Is the P(billionaire|effective altruist) number fairly calculated? It seems most of your model assumes EA billionaires are people who first become an EA, then a billionaire. You do mention this being important but it did not seem like it was integrated into the model. Perhaps it would make sense to instead make two categories and predict each one separately:
One category is billionaires that become EA later (like Moskowitz). Maybe here a base rate could be EA billionaires / all billionaires in the world. Then you can calculate how many more billionaires there will be until 2027 and get an estimate for how many of these will later decide to donate to EA.
Another refinement here could be to get a sense about how many of current billionaires have heard of EA—maybe outreach here is poor and it might be that current non-EA donating billionaires start donating in the future (like Musk, although he has head about EA).
Another is EAs that become billionaires. Not sure who that currently is, but would have been SBF if that didn’t go so badly. I would perhaps even try to use the number of E2G EAs, and not the overall number of EAs to calculate the base rate here.
Related to point 1, a, ii, above—I feel like using the Ivy rate introduces bias. At least at UPenn where I went, a seemingly very high proportion of graduates tried to get rich. I feel like this proportion might be in the 20%-70% range. I think this differs from EA where I feel like the number of people really trying to get rich is probably closer to 5%-20% (maybe e.g. the EA survey has ways to find out). Also if the proportion of Ivy League graduates who become billionaires is significantly higher than graduates from all universities, perhaps this is not applicable to EA? Not sure what the makeup in terms of academic credentials is in the 9500 EA number you use, but it might be reasonable to be less optimistic about EAs ability to become billionaires (one anecdotal reason from my experience is that it seems the network one builds at an Ivy League is a major factor for billionaire success).
A last, and I think minor point is the timeline. For the number of future EA billionaires you expect to come from future EAs becoming billionaires, you might want to take into account the time it takes from becoming an EA, through to deciding to E2G and then finally starting executing on a plan to become rich. That could take several years meaning your number might be a bit optimistic for 2027. But then it might be a good estimate for ~2035. Ways to get info on this parameter could be to look at CrunchBase and look at time from incorporation to billion dollar valuation. And then add a few years before incorporation before that.
That said, I find your analysis super helpful. I am using it to get a sense of the likelihood of financing a bioweapons shelter/refuge in the next few years and after reading your analysis I became much more enthusiastic about this being possible (I previously used the expected value of Founder’s Pledge and was quite pessimistic about significant funding being available 5-10 years from now) . So thanks a ton for doing this analysis and posting it!
I’m not sure what you mean by fair exactly, but you’re right that I don’t distinguish the billionaire → EA pipeline from the EA → billionaire pipeline in the model (only mentioning it in text). It seems possible that your proposal of splitting these is good, though that may make it harder to calculate reasonable base rates (are there even any examples of EA → billionaire people left post-FTX?). Numbers on earning-to-give EAs could definitely be useful here if you have them.
Well, you’re probably right that there are reasons why we might expect the Ivy League base rate to be skewed high when using it for EA (the average age of Ivy League alumni being much higher than that of EAs is the most obvious one IMO), but also, the Ivy League base rate is actually substantially lower than the EA base rate (at least it was pre-FTX), and so including it has the effect of reducing the overall estimate, and if you didn’t include it you’d increase the overall estimate, which seems like the opposite of what you’re arguing for? It’s definitely possible that EAs are less likely to become and/or be billionaires even than the average Ivy League alumnus/a, but in that case it’s pretty suspicious that there have been several EA billionaires already.
Yeah this is true—there’s definitely some sort of diffusion/momentum thing going on that my model isn’t fully accounting for. (Then again, I think there’s something to be said for “simple model + qualitative reasoning around it” over “complicated model that accounts for everything”!) I guess when I calculate base rates I do only include actual billionaires, not about-to-become billionaires, so it shouldn’t be that biased.
Thanks for responding. Take only what you think is useful from my comments—you have thought much more deeply about this than I have and seem on top of the issues I have raised. Just a couple of responses in case it might be helpful (otherwise please disregard them):
Sorry, I have not seen such numbers. Just thought perhaps there might be some numbers lying around somewhere, e.g. results from surveys. I actually think perhaps the best number would be E2G EAs that pursue for-profit entrepreneurship—not sure if even the quant traders have a high probability of becoming billionaire donors. But this number might be even harder to come by.
I think I would not exclude the Ivy League base rate. Instead some possibilities could be (and please disregard this if it does not seem promising—I have not thought deeply about it!):
Perhaps one path could be to actually discard the EA base rate. My intuition here is that the number of EAs who later become billionaires is so low that the base rate calculated from it does not carry much weight (not sure if statistical significance is the right term here, and if not something close to it). Instead one could use an adjusted Ivy League base rate. And adjusting it based on some assumptions about “strength of talent”, fraction of population that pursues becoming rich and maybe some other adjustments, which would lower the final estimate.
Alternatively keep both base rates but still adjust the Ivy League base rate downwards due to the observations I made. That should also lower the final estimate.
Your point of having a simple model is a good one—I am not sure how much more accurate the forecast would be by making a more complex model. And I think you point out well in the post that one should not lean too heavily on the model but take into consideration other sources of evidence.
