On Intelligence

Rating:
Summary: The Amazing Prediction Machine
Review:
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.




Rating:
Summary: Thought provoking.
Review: In the prologue, the author states: "The agenda for this book is ambitious. It describes a comprehensive theory of how the brain works. It describes what intelligence is and how your brain creates it.". That second sentence is misleading. This book does not actually cover emotions, consciousness, or Self, among other things I'm possibly missing. The author does present a thesis and exposition of how the brain manages to be 'intelligent' which basically amounts to the capacity for problem-solving, where problem refers to a very broad category, covering most day-to-day mundane as well as challenging activities.

The writing is lucid and the ideas concerning the brain's implementation of 'problem-solving' are well-explained without recourse to much actual neurobiology, despite some of the reviews suggesting otherwise. As far as the exposition is concerned, I found the writing somewhat redundant with various examples all illustrating the same point, but it's not too much of a drag. The book is aimed at a wide audience.

Criticism of this book's content could fall into 3 categories,
----
1)problems with the theory of neocortical operation i.e. memory-prediction framework - I'm not well versed with neural architecture to find any. My intuition had led to the single cortical algorithm principle before I had heard of this book and naturally I agree with it. Hawkins touts Vernon Mountcastle's Organizing Principle of Cerebral Function as revolutionary, but that's from a neurobiological analysis. A rational psychonaut may arrive at the same conclusion via direct experience. The details of this theory are subject to verification of the testable predictions he makes at the end of the book.
----
2)problems of integrating other brain functions especially emotions - Hawkins, more or less, says that emotions are covered by more ancient brain structures, and not the neocortex. In the final chapter, he dispels fears of intelligent robots revolting against their organic masters on the basis that their designed hardware won't give rise to human emotions (fear, greed..etc) On what basis? My neurobiology is rustic, but isn't virtually all grey matter implemented on a similar biological substrate (neurons). So why can't the cortex support emergence of emotional quale? This isn't explained beyond saying that emotion centers are elsewhere other than the cortex. You could argue that the specific wiring is a constraint, but Hawkins doesn't even have a convincing answer to how distinct sensory modalities emerge from homogenuous neural processing of data within the neocortex itself.
----
3)problems of a philosophical nature - why does the neural activity give rise to qualia? How do distinct modalities of qualia arise from essentially homogenuous neural activity? On page 63, he brings up the obvious point - since all perception/cognition is via the mind, how does one ascribe 'real' existence to other minds? i.e. falsify solipsism. His answer: "They are really there.". Nothing further. On pg 75, talking about auto-associative recall: "Inputs to the brain auto-associatively link to themselves, filling in the present, and auto-associatively link to what normally follows next. We call this chain of memories thought, and although its path is not deterministic, we are not fully in control of it either.". He apparently bows to free will, but doesn't actually try to support it. He also doesn't explore the sense of self and how the brain/cortex creates/binds to it.
----

Overall though, this book is engaging and thought-provoking. I suppose the author has succeeded in his central goal i.e. to present his theory of how the neocortex operates.

Rating:
Summary: Pretty good, interesting, but wildly optimistic
Review: Every philosophy, every programming paradigm seems to have some central point of focus, and Jeff's focus is on prediction. He has a prediction scheme based upon a heirarchy.

I say he is wildly optimistic only because he is radically short on details, unfortunately falling into the siren set by real AI.
He states that the primary barrier right now is inadequate hardware (I refer to pg. 213, top).

Not that real AI is impossible, but thinking that real intelligence will fall into place with only a good initial scheme of prediction/heirarchy and fast enough/big enough hardware is wild optimism. Consider: how does he plan on building in such basic things as: representing the difference between part and whole? The difference between a quality and a quantity? How about a basic distinction like continuous vs. discrete? He seems to assume that these issues will take place merely with a cleverly devised prediction scheme and lots of training, but how he will handle these metaphysical issues he does not say...

