Thank you Elliott for another well-thought response!

I agree with you, machine learning isn't exhausted. My point is that it'll be before we achieve AGI. It's reasonable to argue that intelligence arises from connectivity in neural networks, but it may not be the only necessary element. That's what those working at embodied AI think.

The thing is, we may not be able to reproduce adequately by design what evolution did throughout millennia. If we could perfectly copy the brain, then it's fine (although the body may still be necessary to develop the brain's connections). However, I'd argue Turing was right in his approach. It's better to build a child's brain and then educate it than trying to build directly a fully-developed adult's brain.

I read the other day a "debate" on Twitter between Yann LeCun and Gary Marcus exactly about this. Marcus argued that machine learning-based AI could never extrapolate, only interpolate from the training data. Humans can extrapolate and we don't know how to close that gap in AI.

What do you think? :)

Writing at the intersection of AI, philosophy, and the cognitive sciences | | Get my articles for free: