The usual thing that I’ve seen happen in the case of recusals is that the recused person can no longer bring their expertise to the table, and de-facto when a fund-member is recused from a grant, without someone else having the expertise to evaluate the grant, it is much less likely for that grant to happen. This means two things:
1. Projects are now punished for establishing relationships with grantmakers and working together with grantmakers
2. Grantmakers are punished for establishing relationships with organizations and projects they are excited about
3. Funds can no longer leverage the expertise of the people with the most relevant context
In general when someone is recused they seem to no longer argue for why a grant is important, and in a hit-based view a lot of the time the people who have positive models for why a grant is important are also most likely to have a social network that is strongly connected to the grant in question.
I don’t expect a loosely connected committee like the LTFF or other EA Funds to successfully extract that information from the relevant fund-member, and so a conservative COI policy will reliably fail to make the most valuable grants. Maybe an organization in which people had the time to spend hundreds of hours talking to each other can afford to just have someone with expertise recuse themselves and then try to download their models of why a grant is promising and evaluate it themselves independently, but the LTFF (and I expect other EA Funds) do not have that luxury. I have not seen a group of people navigate this successfully and de-facto I am very confident that a process that relies heavily on recusals will just tend to fail to make grants when the fund-member with the most relevant expertise is excused.
have fewer grant evaluators per grant
Having fewer grant evaluators per grant is a choice that Open Phil made that the EA Funds can also make, I don’t see how that is an external constraint. It is at least partially a result of trusting in the hit-based giving view that generates a lot of my intuitions around recusals. Nothing is stopping the EA Funds from having fewer grant evaluators per grant (and de-facto most grants are only investigated by a single person on a fund team, with the rest just providing basic oversight, which is why recusals are so costly, because frequently only a single fund member even has the requisite skills and expertise necessary to investigate a grant in a reasonable amount of time).
and most of their grants fall outside the EA community such that COIs are less common.
While most grants fall outside of the EA community, many if not most of the grant investigators will still have COIs with the organizations they are evaluating, because that is where they will extend their social network. So the people who work at GiveWell tend to have closer social ties to organizations working in that space (often having been hired from that space), the people working on biorisk will have social ties to the existing pandemic prevention space, etc. I do think that overall Open Phil’s work is somewhat less likely to hit on COIs but not that much. I also overall trust Open Phil’s judgement a lot more in domains where they are socially embedded in the relevant network, and I think Open Phil also thinks that, and puts a lot of emphasis of understanding the specific social constraints and hierarchies in the fields they are making grants in. Again, a recusal-heavy COI policy would create really bad incentives on grantmakers here, and isolate the fund from many of the most important sources of expertise.
The usual thing that I’ve seen happen in the case of recusals is that the recused person can no longer bring their expertise to the table, and de-facto when a fund-member is recused from a grant, without someone else having the expertise to evaluate the grant, it is much less likely for that grant to happen. This means two things:
1. Projects are now punished for establishing relationships with grantmakers and working together with grantmakers
2. Grantmakers are punished for establishing relationships with organizations and projects they are excited about
3. Funds can no longer leverage the expertise of the people with the most relevant context
In general when someone is recused they seem to no longer argue for why a grant is important, and in a hit-based view a lot of the time the people who have positive models for why a grant is important are also most likely to have a social network that is strongly connected to the grant in question.
I don’t expect a loosely connected committee like the LTFF or other EA Funds to successfully extract that information from the relevant fund-member, and so a conservative COI policy will reliably fail to make the most valuable grants. Maybe an organization in which people had the time to spend hundreds of hours talking to each other can afford to just have someone with expertise recuse themselves and then try to download their models of why a grant is promising and evaluate it themselves independently, but the LTFF (and I expect other EA Funds) do not have that luxury. I have not seen a group of people navigate this successfully and de-facto I am very confident that a process that relies heavily on recusals will just tend to fail to make grants when the fund-member with the most relevant expertise is excused.
Having fewer grant evaluators per grant is a choice that Open Phil made that the EA Funds can also make, I don’t see how that is an external constraint. It is at least partially a result of trusting in the hit-based giving view that generates a lot of my intuitions around recusals. Nothing is stopping the EA Funds from having fewer grant evaluators per grant (and de-facto most grants are only investigated by a single person on a fund team, with the rest just providing basic oversight, which is why recusals are so costly, because frequently only a single fund member even has the requisite skills and expertise necessary to investigate a grant in a reasonable amount of time).
While most grants fall outside of the EA community, many if not most of the grant investigators will still have COIs with the organizations they are evaluating, because that is where they will extend their social network. So the people who work at GiveWell tend to have closer social ties to organizations working in that space (often having been hired from that space), the people working on biorisk will have social ties to the existing pandemic prevention space, etc. I do think that overall Open Phil’s work is somewhat less likely to hit on COIs but not that much. I also overall trust Open Phil’s judgement a lot more in domains where they are socially embedded in the relevant network, and I think Open Phil also thinks that, and puts a lot of emphasis of understanding the specific social constraints and hierarchies in the fields they are making grants in. Again, a recusal-heavy COI policy would create really bad incentives on grantmakers here, and isolate the fund from many of the most important sources of expertise.