I think I'm tuning in to the frequency Kenneth

[Sal]

It's more than just computation, I think. Today, compared to true AI, the algorithms we use are deterministic; there's no "leeway". It's practically impossible to make "polymorphic" programs work as AI with current technology. ML, GANs, NLPs and whatever are just toy models of AI, they don't actually learn - they just use huge amounts of training data for their generalized function. For example, how do you train an NLP AI to distinguish - in the same way humans do - between a "border collie" and "poodle" belongs in the same group as "dog"? We're basically just using short-handed tricks to solve these things. They work, until they don't. At least I think so. Memristors might potentially change all that significantly - although I don't know well enough about them. That or quantum computers.