“AI will be able to figure out what humans want” (yes; obviously; this was never under dispute)
I think the problem is that how existing systems figure out what humans want doesn’t seem to do anything with your theory of why it supposed to be relatively easy. Therefore the theory’s prediction of alignment being relatively hard is also doesn’t have anything to do with reality.
I think the problem is that how existing systems figure out what humans want doesn’t seem to do anything with your theory of why it supposed to be relatively easy. Therefore the theory’s prediction of alignment being relatively hard is also doesn’t have anything to do with reality.