> The good news is, it makes replacing humans - trait for trait - far more more complicated.

I wouldn't take much comfort in that. You can probably reach something that resembles a human by just providing an alternative set of signals. We already do this with the reward mechanism during training and that is proving very effective.