“ Truth from Trash . How Learning Makes Sense ”

This book is a risky bet. It is difficult to try to make sense from the so complex field that machine learning (ML) is today. It would be even harder if this were to be done in an informal way, avoiding mathematics as far as possible, and in scarcely more than 200 pages. And finally, to make the book amusing and enjoyable seems to me a bet on a losing horse. Even worse when the horse is made from strewn and incomplete pieces: prediction, supervised learning, nearest neighbors, Kepler’s Laws, redundancy, clustering, decision-tree learning, encryption, Bletchley Park, similarity, relational learning, philosophy of induction, compression, Occam’s Razor and creativity. But things are not so predictable as they look at first glance, and a good rider can come up to us as a bolt from the blue. Chris Thornton has been able to combine these a priori eclectic elements in a coherent, enlightening and easy-to-read canter. The journey runs from a short reference to the disappointment of robotics, which have been unable to endow robots with the ability to learn, to the firm rebuttal of mystical and pseudo-scientific arguments against the possibility of intelligence and creativity in machines. As the title pun suggests, the book focuses on clarifying the reasons and implications of two issues which are not well realized by ML beginners: how difficult learning is, and how ubiquitous it is for constructing our view of the world. And the tool for this clarification is the illustration of the most successful techniques and paradigms from ML. They are usually being preceded by either historical or philosophical background and followed by their applicability and limitations. In this way, the author intertwines some literary passages in form of cunning dialogues, historical anecdotes and even sly games with the reader, and lightened but still substantial technical material. Chapter 1 is apparently bait to draw in the reader. It introduces an imag-