[skue]

A foal can stand shortly after birth. It doesn’t learn that from its environment. Say we train a quadruped AI, then what is our training simulating if not evolutionary development? It’s no different for an AGI and whatever analogue it ultimately has for our evolutionarily derived neocortex.

[dodslaser]

Walking and having feelings are two very different things though. Walking is simple enough that we can program a robot to do it without even having to resort to a NN. The concept of having feelings is so difficult that we barely understand how it works in humans, let alone how we could train a computer to have them.

[hombre_fatal]

Maybe emergent algorithms like biological evolution never “understand” the thing they have made, but the thing still has been made. The same might apply to our steps toward AGI.

We can’t even predict all the emergent patterns in current AI much less future, more complex systems.

Maybe there never has to be a time where someone sits down to create some sort of “feelings module” for something akin to feelings to emerge in a complex system. Same, presumably, even for sentience.