I feel NickLaing is encoding an implicit graph-theoretic belief that may not be factually accurate. The premise is that CoI opportunities fall with decentralization, but it may be the case that more diffuseness actually lead to problematic intermingling. I don’t have super good graph theory intuitions so I’m not making a claim about whether this is true, just that it’s a premise and that the truth value matters.
My graph-theoretic intuition is that it depends a lot of the distribution of opportunities. Because EAs tend to both fund and date other EAs, the COI increase / decrease probably depends to some extent on the relative size of the opportunity / recipient network.
My premise may well be wrong, but all I have heard to date is that the conflicts of interests aren’t that big a problem, not a clear argument that more diffuseness could make COI worse.
if we take an imaginary world where there is only one donor org and many donee organisations, within a small community like EA it seems almost impossible to avoid conflicts of interest in a high proportion of grants.
But I have low confidence in this, and would appreciate someone explaining arguments in favor of centralisation reducing potential for COI
I think Nick is suggesting that if we had Open Phil split into funders A and B (which were smaller than Open Phil), then A declining to fund an organization due to a COI concern would be somewhat less problematic because it could go to B instead. I’m not a graph theory person either, but it seems the risk of both A and B being conflicted out is lower.
I don’t think that’s a good reason to split Open Phil, although I do think some conflicts are so strong that Open Phil should forward those organizations to external reviewers for determination. For example, I think a strong conflict disqualifies all the subordinates of the disqualified person as well—eg I wouldn’t think it appropriate to evaluate the grant proposal of a family member of anyone in my chain of command.
I feel NickLaing is encoding an implicit graph-theoretic belief that may not be factually accurate. The premise is that CoI opportunities fall with decentralization, but it may be the case that more diffuseness actually lead to problematic intermingling. I don’t have super good graph theory intuitions so I’m not making a claim about whether this is true, just that it’s a premise and that the truth value matters.
My graph-theoretic intuition is that it depends a lot of the distribution of opportunities. Because EAs tend to both fund and date other EAs, the COI increase / decrease probably depends to some extent on the relative size of the opportunity / recipient network.
My premise may well be wrong, but all I have heard to date is that the conflicts of interests aren’t that big a problem, not a clear argument that more diffuseness could make COI worse.
if we take an imaginary world where there is only one donor org and many donee organisations, within a small community like EA it seems almost impossible to avoid conflicts of interest in a high proportion of grants.
But I have low confidence in this, and would appreciate someone explaining arguments in favor of centralisation reducing potential for COI
I think Nick is suggesting that if we had Open Phil split into funders A and B (which were smaller than Open Phil), then A declining to fund an organization due to a COI concern would be somewhat less problematic because it could go to B instead. I’m not a graph theory person either, but it seems the risk of both A and B being conflicted out is lower.
I don’t think that’s a good reason to split Open Phil, although I do think some conflicts are so strong that Open Phil should forward those organizations to external reviewers for determination. For example, I think a strong conflict disqualifies all the subordinates of the disqualified person as well—eg I wouldn’t think it appropriate to evaluate the grant proposal of a family member of anyone in my chain of command.
Correct: a treatment of this question that does not consider BATNAs or counterfactuals would be inaccurate.