Views are my own.
Thanks very much for the update!
Do GiveWell's published cost-effectiveness estimates already include an adjustment for the optimizer's curse? Or is the idea that donors should treat estimates like 35x cash as "raw" expected value calculations, to which they apply their own informal Bayesian adjustment along the lines of Holden's post?
Similar thoughts crossed my mind (including the thanks!).
For a donor whose values are closely aligned with GiveWell's and who trusts them to spend wisely on operations, it seems like an unrestricted donation might actually have the highest expected impact.
But it also seems like there's potentially a funging "cascade" across the different options such that marginal donations would be equivalent under certain circumstances, depending on details like the Excess Assets Policy.
I'd be very interested in an in-depth comparison of the different options for giving through GiveWell in terms of expected impact, funging, optimizer's curse, value alignment etc.
In Joe’s thought experiment, the “party” is (happy) life itself, and creating a new happy life is “inviting” an extra person to that party.
I agree with you that people who don’t exist are unable to care about things. There isn’t a concrete list of hypothetical people who we can say are “missing out” on a happy life by not existing. And we can’t ask the non-existent version of Michael whether he would like to exist.
But I don’t think it follows that there is no benefit to creating additional happy lives. I think the extra happiness is still valuable even if we can’t point to a coherent counterfactual where someone in particular would have “missed out” on the happiness. We can ask the version of Michael who does exist whether he’s glad to be alive, and by assumption his answer will be positive.
My intuition is that it does matter whether there are 10 or 1000 people enjoying a party because those extra 990 enjoyable experiences are separate from one another (and from the first 10) and each is valuable by itself. (For what it’s worth, my intuition would be much weaker if we specified that all 1000 experiences were completely identical.)
If you want to understand how someone might not be indifferent about creating happy people, I recommend Joe Carlsmith’s post, “Against neutrality about creating happy lives”. Here’s the passage that resonated most with me (referring to a thought experiment about having the choice to create a new person called Michael):
I feel like I have a chance to invite Michael to the greatest party, the only party, the most vast and terrifying and beautiful party, in the history of everything: the only party where there is music, and wet grass, and a woman he’ll fall in… love with — a party he would want to come to, a party he’ll be profoundly grateful to have been to, even if only briefly, even if it was sometimes hard.
I don’t have kids. But if I did, I imagine that showing them this party would be one of the joys. Saying to them: “Here, look, this is the world. These are trees, these are stars, this is what we have learned so far, this is what we don’t know, this is where it all might be going. You’re a part of this now. Welcome.”
Aside from starting from a low baseline and adopting good mental health habits, I'd be interested to know how much of the 5–10x happiness multiplier Will would attribute to his professional success and the growth of the EA movement. Is that stuff all counteracted by the hedonic treadmill?