How a New Understanding of the Brain Will Lead to the Creation of Truly Intelligent Machines

I am not convinced that Hawkins has "the right answer" to how intelligence works or "the best strategy" to developing intelligent machines; I simply don't think we know either of those things yet. But he lays out his ideas in clear, readable prose and leaves a moderately informed, intelligent reader clear on what he is asserting, why he is saying it, and what the implications will be if he's right. Given the complexity of the field that is an achievement in and of itself. And he may turn out to be right as well, which would be a big bonus. :-). If you want to know the state of the search for how intelligence works and how intelligent machines might come about, this only one of the books you'll need to read, but it is certainly on that list.

Hawkins has created an electrical engineer's view of how the human cortex creates patterns from sensory input, stores the information, and uses patterns predictively. The book is an easy read and very entertaining. A recently published peer-reviewed paper by the author (2009) contains the messy details and should cause a stir in the artificial intelligence community if nothing else.

In his last book, Ray Kurzweil explain his own theory of the brain, regognizing that it's mostly based in Hawkins ideas, (with some add-ons based on his own experience in pattern recognition). Kurzweil has recently been hired as director of engineering at Google, to lead some of the most ambitious AI projects of the new century.
Recently, Elon Musk (the millionary founder behind Tesla Motors, SolarCity, SpaceX and Paypal) and Mark Zuckerberg invested 40$ million in Vicarious (according to WSJ), an AI company aiming to replicate the human neocortex as computer code. Their bet and line of investigation is mainly a modification based in Hawkins vision and works.
Just two capsules of information that gives an indication of at wich point the ideas and theories of Jeff Hawkins, summarized in this book, have inspired many of the big players in the new renaissance the AI field is experiencing.

Very, very interesting! Now you can understand much better how our brain may work and why decisions can be made so quickly!
This ideas fit to the independent observations and research you will find in this book: Sources of Power: How People Make Decisions by Gary A. Klein.

I have always been curious about how the brain works, and why it can do things that computers that work a million times faster cannot begin to achieve. I thought this book might be too difficult for me because I am not a scientist, but it was not compicated. It tackles what I used to think must be an infinitely complex subject in a simple and interesting way for the average person, through great writing, common sense explanations and simple analogies. They do an excellent job of avoiding getting into the details that confuse and bore non-scientists. I started reading it out of curiosity, and then I could not put it down. If you have a brain, hope to have a brain someday, or have any interest at all in how brains work, you should read this book!

Jeff Hawkins, an academic outsider, offers a fresh perspective on how the brain works in an attempt to build intelligent machines. He's since started a new company, Numenta, where you can download a Hierarchical Temporal Memory toolkit to try out some of his handywork and test his ideas. Written for anyone to read.

As a professional with an extensive resume in engineering, education, and publication, the Hawkins/Blakeslee endeavor provides a seminal introduction into Cognitive Science. It will be a classic text in the years to come and will appear in many libraries both paper and electronic. Having read it twice I continue to find new, stimulating and thought provoking insights in my understanding of what it means to be human.

Hawkins helps you see how a brain might function, by first contrasting it to the often overused analogy of a computer, or a binary machine. He helps you realize that, while a synapse may be mimicked by an on/off switch, the integration of information of switches (as in a computer) is fundamentally different than that which is done in a brain. He goes on to help you imagine how a brain might function. Neurobiology is a fast-moving science, and developments since this book was published reveals the book to be a useful early insight, and thus also serves as a great introduction. It is also an introduction from an entirely different direction than that which might typically come from neurobiology.

In the field of artificial intelligence it seems there are as many definitions of intelligence as there are stars in the heavens. Each of these definitions seems plausible, and interestingly, they seem to get more difficult to satisfy with time. Thus progress in artificial intelligence seems to be non-existent, since the criteria used to designate a machine as being intelligent ten years ago are no longer used today. Researchers in AI used to believe for example that if a machine could beat a human in chess then it should definitely be deemed intelligent. That belief is hardly held by anyone in the AI community at the present time.

The author of this book proposes yet another definition of intelligence, and it is one that is inspired by his understanding of how the human brain functions. His justifications are interesting, but they are highly speculative, and border on mere philosophical musings. It would have been a better book if the author refrained from the random walks in conceptual space that are characteristic of philosophy, and justified his conception of intelligence with what is really currently known in neuroscience. He does quote the research of neuroscientists that have produced a detailed map of the monkey cortex, which revealed many different regions connected together in a complex hierarchy. The author then makes the assumption that the human cortex hierarchy has a similar hierarchy. This is not really an unreasonable assumption if viewed from the standpoint of neuroanatomy, but from the standpoint of the cognitive abilities of humans versus those of monkeys, it might indeed be an assumption that deserves intense scrutiny.

The author definitely wants to view intelligence as being one that can function over many different domains, i.e. an intelligent machine will be able to not only play chess for example, but could also analyze stock market data or perform some other function typically thought of as requiring careful thought. He expresses this by saying that the human cortex is "universal" in that it can be applied to any type of sensory or motor system, and that the "algorithm of the cortex" can be expressed independently of any particular function or sense. Certainly humans can think in many different domains, but one cannot conclude from this that humans possess the general intelligence that the author believes they do. There is in fact a large body of research that indicates that the human brain has a modular structure (the author discusses this research very briefly), with each module being responsible for functioning in a particular domain. If one of these modules ceases to function, this has no effect on the functioning of the others. This is a view of the brain as having a domain-specific structure. A domain-general notion of intelligence would mean that the brain can deal with several different domains, but that the same reasoning patterns or processes are used to think in these different domains. If one of these reasoning patterns or processes becomes non-functional, the rest of them will suffer. One could still view the brain as consisting of modules expert in different domains, but that these modules are "entangled" with each other in the sense just specified, i.e. damage in one module will affect the others.

In fairness to the author, there is also research in neuroscience that lends support to his notion of general intelligence and a single algorithm that can deal with all of the data presented to the human brain. He gives a few references that discuss this research, and he definitely emphasizes the need for feedback and the related notion of `auto-associative' memories. The brain in his view is a "pattern machine" and if one is to construct truly intelligent machines one must make use of this pattern manipulating capability of the human cortex. Thus intelligent machines will be a result of this "neocortical inspired" computing, and the author spends a lot of time explaining why these machines will mimic the ability of the brain to solve a problem using memory, and not by computing a solution. The cortex, in his view, creates "invariant representations" which can handle the intricate variability of the world it is confronted with. He summarizes this viewpoint by saying that the neocortex stores sequences of patterns, recalls patterns auto-associatively, stores invariant patterns, and stores patterns in a hierarchy. His explanations of how it does this are interesting, but again are very speculative, and in the absence of a prototype for a machine that possesses this kind of intelligence, it is difficult to assess the validity of his assertions.

This reviewer strongly disagrees with the assertion from the author that there are no machines today that express true intelligence. A strong case can be made for the existence of myriads of intelligent machines in the world today, but this case would again be dependent on a particular definition of intelligence. Machines that have intelligence as the author defines it are nowhere in sight, and this is no doubt due to the lack of commercial value in the domain-general intelligence that the author advocates. The intelligent machines of today can learn, adapt, and manage, and do many other different things, but they only do these things in specific domains. There is absolutely no need for these machines to have expertise in more than one domain, both for the sake of efficiency and also because of economics. In managing a network for example, there is no need for a machine to have expertise in some other area, such as chess playing or backgammon. Business demands thus dictate the kind of domain-specific intelligence that is so prevalent in hundreds of intelligent machines performing many useful functions in business and industry.

I found the reviews and blurbs very intriguing, and once I had the book I didn't put it down until I had read it all.

This book does *not* have the kind of science-is-wonder attitude you might find in Sagan; it does not convey the message "Study science, it's a noble thing to do, your curiosity will be greatly rewarded", like some reviewers write. It's not the kind of book that summarizes the state of what's known about the brain today, while getting you excited about finding out more.

This book is really a monograph by someone who thinks they have literally figured out the brain, and it contains mostly what the author has to say. So you won't see a mention of the holographic theory of the brain, or the brain-as-dynamic-system view. In my opinion, in a book addressed to the general public, that's a problem.

It's a problem, because on one hand the book is written for the general public, while on the other hand it presents utterly untested (by anyone, even the author) hypotheses, mostly made by the author himself or hand-picked from existing research.

In this respect, it reminds one of "A New Kind of Science" by Stephen Wolfram, another "scientific" book which aims to directly impress the layman with something he's not likely to understand, while bypassing specialists.

Next, what annoyed me somewhat is the the fact that the subject, admittedly the biggest mystery of our time, is given such a simplistic treatment. No, we're no closer to creating something that can, say, semi-intelligently fly around the room and land on things, or since the author prefers non-behavioral intelligence, detect whether a given picture has a cat in it. So why such extraordinary claims of no results were obtained?

Finally, there is no evidence that the author or his colleagues have actually built any software or hardware that can detect any meaningful patterns in visual or audio streams.

Yet, I'd still recommend this book, because it highly readable, and it'll make you think (even if it's about the outrageous claims). Because of that it gets four stars.

