On Intelligence (Buch)

I bought this book because of a general interest in human behavior, patterns and intelligence. I have often struggled with verbalizing the differences between how computers and people "think" - and I hoped this book would shed some light on that.

Well it did - but it did a lot more.

Now, this is the kind of material that can either spur the imagination or lay in your belly like 3-day old chili. Of course, that also depends on how (and if) the material connects with you.

The primary point of the book is the presentation of a general intelligence theory based on scientific study of the human brain for the purposes of moving forward the basic scientific principles behind the creation of intelligent machines. Well, I for one am not involved in the creation of intelligent machines however, I see the limitations of current computing technology every single day - as I am an industry analyst for enterprise software.

Regarding the theory itself, I found much of the material had a strong intuitive sense of correctness. Not sure how else one could judge this as it is theoretical. I am convinced that this book did shed clear light on my belief that traditional computing and software will not get us there - that what is needed is an entirely different approach. Jeff presents one that certainly seems plausible.

However, the big takeaway for me was the self-discovery this book triggered for me. On Intelligence has significantly changed my understanding of myself, my behavior, actions, reactions and how I perceive the world. But not everyone will get this from this book, so if you read it with expectation of self-revelation but don't get any don't be surprised.

The ramifications of much of what is presented in this book as it applies to me (I know, that was not the stated intent of the book) are still swirling about - and will be doing so for a very long time.

I have made this book a gift to friends and many have had similar response to mine - but some of not. For people like me, this is an absolute must read. If you read it and don't get much from it then you have learned one thing - you are definitely not like me!

It's just about the right time for a framework to hang our knowledge and questions about human intelligence onto. Hawkins makes a good attempt. I find that a lot of my scattered knowledge and ideas fit onto his framework and a lot of questions become clearer. If the framework sharpens open questions that's good; if current or coming knowledge shows his framework is wrong in small or big ways, that'll be good too. Having this straw man will help keep attention on how things fit together. Just the idea that it's really as simple as he says is challenging.

The writing style aims a little low. At first the book reads as if it will be on a vague popular level without any interesting technical detail or arguments. But after a chapter or two it does get into the nitty-gritty, without being too technical for a layman up to speed on the basics.

These are the main ideas (which he doesn't claim to have originated): The neocortex is what's important. It's organized into a fixed hierarchy of areas with more or less fixed relationships between areas. All the areas follow the same "algorithm," although the tuning varies between areas. The center of the book describes his guesses about the algorithm and how it's wired.

He uses the words "memory" and "prediction" to describe what the cortical algorithm does, and that interferes with his main message that it is more general than that, covering both static and temporal relationships, linear and branching relationships, focused perception and directed action. I recommend making up your own terms for the top-down, bottom-up and sideways processes he's talking about and substitute as you read.

He almost entirely neglects two important things: goal- directedness or backward chaining, and recursive structures (in which, e.g., a sentence can include a sentence, or a prepositional phrase can contain a prepositional phrase). Still I've found it very interesting to try to fit these into or onto his model. Maybe they actually aren't handled by the basic design of the neocortex. Goal-seeking may enter by training from below and pushing from the sides, and recursion may be an exceptional hack (self-looped layers? reification of relationships? a separation into blackboard and manager? echoing off the thalmus?).

If Hawkins fails to mention large areas of research, I think it's because of his sharp focus. He often seems to be describing selected pieces of work he knows more about. I was finally quite impressed with the book.

Ray Kurzweil says that as soon as AI achieves some goal, like beating Kasparov at chess, then that goal gets cast out of the AI orbit.

Hawkins describes a powerful filtering mechanism, based on patterns that are "fed-forward" as pedictions, to enable lower brain regions to sift out the mundane while passing the unusual "up" to the higher brain. As a mechanism, then, so far so good, but not likely to escape Kurzweil's curse.

Presumably the really interesting phenomena happen once the higher brain is reached. It is almost universal amongst intelligence/consciousness theories that the brain does various simplifying pattern-matching tricks, so that some inner goblin can make a choice for you. Dennett, for instance, in Consciousness Explained, has a "concert of the mind", where ideas rear up like loud instruments in an orchestra. Nobody has any idea how that inner goblin works, and you won't find one in this book.

By far the most interesting, daring, and probably the most in error, of all consciousness theories is that presented by the eminent physicist and modern-day genius Roger Penrose, who offers a theorem that consciousness is non-computable, and that its solution is wrapped up in an as-yet undiscovered physics.

I went through this book in one sitting because I couldn't put it down. As someone working in the computer field I found it very interesting providing lots of food for thought. It is easy reading. It is not an in depth look at artificial intelligence but an overview. I found it fascinating.

His logic makes sense to me at least in comparison to many of the other AI articles and books I have read. For example I have read many times that when computer power becomes equivalent to the human brain in processing power we will have comparable artificial intelligence. This just does not make sense to me because even if we cluster hundreds of super computers together and give them 300 years to complete something a human can do in 2 seconds the computers still cannot do it. I don't believe one human has billions of times more hardware power than 100's of super computers unless we are using quantum brains which is a small possibility. The problem most likely isn't hardware it is the strategies the AI engineers are using. I have to agree with the author on this point.

All the authors' ideas may not prove to be 100% correct but I don't believe he claims them to be.

The book has got me thinking and got me excited about AI. What more could I ask for? More detail and being a programmer I wish there was at least one example in code.

Overall I recommend this book as an interesting introduction to AI.

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.