When he writes about AI, he's writing about "good old-fashioned AI" -- the kind that was supposed to be advanced by clever representations, recursive data struc

When he writes about AI, he's writing about "good old-fashioned AI" -- the kind that was supposed to be advanced by clever representations, recursive data structures, and layers of abstraction. Contrast "good old-fashioned AI" with "machine learning", where the representations are as minimalistic and low-level as possible, because the only important thing is that when you update them a few billion times the right results emerge statistically. I don't see it as a problem that this part of GEB is "dated", though, because I find GOFAI to be an interesting topic. It involves trying to think about how you think and formalize it, which rarely happens in modern AI. Imagine a world where brute-force alpha-beta search was just not good enough to beat humans at chess; a world where advances in chess-playing computers would require chessmasters to encode their expertise in interesting data structures. This is the world that might have been, and it's the world GEB describes when it talks about AI. It's dated, but it's interesting.