The Society of Mind Video Lectures

I'd say that the book started as a bunch of untried ideas for how to represent interesting processes that show up in cognition... and it's largely the same now.

It doesn't work very well if you take it as an instruction manual for how to make AI, especially because it describes many processes that we know no good way to compute.

But Minsky's writing represents the currently-hibernating field we call "good old-fashioned AI" (GOFAI). What makes it "good"? GOFAI contains ideas about what we wish AI could do that we've kind of abandoned in the present, not because we don't want to do them anymore, but because they're very hard, and there are easier ways to get results.

The current wave of machine learning techniques will probably give diminishing returns one day. Going back to a GOFAI text such as Society of Mind is a way to dream about what comes after that. In the extremely optimistic view, it might help you recognize an opportunity to do something new with AI when a new computational technique makes it possible.

After reading GEB this summer, I got hooked on ideas of AI and knowledge representation. Not an AI expert by any standard, but I am reading it right now and enjoying it quite a bit. As mentioned by others in this thread, the book itself is a set of loosely connected ideas. Seems like a good overview of the field of AI.

In addition, Minsky is a great writer, and brings all sorts of examples of developmental psychology that are worth learning about.

And youtube, which is finally an iTunesU replacement now that it has background listening.