It seems our biological neural networks, probably, aren't all that different from the ones we write. In their 'scope' they basically do just enough, for us to function. Beyond that, evolution probably started to optimise other things.

"Just enough" would mean being able to agree upon whether the object being seen was a danger (say a wild animal, an enemy tribe) or not (a friendly tribe). This goes deep enough for us to agree upon facial expressions and the like.

Interestingly, the 'lack of agreement' would theoretically be impossible to detect, because for consensus, we would achieve "eventual consistency" of conclusions - thereby reforming the neural pathways of all individuals.