The big problem with this decision isn't that it's mistaken, it's that it's corrupt. He should have excluded things that were in his self-interest from consideration, even if he honestly, mistakenly believes they are the best thing anyone can do with their money.
But it would not be a criticism of the general idea of risk-return optimization applied to charity, and would not exclude a rational participant from belief that allocating some capital towards AI research was part of an optimized approach.
The original comment made it seem like the generic idea of choosing to invest in AI safety as part of the EA framework was, in spirit, intrinsically "morally repugnant."
The specific repugnant actions of one party would not necessarily support that, anymore than say a pro football player's racist comments would imply that all of football is inherently morally repugnant.
It would seem valuable to separate and distinguish vitriol directed at the specific suspicious actions of one person from generic and wide-sweeping criticism of an entire framework that person happens to be associated with, especially when the framework itself attempts to be value-neutral conditioned on one's beliefs after looking at some evidence.
The original comment was the one that brought up "morally repugnant" choices in investing. That's what brought in relative value judgments. I was trying to ask how it differs from simply disagree with someone else's assessment of the evidence.
If Person A evaluated the evidence and sincerely believe you should give $3 to X and $2 to Y, based on empirical outcome optimization, they are using the EA framework.
Person B might look at the same evidence and believe you should give $5 to X and $0 to Y, but that hardly makes Person A "morally repugnant" (which might be Person B's unrelated moral judgment) nor does it make their choices fall outside the scope of the EA framework of decision-making.
As far as people disagreeing with posterior distribution over the goal-maximizing choices after seeing the same evidence, this would ideally fall under something like Aumann's Agreement Theorem (of course with the exception that people are not fully rational, Bayesian agents).
Given that, there is some expectation that if both parties are really rational and have the same value of goals, then they ought to come to the same conclusions given the same evidence.
But if they don't have the same goals (e.g. maybe you just happen to care more about animal welfare than me), then the EA framework doesn't say anything about you and I having the same investment priorities.
Really, popular EA press is mostly just saying, "Look, we take value judgment X on issues A, B, and C. If you agree and you start from the same value judgment we do, then based on the following evidence, we believe it's optimal to invest in foo, etc."
It seems like you're saying that it's not possible for two different people, *both making decisions in the effective altruism framework" could come to different beliefs about what to invest money into.
But there's no part of EA that requires that for two agents with different value judgments at the start.