I think âhuman-levelâ is often a misleading benchmark for AI, because we already have AIs that are massively superhuman in some respects and substantially subhuman in others. I sometimes worry that this is leading people to make unwarranted assumptions of how closely future dangerous AIs will track humans in terms of what theyâre capable of. This is related to a different post Iâm writing, but maybe deserves its own separate treatment too.
A problem with a lot of AI thoughts I have is that Iâm not really in enough contact with the AI âmainstreamâ to know whatâs obvious to them or whatâs novel. Maybe âseriousâ AI people already donât say human-level, or apply a generous helping of âyou know what I meanâ when they do?
I think âhuman-levelâ is often a misleading benchmark for AI, because we already have AIs that are massively superhuman in some respects and substantially subhuman in others. I sometimes worry that this is leading people to make unwarranted assumptions of how closely future dangerous AIs will track humans in terms of what theyâre capable of. This is related to a different post Iâm writing, but maybe deserves its own separate treatment too.
A problem with a lot of AI thoughts I have is that Iâm not really in enough contact with the AI âmainstreamâ to know whatâs obvious to them or whatâs novel. Maybe âseriousâ AI people already donât say human-level, or apply a generous helping of âyou know what I meanâ when they do?
Google Doc draft: Stop focusing on âhuman-levelâ AI
Iâll ask specific people to comment and aim to publish in the next couple of weeks, but Iâm happy for any passers-by to offer their thoughts too.
This became When âhuman-levelâ is the wrong threshold for AI