One thing I want to add to my (old school) AI course next year is a segment on NLP through the lens of MDPs that brings back the classic example where you generate a model based on and generate nonsense text that sounds like that book.
Something that I feel strongly about is that AI---criticisms of the name notwithstanding---needs to be inclusive of its history. This replacement and erasure is also something we saw with the advent of deep learning, and I'm sure many other times.