Thanks for this. I do think there’s a bit of sloppiness in EA discussions about heavy-tailed distributions in general, and the specific question of differences in ex ante predictable job performance in particular. So it’s really good to see clearer work/thinking about this.
I have two high-level operationalization concerns here:
Whether performance is ex ante predictable seems to be a larger function of our predictive ability than of the world. As an extreme example of what I mean, if you take our world on November 7, 2016 and run high-fidelity simulations 1,000,000 times , I expect 1 million/1 million of those simulations to end up with Donald Trump winning the 2016 US presidential election. Similarly, with perfect predictive ability, I think the correlation between ex ante predicted work performance and ex post actual performance approach 1 (up to quantum) . This may seem like a minor technical point, but I think it’s important to be careful of the reasoning here when we ask whether claims are expected to generalize from domains with large and obvious track records and proxies (eg past paper citations to future paper citations) or even domains where the ex ante proxy may well have been defined ex post (Math Olympiad records to research mathematics) to domains of effective altruism where we’re interested in something like counterfactual/Shapley impact*.
There’s counterfactual credit assignment issues for pretty much everything EA is concerned with, whereas if you’re just interested in individual salaries or job performance in academia, a simple proxy like $s or citations is fine. Suppose Usain Bolt is 0.2 seconds slower at running 100 meters. Does anybody actually think this will result in huge differences in the popularity of sports, or percentage of economic output attributable to the “run really fast” fraction of the economy, never mind our probability of spreading utopia throughout the stars? But nonetheless Usain Bolt likely makes a lot more money, has a lot more prestige, etc than the 2nd/3rd fastest runners. Similarly, academics seem to worry constantly about getting “scooped” whereas they rarely worry about scooping others, so a small edge in intelligence or connections or whatever can be leveraged to a huge difference in potential citations, while being basically irrelevant to counterfactual impact. Whereas in EA research it matters a lot whether being “first” means you’re 5 years ahead of the next-best candidate or 5 days.
Griping aside, I think this is a great piece and I look forward to perusing it and giving more careful comments in the coming weeks!
*ETA: In contrast, if it’s the same variable(s) that we can use to ex ante predict a variety of good outcomes of work performance across domains, then we can be relatively more confident that this will generalize to EA notions. Eg, fundamental general mental ability, integrity, etc.
My super quick take is that 1. definitely sounds right and important to me, and I think it would have been good if we had discussed this more in the doc.
I think 2. points to the super important question (which I think we’ve mentioned somewhere under Further research) how typical performance/output metrics relate to what we ultimately care about in EA contexts, i.e. positive impact on well-being. At first glance I’d guess that sometimes these metrics ‘overstate’ heavy-tailedness of EA impact (for e.g. the reasons you mentioned), but sometimes they might also ‘understate’ them. For instance, the metrics might not ‘internalize’ all the effects on the world (e.g. ‘field building’ effects from early-stage efforts), or for some EA situations the ‘market’ may be even more winner-takes-most than usual (e.g. for some AI alignment efforts it only matters if you can influence DeepMind), or the ‘production function’ might have higher returns to talent than usual (e.g. perhaps founding a nonprofit or contributing valuable research to preparadigmatic fields is “extra hard” in a way not captured by standard metrics when compared to easier cases).
Thanks for this. I do think there’s a bit of sloppiness in EA discussions about heavy-tailed distributions in general, and the specific question of differences in ex ante predictable job performance in particular. So it’s really good to see clearer work/thinking about this.
I have two high-level operationalization concerns here:
Whether performance is ex ante predictable seems to be a larger function of our predictive ability than of the world. As an extreme example of what I mean, if you take our world on November 7, 2016 and run high-fidelity simulations 1,000,000 times , I expect 1 million/1 million of those simulations to end up with Donald Trump winning the 2016 US presidential election. Similarly, with perfect predictive ability, I think the correlation between ex ante predicted work performance and ex post actual performance approach 1 (up to quantum) . This may seem like a minor technical point, but I think it’s important to be careful of the reasoning here when we ask whether claims are expected to generalize from domains with large and obvious track records and proxies (eg past paper citations to future paper citations) or even domains where the ex ante proxy may well have been defined ex post (Math Olympiad records to research mathematics) to domains of effective altruism where we’re interested in something like counterfactual/Shapley impact*.
There’s counterfactual credit assignment issues for pretty much everything EA is concerned with, whereas if you’re just interested in individual salaries or job performance in academia, a simple proxy like $s or citations is fine. Suppose Usain Bolt is 0.2 seconds slower at running 100 meters. Does anybody actually think this will result in huge differences in the popularity of sports, or percentage of economic output attributable to the “run really fast” fraction of the economy, never mind our probability of spreading utopia throughout the stars? But nonetheless Usain Bolt likely makes a lot more money, has a lot more prestige, etc than the 2nd/3rd fastest runners. Similarly, academics seem to worry constantly about getting “scooped” whereas they rarely worry about scooping others, so a small edge in intelligence or connections or whatever can be leveraged to a huge difference in potential citations, while being basically irrelevant to counterfactual impact. Whereas in EA research it matters a lot whether being “first” means you’re 5 years ahead of the next-best candidate or 5 days.
Griping aside, I think this is a great piece and I look forward to perusing it and giving more careful comments in the coming weeks!
*ETA: In contrast, if it’s the same variable(s) that we can use to ex ante predict a variety of good outcomes of work performance across domains, then we can be relatively more confident that this will generalize to EA notions. Eg, fundamental general mental ability, integrity, etc.
Thanks for these points!
My super quick take is that 1. definitely sounds right and important to me, and I think it would have been good if we had discussed this more in the doc.
I think 2. points to the super important question (which I think we’ve mentioned somewhere under Further research) how typical performance/output metrics relate to what we ultimately care about in EA contexts, i.e. positive impact on well-being. At first glance I’d guess that sometimes these metrics ‘overstate’ heavy-tailedness of EA impact (for e.g. the reasons you mentioned), but sometimes they might also ‘understate’ them. For instance, the metrics might not ‘internalize’ all the effects on the world (e.g. ‘field building’ effects from early-stage efforts), or for some EA situations the ‘market’ may be even more winner-takes-most than usual (e.g. for some AI alignment efforts it only matters if you can influence DeepMind), or the ‘production function’ might have higher returns to talent than usual (e.g. perhaps founding a nonprofit or contributing valuable research to preparadigmatic fields is “extra hard” in a way not captured by standard metrics when compared to easier cases).