Response to reviews by Feldman, Perlis, Taylor

I want to thank the AIJ and the three reviewers Jerome Feldman, John Taylor, and Don Perlis for reviewing my book, “On Intelligence”. Perlis’ review speaks of “fascination and frustration” with my book. He finds it fascinating because I “hint at a potentially unifying view” of intelligence and he finds it frustrating because the book is “sketchy” and he finds many of my claims about AI “questionable”. I think I understand Perlis’ frustration and will use his comments as an introduction to my response. “On Intelligence” begins with a critique of AI and ends with a description of the future of intelligent machines, so one might expect this book is about AI. However, “On Intelligence” is actually a book about neuroscience and the nature of intelligence. “On Intelligence” builds upon a premise first proposed twenty five years ago [1], that everything the neocortex does (vision, hearing, touch, motor planning, language, etc.), is done via a common algorithm. I deduce from this idea a framework for understanding what the cortex does (the memory-prediction framework), and I present a detailed proposal for how the architectural components of the neocortex and thalamus implement this framework (Chapter 6). If you are looking for a cookbook on how to implement this framework, i.e., how to build intelligent machines based on my theory, you won’t find it. At the time I wrote the book I was living in the world of neuroscience, developing a theory of cortex, and I did not yet have a mathematical formalism for the biological theory. (Although for some readers what I explained was sufficient for them to get started building systems based on my theory, more on this later.) What the book does offer, and I believe has never been done before, is provide a testable explanation for what the cortex does overall and how the detailed anatomy of the cortex fits within this explanation. Reactions to my book and the proposals in it have been mixed. Some people view the book as a collection of old ideas presented in a non-technical way, which at best might promote interest in this field but otherwise doesn’t move the field much forward. If I am allowed the liberty of painting the three reviews in this journal with a single brush I would put them all in this camp. On the other hand, many people claim that “On Intelligence”