Some perhaps naïve observations/questions:
Is the P(billionaire|effective altruist) number fairly calculated? It seems most of your model assumes EA billionaires are people who first become an EA, then a billionaire. You do mention this being important but it did not seem like it was integrated into the model. Perhaps it would make sense to instead make two categories and predict each one separately:
One category is billionaires that become EA later (like Moskowitz). Maybe here a base rate could be EA billionaires / all billionaires in the world. Then you can calculate how many more billionaires there will be until 2027 and get an estimate for how many of these will later decide to donate to EA.
Another refinement here could be to get a sense about how many of current billionaires have heard of EA—maybe outreach here is poor and it might be that current non-EA donating billionaires start donating in the future (like Musk, although he has head about EA).
Another is EAs that become billionaires. Not sure who that currently is, but would have been SBF if that didn’t go so badly. I would perhaps even try to use the number of E2G EAs, and not the overall number of EAs to calculate the base rate here.
Related to point 1, a, ii, above—I feel like using the Ivy rate introduces bias. At least at UPenn where I went, a seemingly very high proportion of graduates tried to get rich. I feel like this proportion might be in the 20%-70% range. I think this differs from EA where I feel like the number of people really trying to get rich is probably closer to 5%-20% (maybe e.g. the EA survey has ways to find out). Also if the proportion of Ivy League graduates who become billionaires is significantly higher than graduates from all universities, perhaps this is not applicable to EA? Not sure what the makeup in terms of academic credentials is in the 9500 EA number you use, but it might be reasonable to be less optimistic about EAs ability to become billionaires (one anecdotal reason from my experience is that it seems the network one builds at an Ivy League is a major factor for billionaire success).
A last, and I think minor point is the timeline. For the number of future EA billionaires you expect to come from future EAs becoming billionaires, you might want to take into account the time it takes from becoming an EA, through to deciding to E2G and then finally starting executing on a plan to become rich. That could take several years meaning your number might be a bit optimistic for 2027. But then it might be a good estimate for ~2035. Ways to get info on this parameter could be to look at CrunchBase and look at time from incorporation to billion dollar valuation. And then add a few years before incorporation before that.
That said, I find your analysis super helpful. I am using it to get a sense of the likelihood of financing a bioweapons shelter/refuge in the next few years and after reading your analysis I became much more enthusiastic about this being possible (I previously used the expected value of Founder’s Pledge and was quite pessimistic about significant funding being available 5-10 years from now) . So thanks a ton for doing this analysis and posting it!
Thanks, I’m glad you found it useful!
I’m not sure what you mean by fair exactly, but you’re right that I don’t distinguish the billionaire → EA pipeline from the EA → billionaire pipeline in the model (only mentioning it in text). It seems possible that your proposal of splitting these is good, though that may make it harder to calculate reasonable base rates (are there even any examples of EA → billionaire people left post-FTX?). Numbers on earning-to-give EAs could definitely be useful here if you have them.
Well, you’re probably right that there are reasons why we might expect the Ivy League base rate to be skewed high when using it for EA (the average age of Ivy League alumni being much higher than that of EAs is the most obvious one IMO), but also, the Ivy League base rate is actually substantially lower than the EA base rate (at least it was pre-FTX), and so including it has the effect of reducing the overall estimate, and if you didn’t include it you’d increase the overall estimate, which seems like the opposite of what you’re arguing for? It’s definitely possible that EAs are less likely to become and/or be billionaires even than the average Ivy League alumnus/a, but in that case it’s pretty suspicious that there have been several EA billionaires already.
Yeah this is true—there’s definitely some sort of diffusion/momentum thing going on that my model isn’t fully accounting for. (Then again, I think there’s something to be said for “simple model + qualitative reasoning around it” over “complicated model that accounts for everything”!) I guess when I calculate base rates I do only include actual billionaires, not about-to-become billionaires, so it shouldn’t be that biased.
Thanks for responding. Take only what you think is useful from my comments—you have thought much more deeply about this than I have and seem on top of the issues I have raised. Just a couple of responses in case it might be helpful (otherwise please disregard them):
Sorry, I have not seen such numbers. Just thought perhaps there might be some numbers lying around somewhere, e.g. results from surveys. I actually think perhaps the best number would be E2G EAs that pursue for-profit entrepreneurship—not sure if even the quant traders have a high probability of becoming billionaire donors. But this number might be even harder to come by.
I think I would not exclude the Ivy League base rate. Instead some possibilities could be (and please disregard this if it does not seem promising—I have not thought deeply about it!):
Perhaps one path could be to actually discard the EA base rate. My intuition here is that the number of EAs who later become billionaires is so low that the base rate calculated from it does not carry much weight (not sure if statistical significance is the right term here, and if not something close to it). Instead one could use an adjusted Ivy League base rate. And adjusting it based on some assumptions about “strength of talent”, fraction of population that pursues becoming rich and maybe some other adjustments, which would lower the final estimate.
Alternatively keep both base rates but still adjust the Ivy League base rate downwards due to the observations I made. That should also lower the final estimate.
Your point of having a simple model is a good one—I am not sure how much more accurate the forecast would be by making a more complex model. And I think you point out well in the post that one should not lean too heavily on the model but take into consideration other sources of evidence.