I’ve been thinking about different cause areas recently, especially AI Safety, and one question that pops up is “What level of impact can the average worker have on a scientific field?” (I.e, someone who expects to have 50th percentile productivity in <field> among all the people currently working on it, if they were to specialise in this field.)”
Does EA have a model for this? I am imagining a simplified model could consist of:
- Effect on the field from adding more people. If you take a field with 1,000 people in it and double it, you probably won’t get double the impact as you previously had, since you now have more communication required and a lot of the best research is already being worked on by the first 1,000. I’d expect the increased impact to be somewhere between 1 and 2 times the start. (Is this true? Sometimes specialisation actually makes N people more than N times as productive! My intuition tells me this only happens in small groups, but I could be wrong)
- Effect on the field from an average worker. Some people are far more productive than others, and contribute 10-100x (higher?) more to the field. That said, the average worker in these fields still has a contribution to make. If you take the entire field’s number of workers as N, the mean impact is 1/N of the entire field, but the median impact is lower than this—so the average worker would contribute (median / mean) / N of the total scientific research.
So, you could then combine these two things—your impact on joining a scientific field as an average performer would be the marginal difference in impact from N to N+1 people, multiplied by the average worker’s contribution (median/mean).
My intuition tells me AI Safety is small enough, and the problem major enough, that it’s going to be worth going into anyway if I think I can be a middling performer in the area—but it would definitely be nice to have some numbers to check!
[Question] Does EA have a model for scientific impact?
I’ve been thinking about different cause areas recently, especially AI Safety, and one question that pops up is “What level of impact can the average worker have on a scientific field?” (I.e, someone who expects to have 50th percentile productivity in <field> among all the people currently working on it, if they were to specialise in this field.)”
Does EA have a model for this? I am imagining a simplified model could consist of:
- Effect on the field from adding more people. If you take a field with 1,000 people in it and double it, you probably won’t get double the impact as you previously had, since you now have more communication required and a lot of the best research is already being worked on by the first 1,000. I’d expect the increased impact to be somewhere between 1 and 2 times the start. (Is this true? Sometimes specialisation actually makes N people more than N times as productive! My intuition tells me this only happens in small groups, but I could be wrong)
- Effect on the field from an average worker. Some people are far more productive than others, and contribute 10-100x (higher?) more to the field. That said, the average worker in these fields still has a contribution to make. If you take the entire field’s number of workers as N, the mean impact is 1/N of the entire field, but the median impact is lower than this—so the average worker would contribute (median / mean) / N of the total scientific research.
So, you could then combine these two things—your impact on joining a scientific field as an average performer would be the marginal difference in impact from N to N+1 people, multiplied by the average worker’s contribution (median/mean).
My intuition tells me AI Safety is small enough, and the problem major enough, that it’s going to be worth going into anyway if I think I can be a middling performer in the area—but it would definitely be nice to have some numbers to check!