Let me just start out this brief review by conceding that the author of "On Intelligence" is himself rather intelligent. With that concession behind me, let me make yet an additional concession: Notwithstanding my total lack of knowledge of neurobiology and the technicalities of Artificial "Intelligence" (I put "intelligence" in quotes, because from my perspective no machine has one iota of genuine intelligence), let me say that I was largely won over to thinking in terms of hierarchical strategies for neurobiology and for any truly successful machine that effectively simulates intelligence.

Let me, though, point out where I, as an amateur philosopher, take very great issue with much of the thesis of Hawkins' book. Over and over, almost nauseatingly, the author refers to the "brain knows this", or "a computer will understand what is going on", "the intelligent machine will be searching for the meaning", etc., etc. In other words, Hawkins takes a highly functionalist and materialist view of human consciousness. He seems very little bothered by the awesome implications (philosophically, ethically, and pragmatically) of consciousness being nothing more wondrous or mysterious than the complexities of a highly sophisticated algorithm. In other words, he treats it largely as a given that mind has no objective existence, that spirit is no more than an outdated superstition, and that a blind, purposeless, and meaningless universe generates conscious creatures via such fantastic algorithmic processes as to become highly conscious through the accidental processes of blind, purposeless physical forces.

There is not one shred of evidence that any computer can possibly be created by human beings such that the created device has even an iota of sentience. Until consciousness is a phenomenon that is understood, as well as how it is that consciousness comes into being, it's rather absurd to reduce the mysteries and glories of consciousness to a material algorithm. And if consciousness truly is spirit -- as many intelligent thinkers believe to be the case -- then humanity's mastery of any consciousness other than partial mastery of one's own consciousness is a hopeless quest. In reading Hawkins' book, one would get the impression that all such deep and profound questions are not even worth seriously discussing. Therefore, I consider myself generous to give the book 4 stars.

A plug for the book, though, is that it is engagingly written, and if you like to read about the science of what constitutes intelligence, how robots or other devices might become awesomely powerful tools for human use, etc., this book might well be a good buy.

Try Intelligence and the Brain: Solving the Mystery of Why People Differ in IQ and How a Child Can Be a Genius for a more complete story of the brain and intelligence. While Hawkins' book is excellent, Hawkins views intelligence as a certain way, and then shapes his theory to explain that view of intelligence. However, his view of intelligence is incomplete. For instance, how can a satisfactory account of human intelligence not mention abstraction?? Countless commentators and researchers from many different backgrounds have identified abstraction as being fundamental to intelligence and yet, Hawkins does not talk about abstraction at all in his book!

Hawkins also does not even mention IQ in his book, even though IQ tests are the most common measure of intelligence. Avoiding a discussion of IQ is not surprising, as IQ tests involve giving people novel problems to solve, and IQ shows at least partial heritability. These facets need to be explained for any serious account of human intelligence.

Jeff Hawkins presents a new perspective on how the brain processes information. The brain the neocortex in particular does not operate like a computer. Understanding how the neocortex works will help us develop machines which are truly inteliigent.I highly recommend this book .

A very good book for psychologists, neurologists, educators and philosophers but probably of little interest to most others. It is very informative but technical in nature and requires some backgound and interest in the nature of intelligence.

This book was released in 2005. Still there are only close to 250 ratings on this book. The ratings, if not excellent, are good but before deciding to read the book I was wondering why this book may not have gained enough momentum in ten years? It may be that people didn't read it at all, or those who did didn't review it because they couldn't really express what this book was about.

Readers may not have been able to understand the book because almost half of it talks about theory of brain system. As you go through that part, it feels like you are sitting in an algorithms class where the instructor tried to simplify the concepts but in the effort to do so, made it completely mundane and tiresome. The author hints on the fact that this part can be skipped but if you skip it, you will lose the continuity of what the book is trying to talk about, which bring us to the question - what is the book all about?

Well, the author starts with a conjecture that the book will be about why current state of Artificial Intelligence may not be able to design intelligent machines. For some time, the author talks about himself and then tries to explain the basic details of Neural Networks. If you already know about Neural Networks and Machine Learning, the part of the book would make sense. Then the author talks about how the brain works (possibly to explain how it is different from Neural Networks). It is at this point that the author goes into gory details of hierarchical structure, which although makes sense but seems irrelevant to the line of argument.
Nevertheless, by the time this argument ends, you are left with only one more chapter where author tries to condense the gist of the entire argument by talking about what the intelligent machines should look like, how it should be designed and what would it achieve. In between, there is some explanation about cautiousness and awareness which although the author explains clearly has no relevance to the topic in discussion.

In the end, you will be utterly confused about the book. What was it about - Neural Networks, Psychology, Biology, Author? Oh, the book was about how in its current state of Artificial Intelligence may not be able to do what a human brain does - it could probably be explained in 50 pages. Ironically only about 50 pages are devoted to that. Rest of the pages could have been blank as well!

As a neuroscientist, I am a little disheartened by the dearth of popular science explaining the brain. To be sure, there are a number of very good books that will tell you interesting facts about the brain, such as amazing stories about neurological patients who suffer bizarre disorders after lesions or strokes to localized brain regions. But as much as I appreciate these books, I think we are at a stage of our knowledge where we do understand enough about the brain's function to endeavour to explain it. It is a little sad that it took a non-neuroscientist to do this, but Hawkins' 'On Intelligence' is the best popular account that I have come across that really attempts to explain how the brain works. I should qualify that somewhat - this isn't a book about the brain, but only the neocortex, and as such only tells part of the story. But it is a very good popular account of what the neocortex does. That is, it learns internal models to predict its inputs. This is the essence of understanding, analysis-by-synthesis. Hawkins is not the first person to propose this, but nor is he just some writer popularizing others' ideas. On his website he has a paper he wrote as a graduate student in 1986 proposing this basic idea. In a way it was quite visionary, as the dominant view of the cortex was much different back then, and Hawkins' ideas about the role of prediction really presaged the next two decades of cortical theory. The classical view saw the cortex as merely efficiently encoding features of its environment, but the ideas of Hawkins and others (e.g. Mumford, Friston) have made it clear that the cortex does not merely extract features of the environment, but actively predicts them, allowing it to form inferences on the basis of Bayesian ideas (while Hawkins doesn't discuss Bayesian formulations in this book, the view presented here is very compatible with the Bayesian zeitgeist, and Hawkins has recently written papers where he attempts this synthesis).

This book isn't perfect. Hawkins isn't a neuroscientist, so doesn't really have the decades of experience that can fully animate a presentation like this. Instead there are a number of asides to discuss mobile computing. And when the book turns to topics like creativity, and consciousness, the book doesn't have much of insight to add. However the key contribution here is the book's cogent elucidation of what the neocortex does, and for this it is an invaluable resource to the non-specialist.

Finally, I should note that 'Phantoms in the brain' by Ramachandran and 'On Intelligence' by Hawkins are two of the best pop neuroscience books out there. What do they have in common? Sandra Blakeslee as coauthor - i'm sure that's not a coincidence.

In his professional life, Jeff Hawkins is a software engineer and a successful high-tech entrepreneur. Like some computer scientists, he is intrigued by the attempts and failings over the decades to develop computer-based artificial intelligence. This passion led him to pursue additional academic training and to put much thought and research into how the human mind thinks and how that process is different from a computer’s. In “On Intelligence”, Hawkins explains to the layman what is known about the human brain’s workings today, and also proposes his theory as to the primary mechanism by which the human mind performs the higher-level thought processes typically associated with intelligence.

Hawkins calls his theory the “memory-predictive framework”, and it certainly seemed compelling to this reader. The basic idea is that the behavior and organization of the neurons in the neocortex is such that they are continually looking for patterns of stimuli that they have been exposed to before as well as making predictions about what will come next. The neurons remember patterns and predict the next stimuli. Then, through the interconnected hierarchical organization of different layers of neurons, higher levels of intelligence are built up with the end result being a human brain that achieves self-consciousness and abstract thought. Hawkins builds his case with evidence from scientific studies of the brain’s physical structure as well as behavioral examples illustrating how his theory can explain, for example, how an athlete learns to catch a ball.

Not being a scientist who studies the human brain for a living, I can’t render an opinion as to how groundbreaking or not Hawkins’ theory is. But that debate is one for the academics. For laypeople with some scientific background and a desire to learn about how the human brain works, I can heartily recommend “On Intelligence”. I learned a lot, and what Hawkins said seemed to make a lot of sense. He also shows just how different the human mind is from computers, which is why we are still a long way off from a computer replicating the human thought process.

After reading responses from the reviewers who gave this book a low rating, it is clear they do not understand the intent of the book.

Jeff Hawkins walks you through his personal journey for the last thirty years on his research to discover how the brain works. The underlying goal of this understanding is that we will one day be able to create it. The book is not directly about artificial intelligence, or a definitive understanding of the brain. It is the necessary precursor to tackling either of these problems which is defining what "intelligence" truly is.

