Embed Notice

@sam @smartgo
Anders: good blog post.

Sam: in a broad, hand-waving sense, of course everything you say is true. In a specific sense, we can say with some confidence that (1) these Go problems illustrate •intrinsic• limitations in example-based “deep learning” approaches in general, (2) AI is going to need to develop new approaches to overcome them, and (3) this “oops, we still haven’t solved the hard part” pattern has repeated throughout AI’s history.