy you tend to have an actual mechanism in mind though.
“Regression to the mean” is a less ambiguous term and generally means what you describe.
Thanks a lot Joey, this is definitely worth reading for people in the wider EA space, not only larger scale donors or people working in philanthropy directly.
What I’ve found particularly helpful are the rough quantitative guidelines regarding “charity time consumed per amount donated” and “how to donate as a function of annual amount and time spent per week”.
This is very valuable to better position myself from an earning-to-give perspective.
I think it might perhaps be interesting to write a short summary of that for the forum, perhaps targeted more at a median e2g EA? (If that doesn’t exist already.)Separately, it’s great to see that the book really embraces plurality in what areas donors prioritise without too strong a view on what’s preferable in the author’s opinion.
Yeah I agree—expected utility maximisation really starts to fall apart in this existential risk regime, even over trajectories rather than applied statically, and it only makes sense “locally” and at the margin.
Personally I’m very happy to bite the bullet and not be rigorously utilitarian, but I’m also a global health focussed “old school EA” thinking about how much to diversify donations across charities ;)