The book is still well worth reading, is easy to read and I don't mean to "trash" this fine book; I just think that the reader should approach it with caution and remember there are vast amounts of issues that are left to the future.

Rating:
Summary: Jeff Hawkins Is Thinking Ahead
Review: For many entrepreneurs, the passion for their business is fueled by a loftier goal than the wealth garnered from its operations. For Jeff Hawkins, inventor of the Palm Pilot and the handwriting recognition system Graffiti, the goal has been a lofty one indeed. Jeff Hawkins wants to know how the brain works and he has spent twenty years of his life doing the research.

"On Intelligence" from Times Books presents Hawkins' new unified theory of real intelligence. It is Hawkins' hope and premise that if we actually can understand how the brain works to create intelligence in human beings then we will be able to take these principles and use them to build computers that think the way we do. He definitely has his work cut out for him.

Trying to understand how the neo cortex works sounds like a difficult task and one that would involve a large number of long words. It does. However, Hawkins largely avoids the problem. He deftly simplifies his model of how the brain works and through the course of the book's eight chapters makes a potentially difficult topic quite approachable in much the same way Carl Sagan brought astronomy to the masses.

Hawkins opens the book by explaining why artificial intelligence has not lived up to its promise. Twenty years after computerized speech recognition became available it still does not work to our satisfaction. He identifies the major trends in artificial intelligence and neural networks and explains why they have failed to deliver substantial results outside of very highly specialized environments. His explanations are clear and to the point. As Hawkins founded both Palm Computing and Handspring and invented the Treo Smartphone, he knows exactly what technology can and can not do.

Hawkins explains that his theory proposes that the human brain is a pattern recognition machine that breaks complicated topics into smaller chunks of time related events and then compares it to known experiences. He portrays the brain as a series of stored patterns all existing in the neo cortex, a neuron packed organ about the size of a dinner napkin, all crumpled up inside our skulls. His simple analogies to dinner napkins, playing cards, hierarchies and feedback loops explain how we can recognize a song even though we are hearing it in a different key than the first time we heard it, and that we can recognize a friend's face in a crowd despite changing angles, lighting conditions, apparent size, and distance. Hawkins explains that the human computer instantly recognizes the significant difference between a person standing at the front door with a wrapped birthday gift and another with a crowbar, and that no computer built would even have much of a chance of figuring out that the crowbar was not part of the person. The Real Intelligence that Hawkins seeks is not to be found be amplifying artificial intelligence. They are different approaches entirely.

After the preliminary work of explaining his model, Hawkins does lead the reader through one chapter stuffed with schematics, charts, and jargon. The reading becomes difficult for a few pages as the cortex is divided into layers and their various functions described. The payoff for completing this chapter is a better understanding of how our brains decide to form new memories and how subdivision of information within our cortex makes hopelessly difficult tasks possible. The material might be dense in places, but Hawkins does his best to bring it to the reader in common terms.

Hawkins then concludes the book with brilliant chapters on consciousness, creativity, and the future of intelligence. Without becoming deeply philosophical, he discusses the intelligence of animals, the significant shifts of biological memory systems over millions of years and how design professionals can develop systems that integrate more smoothly with the way our brains work. He finishes with speculation on the future of real intelligence and how it might improve automotive safety, weather prediction, and medical research.

"On Intelligence" seems positioned to become a landmark book in the popularization of the study of human brain as a path to understanding real intelligence. Hawkins opens the book with "I want to build truly intelligent machines" and then gives you an explanation of just how much progress has been made in the field. If Hawkins doesn't build intelligent machines in his lifetime, then surely a student inspired by his work will.

Rating:
Summary: Shows the way ahead...
Review: I am a neurosurgeon and I picked up this book with a great deal of scepticism because in the past, neither through my professional studies nor through reading many popular books, have I really been able to answer some fundamental questions regarding our brain. Questions like how do we think? what is imagination? what happens during the so called "a-ha" phenomenon, when you suddenly understand something that you did not a moment ago and many other such questions have been plaguing me for years. And believe me, there arent many satisfactory answers floating around.