The low ratings reviews claim that Hawkins does not provide evidence to support his theory. Rather than theory, it would make more sense to call his theory a hypothesis. Hawkins openly proclaims very often in the book that the exact function of the brain is not entirely understood, hence it is impossible to prove with evidence any current theories. The book is not about establishing a definitive framework for how the brain works and how current artificial intelligent models should be frame, it is as he states a call to the new generations to continue research into how the brain works and a call to the new generations to recreate the structure of the brain for new artificial devices. The theory he has comprised is not a finished product by any means and he never intends it to be so.

As Hawkins steps the reader through the thought processes that led to his conclusions, his hypothesis becomes clear and understandable. I believe many readers will have a bias to his conclusions, because some of what he says defies any personal beliefs in spirituality and religion. Hawkins is clearly not setting out to disprove any ideological beliefs, or even wishing to discuss them. He is solely focused on idea at hand.

I thought it was a phenomenal book and I am really glad I read it.

"Prediction is not just one of the things your brain does. It is the primary function of the neocortex, and the foundation of intelligence." (p. 89)

Perhaps the crux of Hawkins's insight into how our brains work and how that is different from how computers work can be gleaned from considering how to catch a ball in flight.

It used to be thought that such tasks were solved by the brain through calculation. The brain would calculate the flight of the ball, adjusting the muscles of the body appropriately so as to arrive at a spot where the ball would be and grab it. Artificial intelligence people working on robots used this method and found out that it was enormously complex, so much so that the robots remained clumsy (and not about to play centerfield for the New York Yankees).

What Hawkins is saying is that the brain does NOT calculate the flight of the ball but instead recalls from memory similar flights of balls while at the same time recalling again from memory the muscular workings of the body as it went after and caught or did not catch similar balls in flight. After a bit of practice (storing memories) a person can get very good at catching balls.

In other words the brain predicts where the ball is going to be not through a laborious and lengthy calculation but through memories of similar events. This is a startling insight. Hawkins shows how everything we do is based on our brain's ability to predict events based on previous experience. Here's how it works:

First there is a "sequence of patterns" of past events stored in the brain.

Second, the brain has an "auto-associative mechanism" that allows it to "recall complete patterns when given only partial or distorted inputs." (p. 73) Unlike computer intelligence, human intelligence can figure out that "Wass up?" means the same thing as "What's up?" or that a face seen from one angle is the same as that face seen from another angle or even seen in some sort of distortion. This is something computers cannot reliably do.

Third, the brain stores "invariant representations" of things seen, heard, felt, etc. "Invariant" in this context means unaffected by differences in light or tone or inflection or background or any one of millions of small, inessential differences that could throw us off. These representations are not exact. They are in a way like Plato's ideal forms except they are not ideal but generalized. They are memories of the relationships between and among various features. In the case of a human face, Hawkins writes that what makes a face recognizable "are its relative dimensions, relative colors, and relative proportions, not how it appeared one instant last Tuesday at lunch." (p. 81)

Hawkins's definition of intelligence in terms of predictive ability is what I found most exciting in the book. When people talk about intelligence I usually want to demand "intelligence for what?" since the criteria for defining intelligence has always been so muddied. One of the ways of establishing a theory in science is through its ability to make accurate predictions. To judge the brain the same way seems strikingly right. Not only that but no longer do we have to beg the question of what intelligence is. It is the ability to predict.

These predictions are about everything in our lives and they involve all of our senses. As Hawkins puts it, "All regions of your neocortex are simultaneously trying to predict what their next experience will be. Visual areas make predictions about edges, shapes, objects, locations, and motions. Auditory areas make predictions about tones, direction to source, and patterns of sound. Somatosensory areas make predictions about touch, texture, contour, and temperature." (pp. 88-89)

While the first five chapters are eminently readable and exciting, Chapter 6, "How the Cortex Works" (the longest in the book) might be a bit tedious and technical for the general reader. (I know it was for me.)

In Chapter 7, "Consciousness and Creativity" Hawkins writes, "Most of what you perceive is not coming through your senses; it is generated by your internal memory model." (p. 202) We do not experience the world directly and we do not interpret it objectively. Our predictions in a sense are prejudices or stereotypes that sometimes lead us astray. Hawkins writes, "...you could substitute the word 'stereotype' for 'invariant memory'...without substantially altering the meaning. Prediction by analogy is pretty much the same as judgment by stereotype." (p. 203)

In the final chapter, "The Future of Intelligence" Hawkins makes it clear that intelligent machines will not be taking over the world. He writes, "The computer in your home, or the Internet, has as much chance of spontaneously turning sentient as does a cash register." (p. 214) Furthermore, an intelligent machine "will not have a mind that is remotely humanlike unless we imbue it with humanlike emotional systems and humanlike experiences. That would be extremely difficult and, it seems to me, quite pointless." (p. 208). Finally, fears that machines will take over the world "rest on a false analogy...a conflation of intelligence...with the emotional drives of the old brain--things like fear, paranoia, and desire. But intelligent machines will not have these faculties. They will not have personal ambition. They will not desire wealth, social recognition, or sensual gratification. They will not have appetites, addictions, or mood disorders." (p. 216)

Hawkins goes on to predict that, with an approach based on learning and memory instead of brute calculation, we will build truly intelligent machines, the applications of which will be numerous and include applications impossible to predict.

I would like to point out that Hawkins' idea that our cortex is continually making predictions about the environment, predictions that we scarcely notice unless they are wrong, is similar to an idea that John McCrone presented in his book Going Inside: A Tour Round a Single Moment of Consciousness (2001), a book I also highly recommend.

The ability to make predictions about the future is the crux of intelligence!
And Jeff Hawkins book ''On Intelligence'' presents some brilliant ideas on how the brain might actually be doing this.

Sure, some might say that because the brain is so complicated, we will never really understand how it works. But according to Hawkins, complexity is a symptom of confusion.
Indeed, we need some good core ideas that can help us make sense of the whole thing. In Hawkins book, the core idea is seeing the brain as a memory-prediction system. A memory system, that store experiences in a way that reflects the true structure of the world. A system that remembers sequences of events and makes predictions based on these memories. According to Hawkins, such a system is the basis of human intelligence, perception, creativity, thoughts and even consciousness.

The brain doesn't ''compute'' answers to problems. It retrieves the answers from memory. And that is why the brain can be so fast, even though neurons really aren't that fast.
It only takes a few steps to retrieve something from memory. Slow neurons are not only fast enough to do this, but they constitute the memory themselves. The entire cortex is a memory system. It isn't a computer at all.

Hawkins book is a real page-turner. Exciting and fascinating throughout.
A brilliant book that gives some really good insights into how the
brain might actually work.

-Simon

All of human intelligence exists in a 2ml. six sheet layered region of the brain called the neo-cortex as large as a dinner napkin when laid out side to side. This region processes information regardless of sensory form, and is present in all mammals to varying sizes; most of us are lucky to have the largest among all mammals. All incoming sensory information from sight, smell, touch, sound etc. is processed using the same algorithm. Contrary to what most scientists believe, there are no modules specializing in specific tasks such as language. Any part of the neo-cortex is equipped and flexible enough to manage any task, and all parts of this region of the brain process information based on pattern/time recognition. Mr. Hawkins believes throughout the learning process, certain areas of the neo-cortex will specialize to particular tasks, but this specialization is not pre-designed.

The neo-cortex does not operate like a computer; instead of computing exhaustively until a solution is found, the neo-cortex utilizes stored invariant representations of actual events in memory to predict solutions, and continuously compares results to predictions to validate. "Prediction", as Mr. Hawkins asserts, "is the primary function of the neo-cortex, and the foundation of intelligence.... Intelligence is measured by the capacity to remember and predict patterns in the world." Bobby Fischer, one of the greatest chess players of all time had an incredibly retentive memory, able to recall most of his speed chess games move by move. It's not surprising, therefore, to expect Fischer to have an IQ score in excess of 180 as his school records indicated (memory + predictive power from pattern recognition = intelligence).

Mr. Hawkins asserts the inner workings of the neo-cortex is not magic. We can understand it, and ultimately build intelligent machines that work on the same principles. Intelligent machines can only be built utilizing this memory-prediction framework of the neo-cortex with a hierarchial design (different layers of the neo-cortex manage differing levels of information complexity).

Artificial Ingelligence (AI) and Neural Networks (NN), the first two forays into building intelligent machines have failed so far because they focus on behavior and outcomes. Mr. Hawkins believes behavior is a manifestation of intelligence, not intelligence itself. An intelligent being can possess intelligence without exhibiting any kind of behavior. AI researchers believe the only impediment to their effort of devising intelligent machines is lack of processing power. Hawkins is highly critical of these two fields for ignoring the biological aspect of intelligence.

In the latter part of the book, Mr. Hawkins states fear of intelligent machines taking over the world is unfounded because machines will lack desire, ambition, the pursuit of social status and wealth. But it doesn't take much for this doomsday scenario to materialize. It can happen with one well networked intelligent machine that develops ambition or the resentment of slavery to humans. If intelligent machines surpass the power of the human brain some day, which no one doubts they will, then they will also exceed the ability of humans to dominate their environment. Even a malicious electronic virus, most likely the creation of some miscreant is sufficient to generate the ambition necessary to eliminate all conceivable obstacles to dominance.

