At the same time it seems to ignore the automatic absence of certain human features that are clearly net-negative for the human when doing "level" comparison with the bots.
I am thinking of things like: need for sleep or rest, need to eat, need to tend to relations and activities outside of work etc.
When he calls it "human-level" he is ignoring this. Perhaps intentionally.
Imagine the same problem that happened with computation. One human can compute, but slowly, now this can be scaled up by multiple human computers, doing organized task of computing - but even the best large scale human computing efforts can be simply replaced by a single electronic computer.