Leslie Valiant: Probably Approximately Correct: Nature's Algorithms for Learning and Prospering in a Complex World.
Thu, Apr 12, 2018./noise/books | LinkKindle. Valiant's theoretical basis for machine learning is far more real-world plausible than the logical accounts of the 1960s (the learning-in-the-limit model of Gold and Blum) and here he posits it as one of the missing links in Darwin's account of evolution amongst many other things. I took extensive notes as I went but lack the time to write them up; now I wonder where I can find the debate this book must have caused since its publication in 2013. It seems unlikely that his neologism ecorithms has stuck.
Edward Frenkel reviewed it for the New York Times. Marcus Feldman points out some of Valiant's blind spots. Ernest Davis is also skeptical: he observes the lack of a story about theoretical terms (which Davis calls "higher order constructs") and that PAC does not exhaust all forms of learning. It strikes me that ID3 neatly spans information-theoretic and computational readings of learning processes.