A Thousand Brains: A New Theory of Intelligence presents a brilliant and tantalizing theory on how our brains work and what the future may hold. Yet, this book is heavy in opinion and falls short in some key areas. Hawkins starts part 1 of his book with the introduction of the old brain / new brain dichotomy. Brains develop additively, and the mammalian neocortex was added at a much later time in our evolutionary history than our reptilian old brain, which is responsible for our emotions, drives and our survival and reproduction drives. Furthermore, the neocortex and the old brain continuously negotiate for control over the host body — old brain overrides the neocortex by flooding it with hormones (danger, there is a lion over there, run), and the neocortex halts the old brain on its tracks by suppressing its instinctive drives (hold on there, don’t eat that cake, not good for you). I find this framing to be brilliant. Would not it be amazing if the old and new brain could just get along and allow us to just do the right thing all the time? I digress. The neocortex is what endows us (and all mammals) with intelligence. Unlike the old brain, which has specialized components, the neocortex is uniform and comprises of cortical columns. The neocortex learns to navigate the physical (and metaphysical) world by creating models of the world that Hawkins terms reference frames. These reference frames are encoded in the place and grid cells of the cortical columns. This, Hawkins surmises, explains a number of emergent human behavior such as false beliefs and consciousness. Our sense of self is an inevitable consequence of this intelligence and arises from the need to create a model of ourselves and others. I completely agree with Hawkins’ conclusion that this model of self, ie consciousness, exists in all mammals to varying degrees. So far, so wonderful. What I found less satisfying in part 1 is the significance of understanding precisely how our cortical columns store information. How information is stored is likely to be an emergent behavior of the structure and make of the evolved brain without real insight into the design of the overall system or why it works at all. It is akin to trying to figure out how the weights and biases of a neural network store the various aspects of image features in an image classification system. A deeper understanding to me seems to be to figure out what the structure of the neural network is and how it processes information, the number of hidden layers, etc. To truly illustrate the structure of the human brain, one needs to do experiments that are currently beyond our capability. Therein is where part 1 is weakest. This book is short on experiments and large on personal convictions this or that has to be true. Parts 2 and 3 of the book focus on the consequence of intelligence, the prospects for an artificial general intelligence (aka machine intelligence), and how we can ensure human knowledge survives into the distant future. I agree that for now our best hope is to model machine intelligence based on human intelligence. Yet, Hawkins’ claim that machine intelligence could only be developed by modeling the neocortex on a computer to be unfounded. Perhaps more bizarre is Hawkins out of hand dismissal of the risks posed by machine intelligence. Here I have multiple objections. Let’s mention two. First, in apparent self contradiction, Hawkins reminisces our species breaking free from the chains put on us by our selfish genes. Yet, he sees no issue with stating a machine intelligence that is potentially a million times more intelligent than us would never overcome the chains placed on it by its human overlords. Second, Hawkins posits a machine intelligence that is able to learn a million times faster than us would still be unable to overtake us because ultimately all real knowledge requires experimentation that can not be speeded up (for example, rats in a lab). While he acknowledges theoretical fields such as mathematics could be learned thousands or perhaps millions of times faster, he reassures us that there is nothing to worry about. I find this bizarre. Could you imagine arriving at general relativity without calculus? Machine intelligence could discover new branches of mathematics that, far example, could easily unify all of the forces of nature. Perhaps provide a formulation for time travel. What if they discover something about the universe that human intelligence is simply unable to grasp — akin to explaining the photoelectric effect to chimpanzees or worse ants. A better understanding of nature could mean new theories that are unfathomable to human intelligence are discovered more easily by machines. How things would evolve from there would be completely out of our control. The book ends on a chapter about persevering our knowledge. I find the goal of preserving human knowledge to be a meaningless endeavor in and of itself. For one thing, we didn’t invent knowledge, thus there is no reason to “preserve” it as if it would vanish if we were to go extinct. Nature knows how it works. Atoms are the embodiment of the thing we call knowledge. Space, time, and energy know exactly how to behave, and they see no conflict between quantum mechanics and general relativity. My point is there simply is no reason to preserve knowledge were it not ultimately connected to the interests of our old brain: our survival. All in all, I recommend this book for anyone interested in a neat theory about intelligence.