That part made sense to me, but I don't see what bearing it has on the economics of automation. Any unemployment from automation has to be preceded by a large gain in productivity, so large you can do the same jobs with fewer people (or no people at all). E.G like shoemakers being put out of work by shoe factories in the industrial revolution.
If there's a modest gain in productivity from AI then it can't cause mass unemployment, since doing the same labor with fewer people won't be possible to any great extent. This is what I mean when I say he's contradicting himself.
