And that's roughly what I meant by "self-directed." Imagine a concrete goal: efficient fusion power, for instance. It's easy to define, easy to establish succe

And that's roughly what I meant by "self-directed." Imagine a concrete goal: efficient fusion power, for instance. It's easy to define, easy to establish success metrics, and it's even easy to propose methods -- but success has eluded us for decades. A mind that could solve a problem like that would have to have intuition, lateral and parallel thinking, and creativity (or their machine equivalents; I'm open to an AI which might think utterly unlike a human). My point is that we don't know where such traits come from in humans, and thus have no idea how to even begin to attempt to replicate them in computers except in the crudest and most rule-based ways. Note: I would dearly like to be proven wrong, I'm just parroting things I've heard.