Mr. Hawkins also neglected to mention the prospect of merging biology and machine. A future man-made intelligent entity may turn out to be part human, part machine. Brain cells, axons, neurons etc. may be reproducible in petri-dishes with minor tweaks and additions of devices to optimize algorithms and neuron connections to magnify the potential of the brain multi-fold and create a bigger, faster, stronger version of us.

"On Intelligence" is a readable and novel publication. Chapter 6 was the only part of the book with technical material not well suited for the lay person, and skipping over the entire chapter doesn't take away from understanding the rest of the concepts presented. The more advanced material was thankfully included in the Appendix section only.

Although presenting more of a wiring diagram than an actual algorithm, this book conveys (quite understandingly) a synthesis of the brilliant discoveries and insights into the activity of the human neocortex. It is a great starting place for fresh directions in creating what most call artificial intelligence (perhaps more aptly called non-biological intelligence).
Although not the point of the book, Hawkins's framework of a hierarchical temporal memory prediction model, also goes a long way to explaining all kinds of human behavioral phenomena that one can observe in everyday life: interpersonal relations, perception, cognitive development, learning processes, false belief persistence, communication, miscommunication, personality disorders, mental illnesses, even dreaming.
I was also very glad to see the emphasis Hawkins puts on the hierarchical and temporal aspects of human perception, the importance of which so many people working on pattern recognition seem to underestimate. There is no such thing as "a" thought in a cardinal sense. It is all dynamic; it is all process; it is all interconnected.
While some readers have detected a mild arrogance in the tone of Hawkins' introduction, I found the tone to be rather matter-of-fact and nowhere near as self-absorbed as, say Benoit Mandelbrot, or Douglas Hofstadter.
Perhaps best of all, in the true scientific spirit, Hawkins makes testable predictions based on his framework, so that is theory can be disproved or refined.

This is an extremely important book in the field of Artificial Intelligence. The author reject this Artificial Intelligence because it identifies intelligence to the behaviors produced by this intelligence. Hence the machine simulates intelligent behavior but is not intelligent. Three things are essential goals to satisfy if we want to move towards intelligent machines. We have to take into account and integrate time. We have to include as architecturally essential the process of feedback. We have to take into account the physical architecture of the brain as a repetitive hierarchy. Strangely enough the main mistake is already present in this first programmatic intention. Jeff Hawkins does not include the productions of that intelligent brain. I mean language, all ideological representations or models of the world from religion to philosophy and science, not to speak of arts and culture. And strangely enough this mistake is locked up in an irreversible declaration:

"A human is much more than an intelligent machinre . . . The mind is the creation of the cells of the brain . . . Mind and brain are one and the same." (41-43)*

We cannot but agree with the first sentence, but the mind is not "created" by anything. It is produced, constructed by the brain from the sensorial impulses it gets from the various senses and the way it processes them in its repetitive and parallel hierarchical architecture. But the mind is a level of human intelligence of its own. Unluckily Hawkins will not see it. I have already said what it excludes from this human intelligence, but we must add the fact that this human intelligence lives in a situation that enabled this intelligence to develop and invent its first tools when Homo Sapiens started its journey on earth some 300,000 years ago. This situation requires from the weak animal that Homo Sapiens is to develop these tools to compensate for its weakness, and to coordinate its survival and development with communication and social organization which implied and required a culture, a model of the world to migrate, develop new productive means, and be able to develop as a species in order to expand all over the world: Homo Sapiens was a migrating species from the very start because of his very brain and the mind it could procude. Jeff Hawkins forgets about the phylogeny of Homo Sapiens. He takes intelligence as existing in itself without a genesis from nothing to what it is today. In other words he speaks of evolution but he does not study it and how this evolution brought this human species into developing intelligence, means of communication and means of production that did not exist before.

At the same time he does not consider the feelings and emotions of that human being and he at best locates them in the old brain, the brain inherited from the species before mammals since the cortex only developed with mammals. It is also obvious this is a mistake. Due to mirror neurons man is able (with some top mammals along with him, to develop empathy, the possibility to imitate (hence to learn through imitation and when language was invented to learn through repetition) and to share the feelings of others and one's own feelings with others. It is this ability more than the old brain that is at stake here and is neglected. That makes Hawkins neglect social aims, productive objectives, cultural targets, ideological psychological social motivations and of course social organization. To invent and develop intelligent machines would not even exist as a plan or a project or even a desire if Homo Sapiens had not been able to blaze and then run the track leading to development.

He is sure right on the fact that behavior is only the consequence of all this but by rejecting behavior because he rejects behaviorism (which is purely ideological on his part) he also locks himself out of the possible approach of human relations, human motivations towards others, hence concrete, material and also emotional and intellectual behaviors. And that prevents him from coming back to the situation that has to be controlled and set up collectively to reach collectively defined objectives. Globalization is right now the best example of how objectives have to be defined at the level of the planet and no longer at the level of particular countries or groups.

But apart from that the whole book is essential because Hawkins concentrates on the study of the brain and its hierarchical architecture, and I should say its double architectural structure, not double in nature but double in working.

The whole adventure starts with the senses and he straight away says there are a lot more than five senses even if we can consider there are only five basic sensorial organs: the eye, the ear, the tongue, the skin and the nose.

At the level of the eye we have to add motion, color, luminescence and spatial orientation. At the level of hearing we have to add pitch, length, intervals, timbre, spatial orientation and balance (vestibular system). At the level of touch we have to add pressure, temperature, pain, vibration but also spatial orientation and movement on the skin that will be useful both in torturing (along with pain) and eroticism or emotions (along with pleasure). At the level of smell we have to consider intensity, appeal (good, bad or somewhere in-between), spatial orientation. At the level of taste we have to add temperature, texture, appeal (good, bad or somewhere in-between), and even finer elements like sweet, salty, acid, alcoholic and many others. But, and he insists on that, the general senses of the body are essential too. The whole body is a network of sensors that checks and measures our joints and joint angles, all our bodily ,positions, and all proprioceptive receptors (sensory receptors, in muscles, tendons, joints, and the inner ear to detect motions or positions of the body or the limbs, that respond to stimuli arising within the body.) Note these are indispensible for walking, running, swimming and all movements, particularly coordinated movements like gymnastics and all kinds of martial arts And we should add the physiological sensors and mechanisms that measure our inner level of satisfaction, dissatisfaction, balance and unbalance of every single organ of ours. These last sensors are essential for a new born child since it is those he/she will use from the very start and that will prompt his first cry or call. And every single of these senses and sensors sends messages to the brain in temporally organized sequences. The eye reboots its vision three times per second, what is called a saccade.

The first hierarchy he takes is exemplified by vision. I will integrate the eye into it right away though the eye is more or less marginalized in Hawkins's approach. And here the eye sends many messages according to the particular abilities of the various retinal cells that capture the signal. I will insist on the fact that he neglects: the signals are sent from the retina and are spatially oriented right-side right and upside down. He neglects it because we do not have an "image" on the retina and it is not an "image" that the retina sends. But the spatial orientation of this "pattern" as he calls it is essential. The brain will have to interpret this orientation to reestablish the proper one thanks to the signals sent to the brain by the other senses and thanks to its experience starting right after birth. Experiments have been performed using glasses that inverted the orientation of the "pattern" on the retina and after a short while the brain corrected the initial correction and provided the mind with the proper spatial orientation.

In the neocortex, the capture of a visual stimulus is hierarchically organized and we must keep in mind that the signals are renewed three times a second. In the V1 area only many small segments and isolated characteristics like colors are deciphered. These numerous small elements are sent to the V2 area where they are regrouped into larger elements. Then they are sent to the V4 area where they are regrouped into recognizable elements like a nose, an eye, etc Then they are sent to the IT area where they are reconstructed into a face for example. Here Hawkins defines a pattern as being "a stable cell assembly that represents some abstract pattern" (p. 80). At each level after learning, hence after first stimulation by one unknown element (which is sent unanalyzed to the hippocampus that takes over, identifies it and sends it back into the system), an invariant representation of each identified pattern is memorized (cortical memory, p. 100) in the cells (he does not specify the electrical and chemical procedure nor the molecular level of it). The cortical procedure then, after learning, is a recognition procedure: the pattern received corresponds to one invariant representation previously memorized, otherwise it is sent up as far as the hippocampus if necessary. The last element we have to understand is that the identification is not done in detail but as corresponding to an invariant sketch of the element and that sketch accepts variations. That explains why we can recognize someone and yet be mistaken. The mind did not make a mistake it used some elements that corresponded to the sketch it had in memory, and that was the wrong sketch.

The three basic characteristic of this hierarchical functioning are:
1- its sequential memory (sequences of patterns hence spatial in the pattern and temporal because serialized);
2- its autoassociative nature (it memorizes a sketch and not the real detailed pattern when learning, though this detailed pattern is also memorized which enables us to realize we made a mistake when we took someone for someone else, and then it recognizes this sketch in the real pattern it receives after learning);
3- and finally its "invariant representation" dimension which is the identification of these sketches as referents for further use. Here instead of saying that these sketches have to be "named" he should have said that they have to be identified at each level with some kind of Cortical Identity (CI) and this when connected with the invention of language by Homo Sapiens, or the learning of language by children would have led him to the word "concept" that he uses rarely, and the operation of "conceptualization" that he does not use at all. Homo Sapiens seems to be the only animal who managed this conceptualization power of the neo-cortex (dominated by the hippocampus) into producing language.

