Lately I've been staring at the Codex Seriphinianus quite a bit. Worth finding a copy if you haven't seen it before. From a technical point of view, The Elements of Statistical Learning, by Tibshirani, Friedman, and Hastie. Far and away the most illuminating demonstration that so many ML / AI techniques have a long-standing statistical foundation, and, essentially, everything boils down to the linear model.
Also recommended in this reply