Talk

There is now a lot of discussion on whether AI is a welfare subject and, relatedly, on whether AI has moral standing. But there is barely any research on the scale of AI welfare or the moral weight of AI welfare. Prima facie, this seems reasonable: how should we determine the scale or moral weight of AI welfare if we don’t even know if AI has any welfare at all? But this line of reasoning is mistaken: even though we are highly uncertain about whether AI has any welfare, we should already discuss its expected scale and moral weight. In fact, there is a case for prioritising the moral weight question. After making this argument, the paper distinguishes between different welfarist concepts based on which the scale of AI welfare can be assessed: welfare potential, welfare efficiency, and welfare alignment. How exactly the moral weight of AI welfare is determined by these welfarist properties depends crucially on how aggregative morality is. But either way, there are plausible scenarios in which AI welfare could be morally decisive in important decision contexts.