We come then to the heart of the volume:

"The three properties of cortical memory . . . (storing sequences, auto-associative recall, and invariant representations) are necessary ingredients to predict the future based on memories of the past . . . Prediction . . . is the primary function of the neocortex, and the foundation of intelligence . . . Evolution discovers that if it tacks on a memory system (the neocortex) to the sensory path of the primitive brain, the animal gains an ability to predict the future . . . This new idea of the memory-prediction framework of the brain . . . " (p. 84-105)

We can notice there is an intellectual drift in his reasoning. Evolution does not have a mind or intelligence. Just as we can prove human articulated language is the result of the conceptualizing power of the brain on one hand, and of other physical mutations dictated by the long distance bipedal nature of Homo Sapiens (not the first hominid to have that characteristic but the first to be endowed with mutations that go a lot farther than before) that are absolutely necessary for survival on the other hand (low larynx, high level of innervation of the laryngeal-glottal-buccal masticatory and articulatory apparatus, high level of coordination of various organs and functions), we have to consider evolution as being a blind and unguided process that selects haphazard mutations when they are propitious to bringing a higher survival potential to a given species. It is quite obvious that the development of the neocortex of mammals into human neocortex provided Homo Sapiens with a higher survival potential. In other words Hawkins suffers of some teleological bias which is a way to escape from asking who did it and hence a way to exclude the possible religious answer. But that is wrong. We don't have to answer the question of where does the logic of evolution comes from because we cannot answer this question with any scientific final elaboration.

Then Hawkins moves to the second hierarchy, that of the neo-cortex structures. The neocortex is divided into columns that are perpendicular to the surface of it. It has six layers. The first layer has few cells that have myriads of small dendrites connected to their neighbors by synapses that can build and rebuild themselves. Then they have three axons, two horizontal and lateral in the first layer connecting this cell to distant other cells all over the brain on one side and on the other side, the famous spindle cells, and a third one going down into lower layers of the neocortex. When layers 1, 2, 3 are activated the activating pattern goes to layer 5 and then layer six. In layers 1, 2, and 3 the pattern is analyzed to be finally identified in layer 5. Then it is moved to layer 6 where a prediction might be performed about what may come next from this identified pattern. Then the transmission branches into part of it being sent to the Thalamus and then back to layer 1 as a feed back and part of it being send simultaneously to motor areas for processing. Layer 4 is the layer where a newly learned pattern, identified by the Hippocampus arrives to activate the column, that is to say layers 5 and 6 and beyond. This can be summarized in a triple hierarchy: the mind must first discriminate an element, then identify and eventually name that discriminated element, and finally classify ort conceptualize this identified and named element. This basic conceptualization that has to be constructed in a child through education, just the way it was constructed in Homo Sapiens through experience.

It is important then to cross this approach with a phylogenic and psychogenetic approach of language to understand how language was invented and how it is learned. That of course would require a lot of space and it is not here it can be presented. But let's say that three hierarchies can be seen in language and all of them can only be understood as the crossing of the neocortical capabilities of Homo Sapiens on one side, and the highly frail state of Homo Sapiens or the highly dependent state of a human newborn on the other side. These hierarchies are that of the word: consonantal roots, isolating characters or themes, and conjugation-declension fronds giving the three (maybe four) vast phylogenic families of languages: consonantal Semitic languages, isolating Chinese, Tibeto-Burman and Khmero-Vietic languages, and agglutinative (the vast Turkic family from Turkish to Siouan) or synthetic-analytic languages (Indo-European and Indo Aryan languages).

The triple syntax of any language: Categorial syntax (discriminating nouns and verbs, spatial units and temporal units), Functional syntax (building the sentence on the pattern [AGENT (feed-ER) - RELATION (feed) - PATIENT (feed-EE) - THEME (feed-Ø, food, fodder)] and finally Expressive syntax (expressing the mood and modalizations imposed onto the utterance by the speaker and his relations to his environment. These three syntactic functions are mapped onto the first hierarchy by making it all discursive in root-languages, making the last two discursive in theme-languages and only keeping the expressive level for discursive means in frond-languages. Note each one of these three syntaxes is a hierarchy too by themselves.

Taking language into account would have enabled Hawkins to understand that he cannot consider the mind is the brain. The mind is an abstract and absolutely virtual construct of the brain from the various patterns the brain has registered in its own cells and molecules. I insist here on molecules because Microtubule Associated Proteins have been proved as having a role to play in various mental operation, particular with the loss of ,their phosphorylation when activated by some stimulus, for one example. The mind is based on the hierarchical potential and architecture of the brain and this potential and architecture produce the conceptualizing potential that will produce the virtual mind and its tools. These tools are essential if we want to understand the emergence of Homo Sapiens as the superior intelligent mammal on earth and if we want to understand today's man and human society. The first of these tools is (spoken) language (note written language was invented only around 5 or 6 thousand years ago some 300,000 years after the invention of spoken language). Then Homo Sapiens invented all "ideological" tools to understand and explain the world in order to survive and expand in a state of great physical inferiority as compared to most of his predators. These tools are religion, astronomy, science, history, all constructed models of the world produced or that could be produced with the conceptualizing power of the human brain. Note here Neanderthals could not even invent fishing whereas Homo Sapiens just started with fishing to move onto agriculture, herd-husbandry, and so, and all that before inventing written language.

So I do not believe "the mind is just a label of what the brain does." (p. 204) and the mind the way I have sketched it is something that might be one day equaled by machines. But these minded machines will not be human since they will not be able to learn and develop their brain and mind the way man does it, from scratch and as the result of an intense and highly emotional intercourse between an individual and his/her linguistic, cultural, social and emotional environment. We are not speaking of a machine loving a man, but of a machine loving a machine not as something programmed but as something learned from experience. As a matter of fact the Terminator saga is a lot more instructive on that point than what Hawkins says. In the same way the intelligent machines are not the machines themselves but all the Mr. Smith taking over the earth by decision of the Architect who manipulates machines into attacking humans till one, two or three humans are able to negotiate the end of the war with machines who accepts on the basis of Neo being crucified in order to be able to defeat all the Mr. Smith and the Architect's matrix. Once again we are far away from what Hawkins says.

To conclude, Hawkins's book is the first important step against the apocalyptic messianic prophetic prediction the engineers turned theoreticians like Ray Kurzweil who are already taking all the necessary pills to be able to live long and merge with intelligent machines in less than fifty years, and thus become the nurtured cows of these intelligent machines, who would not be intelligent enough to understand that kind of slavery would be doomed to destruction just like any other form of slavery was and has been doomed to destruction. If these machines were humanly intelligent they would understand that as a basic requirement to qualify for intelligence.

But at the same time Hawkins does not reach the level of the mind. He locks himself in the physiological and biological brain pretending it is the mind mixing up the capacity and the potential. He thinks too much with metaphors and comparisons. To use one I would say that a plane CAN fly but that this plane is not the FLYING POTENTIAL itself. The plane has that potential but to realize it a whole procedure is necessary (with kerosene, air strips, engineers in the air traffic control tower, pilots, passengers, freight, stewards and stewardesses, etc) and flying can only become a reality when that procedure has been performed. Hence the FLYING POTENTIAL is a VIRTUAL capability of the plane, just like the MIND is a virtual construct of the brain using its POTENTIAL INTELLIGENCE, and this POTENTIAL INTELLIGENCE cannot produce any INTELLIGENT ACTION if the VIRTUAL MIND is not activated and used by the brain.

The first intelligent machine invented by man was language in order to satisfy the need for communication Homo Sapiens had. That language has had a long career in improving and developing man's lot. It has also transformed its inventor and his/her society.

There still is a long way to go to even approach such humanly intelligent machines. In the meantime we will invent and use more and more intelligent machines that will liberate our brain and body of innumerable tasks that would otherwise use our mental and physical time and energy. With this mental and physical time and energy we will develop new forms of intelligence that we cannot even imagine today, and we must not forget that evolution goes on and man is a natural species. The more contact he/she will have with intelligent machines the more chances there will be he/she will go through mutations and developments that will be retained by evolution and education as vastly increasing human intelligence. The more intelligent machines, the more chances man will become more intelligent.

Dr Jacques COULARDEAU

For those who are interested in this field, it is hard finding good books. Many writers are technologist and many of them try to defend positions hard-to-defend (Brooks or Minsky should be good examples). Others are cognitive-psychologist and have bought the merchandise of the first ones about the brain as an information processor (Pinker should be an example of these). Others are essentialists and refuse to discuss the idea of intelligent machines because intelligence is human and that's all. At last, some of them are visionaries like Kurzweil or others.

Luckily, there are writers, coming from technology, philosophy, sociology or whatever, that escape from that classification and it is a real pleasure reading them: Dennett, Searle, Maturana, Varela, Hofstadter, Dreyfuss, Hillis and....Hawkins.

