Earlier quoted context omitted.
Isn’t this just a statement that you disagree either with the relative probability of an AI-related disaster or the relative degree of harm it could cause? It seems harsh to say it’s “morally repugnant” when in the end you’re just saying the way you assign probabilities and degrees of negative outcomes would lead you to invest in a different portfolio of charities or efforts than what this other group would do. (I’m…
IMO, the problem with AI risk funding isn't that it is 'wrong', per se. Instead it is that it is very speculative and not grounder in real world data, or measurable benefits that can be achieved TODAY. The whole point of effective altruism is to be very ground in evidence based causes. IE, if I spend 10K on this cause, then it will definitely save X lived by next year. AI risks can't be reduced to these kinds of numb…
Then in the optimization problem of determining what to invest donations into, the amount of gain from putting money into AI research would be correspondingly penalized by the amount of uncertainty (this is known as mean-variance optimization).
It's very much like a choice to invest in a solid, fundamental company which you know will give you 5% return next quarter, or invest in a very uncertain start-up which might give you 1000% return next quarter or might give you -50%.
The start-up investment might have a higher expected value, but also a much higher risk. At that point, the way to distinguish between whether you prefer to invest in the "sure bet" company or the "risky and ambiguous" start-up would be down to your personal tolerance for risk.
It could be perfectly rational to invest in the start-up in that situation. If you personally have a high risk tolerance. If you don't, then the start-up would look like a crazy, speculative bet with huge downside.
So some people might look at AI and say, "We don't have a good idea what the capabilities will be in ~50 years. So there is a huge risk that my charity donations will be wasted because in 50 years we realize we never needed to worry about this problem. Or we might use current AI research to thwart an unimaginable global crisis."
That person then looks into various sources of thought on putting actual numbers and actual error bars, the best we can, onto the problem, say by reading the Global Catastrophic Risk book, or stuff by Bostrom, or consulting surveys of current practitioners' estimates.
At the end of that process, it could totally end up being the rational course of action for that person to still invest in AI charities. Yes, their estimate for the expected value of the donation might have huge error bars around it, but under their particular beliefs about risk-return trade-offs, that might not cause their optimized decision to change.