Hawkins has a double background: Technology and Neuroscience. His definitions of intelligence and his explanation about how the brain works and how this knowledge could be used to build intelligent machines is outstanding.

Before reading "On intelligence" and being familiar with the state-of-art in technology, I was convinced that building an intelligent machine was impossible. After reading this book, I am almost convinced that it is possible to build intelligent machines. This is only a matter of time and having people like Hawkins.

After listening to a podcast interview with Jeff Hawkins, I picked up this book because although I'm by no stretch of the imagination an AI expert, Hawkins' arguments regarding the failures of AI research over the decades made sense, and so I thought it would be interesting to learn more about his particular take on the topic. Hawkins certainly delivers in that regards, offering a theoretical framework for his conclusion that the brain is essentially a highly organized prediction machine which manages to outperform even the most powerful of today's computers despite the brain being woefully slow comparatively.

To be clear, this is not a book about "artificial intelligence", but rather focuses on how the human brain operates. The authors devote a mere 30 pages (Chapter 8: The Future of Intelligence) to a specific discussion of how Hawkins' theory might apply to technology development. Do not construe this as being a shortcoming, because Hawkins' intent (as I understand it) is to right his perceived listing of the AI ship by first rethinking the concept of intelligence before applying these theories to silicon.

If you've any interest in the science behind what may one day make the machines of "I, Robot" a reality, consider reading this book.

Back in 1977 Carl Sagan's Pulitzer Prize winning Dragons of Eden introduced us to Reptilian brain and the revolutionary idea that most of our brain is much older than humankind. Hawkins takes this fascinating subject to the next level by asking what it is that the neocortex, that part of the brain that only mammals possess, does. Just like Sagan, his speculations give us a new way of looking at ourselves and what it means to be an intelligent being.

He comes to the subject with a unique background that combines inside knowledge of how computers work with a lifelong passion for understanding the brain. This enables him to explain his ideas in a way that is approachable to all readers.

His explanations of how artificial intelligence and neural networks fail to consider what brains "really do" is very valuable. Finally he makes a convincing argument that intelligence should be measured, not by behaviour, but by the ability to make predictions.

His "theory" may not pan out in all its details but he makes an important contribution by looking at the subject from a productive point of view. Also, the book is a wonderful introduction to how the neocortex functions. It is true that he does not go into the philosophical implications of his ideas, but he admits outright that he is not a philosopher.

On Intelligence takes us down two paths. The first and least interesting is a survey of the moribund state of artificial intelligence, or AI. Jeff Hawkins claims that AI applications haven't lived up to their hype because they focus on machine logic, connectivity and processing power instead of understanding and replicating the decision-making capabilities of the human brain. Which is his second and much more compelling path: an exploration of how the human brain produces intelligent thought.

Hawkins focuses on the neocortex, a relatively recent evolutionary addition to the human brain. The size of our cortex (a sheet of cells about the size of a dinner napkin) relative to our body mass gives us a huge advantage over other mammals and other species. Hawkins' central assertion, derived from the theories of Vernon Mountcastle, is that the human cortex functions in the same way no matter what kind of data it's processing. Details brought in through our sensory organs - seeing, hearing, touching - are converted into neural patterns, then processed up through a hierarchy of cortical levels. Information gathered in the present moment is matched against previously built neural patterns stored in the cortex.

Intelligence, then, is matching the data you pick up from your current environment against the representations of reality stored in your cortex in order to make predictions about the future. Your cortex is essentially a prediction machine. Its complexity lies not in its structure but in the trillions of neural connections linking different cortical layers and levels to one another. As you learn something, you drive its stored representations down to lower levels of the cortex, freeing up capacity to learn more sophisticated representations. An expert, whether it's a plumber, stockbroker, or oboe player, is someone who has stored more representation, and can therefore make more sophisticated predictions.

What Hawkins proposes makes intuitive sense. You can test it out yourself by taking a walk and observing how your mind works. On a beach, for instance, you'll take in the warmth of the sun, the force of the wind, the sound of the waves, and match them up against mental templates to make a prediction. If the weather is benign, you'll predict a leisurely walk down the beach. If something is off - a cloud blocking the sun, a stronger wind, a bigger boom when the waves crash - you'll run this up against other mental patterns and predict the onset of a storm. In all cases you are relying on the inputs of your senses and matching them against stored representations of past experiences.

The book becomes dense with scientific detail only in Chapter 6 which is a level by level description of how the cortex works. Even here, the prose is relatively jargon-free, and the essential points are accessible to the dedicated non-scientist.

Hawkins' goal - a unified theory of how the brain works - is admirable, and his hypotheses seem well-reasoned, at least to this non-scientist. In the interests of simplicity and clarity, though, he steps lightly over some of the messier aspects of mental functioning such as individual consciousness (how do I know it's me writing this?) and the subjective quality of human perception (is the red I see the same red you see?). Hawkins doesn't ignore these issues, but he doesn't address them in any detail either.

Except for the thalamus, (seen as an essential organ for sequencing information in representational patterns) he also doesn't spend a lot of time dealing with the ways in which other brain and body parts affect our pattern-making processes. Emotions and mental pathologies, not to mention false memories, have a huge impact on our ability to predict future outcomes. Which is why some writers and philosophers see the brain not as a well-oiled prediction machine, but as a hallucination engine. Mental pathologies aren't the focus of Hawkin's investigations, but some good companion works to read would include Antonio Damasio's The Feeling of What Happens, and Oliver Sack's The Man Who Mistook His Wife for a Hat.

These issues aside, Hawkins deserves a lot of credit for injecting some provocative new thinking into the neurobiological debates, and for doing it in clear and accessible prose. He even provides samples of testable predictions that can be run as experiments to prove or disprove his hypotheses. Whether you ultimately agree or disagree with Hawkins, you'll come away with some new ways of looking at that amazing arrangement of neurons situated between your ears.

An extremely impressive book, which only a polymath thoroughly versed in both computer science and neuroscience could have produced. While reading On Intelligence I felt for the first time that the age-old fog surrounding the brain’s ability to recognize patterns is finally beginning to lift. Hawkins shows how a single basic cortical structure, repeated millions of times, can account for our ability to handle sensory input of all types, build up intellectual edifices of arbitrary abstractness and complexity, and choreograph our muscles’ most delicate routines. By the end of the book I felt convinced that his specific proposals for further neurological research deserve vigorous pursuit, as does his vision for intelligent machines that incorporate the architecture he has unveiled.

I have only a couple of quibbles. The sixth chapter, which gets into neural detail, is heavy going, and I believe it could have been organized somewhat more clearly. And I find Hawkins’s charitable attitude towards John Searle’s notorious Chinese Room “paradox” somewhat perplexing. I view this scenario — in which Searle, who understands no Chinese, imagines himself producing convincing answers to written Chinese questions by using a rule-book to shuffle around slips of paper stamped with symbols — as a pernicious and misleading attempt by a Luddite to reclaim some of that dualistic mental magic that modern cognitive science seems to be stealing from us. As many others have observed, for Searle to produce convincing answers to complex Chinese questions would require a mind-boggling array of symbols (not all of them Chinese characters), years of processing time as he shuffles them about, and a vast rule-book that plausibly represents the life experiences and dispositions of an actual Chinese speaker. The fact that in this farfetched scenario Searle, mindlessly manipulating slips of paper, comprehends no Chinese has no more significance than the fact that individual neurons in my brain are too stupid to understand English.

In a similar vein Hawkins is critical of the Turing Test, which Alan Turing devised as a way to assess whether computer programs have truly achieved intelligence. His main beef seems to be that by focusing on a computer’s behavior, in particular its ability to mimic human linguistic behavior, Turing overlooked the critical ingredient, whether the computer truly understands English-language concepts. Responding to this criticism, let me first concede the possibility that, by focusing too heavily on the Turing test and misconstruing it, subsequent AI researchers have headed down a blind alley. Nonetheless, I think Hawkins has (at least temporarily) overlooked Turing’s main insight here: the surest way to assess whether a computer has understanding is by questioning it and observing its reactions, and without understanding no computer will fool us for long. The same of course applies when we want to investigate the intelligence of our fellow humans. While it might nowadays be theoretically possible to perform high-tech brain scans to confirm that your friend Pat has understood the sentence you just uttered, the old-fashioned behavioral tests are surely superior from a practical standpoint (barring cases where Pat has debilitating physical impairments), and probably from a philosophical point of view as well.

The curious thing is that Hawkins demonstrates in many parts of On Intelligence that he appreciates these points. For instance, he is happy to contemplate the possibility of a computer simulating the brain’s operations and achieving understanding, notwithstanding Searle’s insinuations that understanding inevitably eludes “mere automata.” Similarly in his discussion of consciousness, where he rightly rejects the mystical mumbo-jumbo customarily associated with the topic, he effectively implies that behavior is all we have to go on when ascribing it (or other mental states) to other people. So in all, it is difficult guessing what to make of Hawkins’s soft spot for Searle and harsh words about Turing. Perhaps his main point is that some old-fashioned AI researchers have given “understanding” short shrift, and in that case Hawkins’s program is a welcome corrective. But the faults in traditional AI are not fairly ascribed to Turing; nor will Searle provide us with any constructive insights on the road ahead.

Whether or not JH's hypothesis proves correct (see other reviews for details of the hypothesis), it certainly demands attention. Those who level the criticism that JH does not prove his claims have a point. But they're also beside the point. JH does not assert that his hypothesis is true, only that it or something like it, is VERY LIKELY true. At the conclusion of this highly readable book, JH is careful to remind the reader -- per philosopher of science Karl Popper -- that no hypothesis can be proven absolutely true; it can only be proven false. What JH says, essentially, is 'The brain could work this way, and it well might, and there's some compelling evidence to suggest that it does. Neuroscientists, prove me wrong.' At the same time, he levels a serious critique of AI research, and goads AI researchers to move down the (potentially) very productive new path he lays out. Whether or not any of JH's speculations will bear fruit, no one -- not even JH -- can say. But my intuition says JH is on to something. Big. If nothing else, OI is great food for thought. A must read.

In "On Intelligence," Jeff Hawkins presents a new theory about how the brain works and how we can finally build "intelligent" machines. The neocortex, the center of higher thought, is the focus of attention here. Hawkins says that neuroscientists are lost in the complexity of mapping out neural pathways, and are not coming up with compelling overarching theories that begin to explain how we think and learn.

He believes there is enough evidence now to posit a common cortical algorithm, as first proposed by Vernon Mountcastle, a neuroscientist at Johns Hopkins, in 1978. The algorithm is hierarchical, with lower layers encoding data from a sensory organ, but higher layers dealing with abstract signals that bear little resemblance to the sensory signals. Hawkins asserts that brain researchers got sidetracked partly due to the experimental difficulty of taking measurements. The standard approach is to present a static sensory stimulus and take readings of resulting cortex activity. It is too difficult to work with dynamically changing stimuli, so researchers have missed a point that Hawkins believes is crucial: the brain can only perceive dynamic stimuli.

Hawkins' theory, called "Memory Prediction Framework," defines intelligence as "the capacity of the brain to predict the future by analogy to the past." According to him, there are four key attributes of neocortical memory that differ from computer memory:
* All memories are inherently sequential.
* Memory is auto-associative; a partial memory can be used to retrieve the full memory.
* Memories are stored in invariant representations.
* Patterns are stored in a hierarchy.
Support for the theory is most concretely expressed in chapter six, the meatiest part of the book. This is where the author describes in some detail his vision of how the neural circuitry in the layers of cortex works. The description is compelling, but takes more work to follow than the other chapters.

Chapter six ends with several fascinating observations that are built on top of the neural circuitry described earlier. It emphasizes that perception and behaviour are highly interdependent because they both originate in a detail-invariant representation that is then transmitted through both motor and sensory cortex. Also, although many researchers have discounted it, Hawkins argues that feedback and the importance of distant synapses in cortex is essential to explain the Memory Prediction Framework theory, and should be reconsidered. The theory includes the broad principles of how hierarchical learning of sequences explains how children first learn letters, then words, phrases and finally sentences, and as adults we can speed-read without needing to study every letter. The author believes that the memory of sequences re-forms lower and lower in cortex, allowing higher layers to learn more complex patterns. Finally, the hippocampus is briefly described as logically residing at the top of the cortical hierarchy: the short-term repository of new memories.

An impressive result of the speculations in chapter six is a list in the appendix of 11 specific, testable predictions made by the theory, which is an invitation to brain researchers. And Hawkins founded a company, Numenta, to develop the Hierarchical Temporal Memory concept based on the theory.
Chapter six also hints at how daydreaming or imagining occurs, when predictions from layer 6 of a cortical column are fed back to layer 4 of the same column. Cortical modeller Stephan Grossberg calls this "folded feedback". In chapter seven the book expands on philosophical speculation about the origin of consciousness and creativity that arise from the Memory Prediction Framework theory. Creativity is defined here as "making predictions by analogy". As the author says, there is a continuum of creativity, from mundane extrapolations from learned sequences in sensory cortex to rare acts of genius. But they have a common origin. This is how a piano player can quickly figure out how to play simple melodies on a vibraphone, or a customer in a strange restaurant can figure out that there is probably a restroom in the back. Creativity is so pervasive that we hardly label it as such, unless it violates our predictions like an unusual work of art. There are practical suggestions in this section for how to train oneself to be more creative, and an interesting story of how Hawkins conceived the handwriting recognition system, Graffiti.

Chapter seven ends in speculation about the nature of consciousness, imagination and reality in response to the inevitable questions to which this type of work gives rise. A review on the the store website by Dr. Jonathan Dolhenty takes issue with what he describes as "plain old-fashioned metaphysical materialism and, probably, old-school psychological behaviourism," which are largely discounted theories today. Dolhenty is a philosopher who thinks human intellect at the higher abstract and conceptual levels cannot be described by such a simple extrapolation of the Memory Prediction Framework. But I found the connections made between brain theory and "mind" reassuring. Leave it to others to build on this foundation. In fact, Hawkins does hint at a broader source of the mind in chapter seven, where he says that it is influenced by the emotional systems of the old brain and by the complexity of the human body.

The last chapter in the book contains another vision, of how intelligent machines might be built in the future. This is back into the Popular Science mode. Unlike many current roboticists who believe humanoid robots will be needed to interact with humans, Hawkins believes humanoid form is pointless and impractical. He advocates working from inside out, by building sensing mechanisms and attaching them to a hierarchical memory system that works on cortex principles. Then by training the system he believes it will develop its own representations of the world. This system can be built into any sort of machine, and the sensors can be distributed if desirable.

The technical challenges of building an intelligent machine include capacity, which by analogy to the brain, at 2 bits per synapse, would require 8 trillion bytes of memory or about 80 hard drives. Connectivity is a larger problem, since it would be impossible to provide dedicated connections. Hawkins believes the answer would be some sort of shared connections, like in today's phone network, but this is still a challenge.

As an aside, there is no mention of the Cyc project, which has been working since 1984 to build a mammoth semantic knowledge base. But unlike the automatically learned representations in Hawkins' proposed artificial brain, the ones in Cyc are hand-input in a preconceived structure as a vast quantity of terms related by assertions.

The last chapter ends with a very positive view of the potential of intelligent machines to solve problems humans cannot, because they can be equipped with custom senses, immense memory, and even be networked to form hierarchies of intelligent machines. Hawkins believes that intelligent machines will be a hot topic in the next ten years. It is easy to get caught up in his excitement.

During the past half-millennium the history of anatomy documents the peculiar custom of using the most advanced technology of each era as the definite model of the human brain. The first match was with clockworks during the sixteenth century; then with the steam engine, in the nineteenth century; one hundred years later with telephone switchboards in the first half of the twentieth century, and in the recent decades, naturally and expectedly, with electronic computers. However sound they might have appeared at each time, all these comparisons proved inadequate after a while. All have fallen short when matching up manmade machines with the extraordinary prodigy of the human organ that designed them.
Twenty years ago Jeff Hawkins, the architect of many technologies and a successful Silicon Valley entrepreneur, decided to turn the metaphor all the way around and walk it in the opposite direction. Instead of starting from already invented equipment to develop explanatory models, Hawkins decided to first understand the way the brain operates--more specifically, how the cerebral cortex works--and design from there on a new technology. With such a challenge in mind, after studying neurology on his own and co-working with many scientists, the ambitious businessman initiates a monumental (if not chimerical) project to design and build electronic equipment that is to operate similarly to the human brain. Numenta, a company founded by Hawkins in 2005, has the mission to make this initiative a reality. His book ON INTELLIGENCE, written with science journalist Sandra Blakeslee, describes the reasoning behind his adventure, the factors that support the idea, the obstacles that make it extremely complex and the scientific developments that will contribute to its realization.
There is only one chapter in the book complex and difficult to read (the author warns about this) that presents his view of a detailed model of the functioning of the cerebral cortex, the thin layer of thirty billion neurons that surrounds the brain. Even with this exception, ON INTELLIGENCE is an entertaining and educational book. The description of the four attributes of the cerebral cortex that make it radically different from electronic computers is fascinating. The first attribute is the storage of sequences of patterns (instead of isolated data interrelated by data models and database software) that enables the recording and recalling of stories or sequences. The second is the ability to pick the full story or sequence from only a fraction of any part of whole without the need to access the complete pattern (we recognize a song by just listening a bit of it). The third is the conservation of the essence of every pattern although the rest of the information might be variable (this is why we recognize incomplete objects or identify people we have not seen in years despite changes of age, contexture or makeup). The fourth, the difficult-to read chapter of the book, is the storage of the patterns in a hierarchical structure.
These attributes provide the cerebral cortex an intellectual capacity quite different from those put forward in previous interpretations. According to Hawkins the cortex is an organ of prediction; predicting is the main function of the human brain and this capability is the very foundation of intelligence. The neurons involved in any activity (or some associated neurons yet to be discovered) are activated prior to the arrival of the corresponding sensory signals, be they visual, auditory or tactile, anticipating the coming events from some sort of extrapolation of all the patterns that the cortex has already in its memory. For example, when someone enters a restaurant where he never has been, he can "predict" with a good degree of certainty in what direction are the bathrooms. When the event is completed, if the result matches expectations (this happens most of the time), the owner of the brain does not even realize that a verification transaction was performed. If, on the contrary, expectations do not coincide with reality, there is a surprising reaction, followed by corrections and learning lessons that eventually lead to the creation of new patterns.
In Hawkins's perspective, the human brain is an organ that builds models based on patterns and analogies and generates with them creative predictions. When it does not find correlations, the brain invents them anyway with minimum consideration on how preposterous they may turn out. Pseudoscience, prejudices, intolerance and religions are the result of these inventions.
The concept of prediction that Hawkins developed in 1986--we should remember that he did not graduate in neurology--was later confirmed in independent scientific studies. For example, Rodolfo Llinás, a neuroscientist at the New York University School of Medicine establishes in 2001: "The capacity to predict the outcome of future events--critical to successful movement--is, most likely, the ultimate and most common of all global brain functions."
I believe the development of truly intelligent machines is an unfeasible project. Its endeavor, nevertheless, will lead to many new scientific discoveries. The brilliant entrepreneur recognizes that his target is neither the invention of an electronic model of human consciousness nor the production of machines that arrogantly say "I." His main interests aim at the development of computers with vision, the design of thinking robots and the construction of machines with capacity to learn. The invitation to the greed of the young generations to join in some way the great idea is outside the context and beauty of the whole project. Contributing to human growth or making a difference--not plain utilitarianism--should be the driving forces of scientific research. Still, from my perspective of cognitive science enthusiast, I consider that the very description of the functioning cerebral cortex (I suppose that a few neuroscientists may disagree with it) and the concept of prediction as the fundament of human Intelligence far deserve the reading of this excellent book.

This engaging, non(too)technical book offers a new and plausible theory of how the brain, or more specifically the neocortex, works.

When I learned about the existence of this book, I was drawn to it for a number of reasons. For one thing, I'm intrigued by the faculty we call intelligence: what is it, exactly? For another, I, like the author Jeff Hawkins, have long been fascinated by the brain and how it works. And finally I was eager to read a book on neuroscience by a nonscientist, for Hawkins, inventor of the Palm Pilot and other things, is a technologist who has long pursued brain science as a hobby. I love the idea of contributions to knowledge being made by amateurs, for they seem best able to think outside the box.

And thinking outside the box is what Hawkins has done here. His point of attack was to discover whether it is possible to build an "intelligent machine," and how this might be done. He noted the relative unsatisfactoriness of the results achieved by "artificial intelligence" in the computing world, and wondered why this was. How was it that a computer, with processors executing millions of instructions per second, could not seem to remotely approach the prowess of the human brain at most tasks requiring "intelligence," when the cells in that brain could only execute a few tens of "instructions" per second? Even relatively simple perceptual tasks, like recognizing faces and chairs, are done effortlessly and almost instantly by humans, while machines toil to achieve a success rate well below 100%. What have humans got that computers don't got?

Humans have got a way of processing information that is completely different from the way computers process it. The brain, unlike a computer, does not run on the instructions of a single master program controlled from the top. The brain, says Hawkins, operates as a vast array of small, localized processing systems. In particular, the neocortex--that sheet of neurons that covers the upper frontal part of our brain, and is responsible for all of our human intelligence--is set up as an intricate, interconnected feedback system that can be boiled down to performing two functions: memory and prediction. He's saying that what we call intelligence is the interplay of memory and prediction.

To defend and illustrate this thesis, he goes into some detail on exactly how he thinks the neocortex is wired up. It is known to consist of 6 layers of neurons, which are all interconnected in certain characteristic ways. Hawkins shows why they are so interconnected, and how this results in the formation of memories at increasingly high levels of abstraction. What we call intelligence is the recognition of a current sensory input as belonging to an abstract category already in memory. According to this view, animals that also possess neocortexes have this same intelligence, but in lower degree than humans, who have the most sophisticated neocortex (one interesting fact in the book was that dolphins, which are intelligent and also possess large brains, have a neocortex with only 3 layers, as opposed to the humans' 6).

Hawkins makes his case very well; I found it persuasive. Where I found myself less persuaded was in what I would call the philosophical side of the book, where the author addresses questions such as, What is creativity? What is imagination? What is consciousness? And, of course, the basic question: what is intelligence itself? I think that Hawkins, an extremely able technologist and even scientist, overplays his hand as a philosopher.

Along the way, for example, he talks about Plato's theory of forms as an explanation of how we are able to recognize sameness in the hurly-burly of our ever-shifting sensations. Hawkins notes offhand, "His system of explanation was wildly off the mark." Well, maybe it was and maybe it wasn't, but what makes Hawkins so sure? I recall that Roger Penrose, in chapter 1 of his Road To Reality, treats the world of mathematical truth as one part of a Platonic world of forms, seemingly real but also different from the worlds of mental experience and of physical things. My point here is just that Plato's ideas live on; they'll keep climbing back out of the dustbin of history.

I had similar feelings about Hawkins's take on the other philosophical questions. He contends that there the difference between the intelligence of, say, a rat and of a human is purely one of degree. But Mortimer J. Adler, in his book Intellect: Mind over Matter, contends the opposite. According to him, intellect--which was the word formerly used to label the faculties that we now point to with the word intelligence--is something more than the rudimentary power of abstraction used by brutes. In this view, animals are able to respond to individually differing things in the same way, as when a rat is able to press different triangular buttons to get food, but this is not the same thing as

"cognizing what is common to them or knowing them in their universal aspects. . . . By means of concepts, and only by means of concepts, we understand kinds or classes as such entirely apart from perceived particulars and even though no particular instances exist."

Adler argues that the brain is a necessary but not a sufficient condition of the human intellect. The existence of the intellect, he thinks, is a sign that a human being is something more than just a body.

Is Adler right in this? I don't know. And I don't think Jeff Hawkins knows either, no matter how confident he is in his assertions. But for someone who wants to build intelligent machines, I think a cautious outlook would be fitting. Hawkins dismisses people's worries that superintelligent machines might become our overlords or our executioners, like HAL in 2001: A Space Odyssey or the Skynet computers in the Terminator movies. He thinks that such behavior would require the presence of the equivalent of the emotional centers of the brain in addition to the neocortex, and he's only planning to build an analog of the neocortex. So don't worry, folks.

I recall reading a comment from the Dalai Lama, apparently changing his mind about whether robots or machines could become sentient. He said that if some being had the necessary karma to take birth or manifestation in a machine, then that would happen. I note that karma is not a word that occurs in the index of Hawkins' book.

But this is a good, clear, strongly argued, plainspoken, provocative, and, yes, intelligent book. Hawkins has persuaded me that "intelligent" machines are very likely in our near future. And I'm sure they will be very helpful and will have no reason to do anything mean to us, their intellectual inferiors.

If they do turn on us, then, well, maybe God will help us.

The main premise of the authors is that brains are predicting "machines," or, as they say, some kind of box in the dark (but not a "black box" they argue). Well, pardon my ignorance on the history of neuroscience, but, um,...duh? Of course brains make predictions. And of course those predictions are observed in behavioral patterns, which is just another form of the Turing Test Hawkins seems to be really against.

I read the book cover to cover and found the lack of detailed explanation frustrating. I wanted to hear more about this big idea but the more I read, the smaller it seemed. There were blatant contradictions, as well. For instance, he goes on about how the brain requires its developmental history to attain the patterns which help it act "intelligently" and that sensory input is a big part of this so-called training. Well, unless I'm just a complete ignoramus, I don't see how you can isolate just the circuits that need to be trained and give those the mysterious inputs so that they achieve "intelligence" without doing anything else to the rest of the brain. He says as much (kind of). But then he talks about how we can make systems that have completely different (e.g., 10-dimensional) sensory worlds that will learn like us.

Wha?? Did I miss something in evolution class? Our brains came out of the process and history of life, and, as far as we know, we like three dimensions and 5 senses (yes, I know there are more, but humor me) and certain other "inputs" our brains have evolved under. So, to say that you can (1) isolate these systems somehow (magic!), (2) develop a complete description of their behavior across several scales (remember, it has to be isolated because you've removed them out of an animal's skull, where they have ALWAYS been), (3) build a non-organic replica that responds to hypersensory input (from the 10th dimension), and, finally, (4) claim that it will acquire "intelligence" the way we do...to say all that is just science fiction. Sorry, I'm no expert in the field, but I know the smell of BS.

The variables are infinite, and this is assuming you can even isolate stuff from the brain, let alone reproduce it exactly in silico. I'm no dummy. Those systems are part and parcel to the whole brain and the whole brain belongs to the whole organism. His claim that the brain is a like a dark box, but not a black box, is tantamount to saying you can put a mouse brain inside of a man's skull and the man would start running around looking for cheese and become scared of cats.

That's just my nonexpert opinion on it. I almost gave the book 1 star, but I happened to quite enjoy his 2-page epilogue, which, I felt, was very inspiring.