Hawkins has made a fantastic contribution by giving us A model to think about these questions. His memory-prediction paradigm is very attractive intuitively because it automatically explaines so many facts about our brains and their evolution that other theories just ignored. But even when tested in hard scientific experiments, I predict that the basic structure of his arguement will remain intact, though details may differ.

We still do not know a lot about the architecture of the brain, the way neurons are connected to each other and the way brains develop their enormous computive capabilities. As we learn more, it is likely that Hawkins paradigm will be refined, but in the long run, we will owe Hawkins gratitude for allowing our brains to understand how our brains work!!

Rating:
Summary: A Review of ON INTELLIGENCE
Review: On Intelligence would be of interest to anyone studying cognitive science.
Many researchers believe that it is the proper design/choice of architecture
that makes a cognitive system intelligent (Architectures for Intelligence, VanLehn
Lawrence Erlbaum, 1991). Hawkins attempts to outline the architecture of
the human cognitive system. Since no actual algorithms are presented it is
not possible to actually code and run his suggestions. The hierarchy he
discribes is probably closest to my own Asa H (Autonomous Software Agent -
Hierarchical). Algorithms for Asa H have been published (A Hierarchical
Architecture for Software Agents, sci.cognitive, 14 Nov. 2003, parts 1 and 2,
R. Jones) and code has been run successfully.
Hawkins argues for simple general principles of human cognitive function.
I am not convinced. Nature (evolution) is more likely to produce flawed
complex kludges.
I may have detected a contradiction in Hawkin's argument:
How is it possible that "an unexpected pattern will keep propagating up the
cortical hierarchy until some higher region can interpret it" (pg 159)
Only one level has the right degree of abstraction "large-scale
relationships are stored at the top" "small-scale relationships are stored
toward the bottom" (pg 126). Also, the time scale/granularity only
matchs at one level. At the top of the hierarchy lengthy sequences
are stored, at the bottom, brief sequences (pgs 130-131).
The concepts represented in each layer of the hierarchy defines
the vocabulary that is available to the next higher layer. Each
layer speaks its own language. How can a pattern propagate
up the hierarchy?


Rating:
Summary: Fantasy book fit for high school students
Review: Scientists who study AI, vision, language, memory, and the brain (I include myself in this category) will find the fantastic claims in this book hard to swallow. It could be old age or an unwillingness to believe that a neophyte could come in and revolutionize the industry, but stranger things have happened, and we shouldn't exclude him. To say that the brain is a prediction device is one thing. To actually determine how representations needed to predict accurately actually exist the brain is another. You can't just point to anatomy and figure it out -- you need a very different information-processing view of the brain, something that cannot be revealed by the anatomy. The frustration is that there is a massive disconnect between what the cognitive scientist does and what the neuroscientist does, and while the effort, the optimism, the funding is admirable, the actual science of trying to decipher how to solve the bridge problem between cogsci and neuroscience cannot, should not, be trivialized the way this book does. So while I applaud the big dream, Hawkins' cannot solve the bridge problem by ignoring it. I took some time to read some of the early people hired at the Redwood institute and while they are proposing novel computational visual neuroscience theories, the problem cannot be addressed through neuroscience alone. I would recommend this book for high school students with a fascination with AI and the brain, but would recommend many other eye-opening classics (Godel Escher Bach, Society of Mind) that do not regard the existence of brain anatomy as some kind of fantasic cure-all. I suggest Hawkins spend some time doing some single-cell recording in the macaque monkey, some machine vision, some linguistics.

Rating:
Summary: It got me thinking...
Review: This book is thought provoking. First, let me qualify myself, that I am not a neuroscientist but an engineer and an experienced programmer. Like Jeff Hawkins, I share in his belief that intelligent machines can be built. However, this is only a cursory interest of mine.

As a non-medical person, I learned a lot about the neocortex from this book. As a programmer, I found great insights about the importance of stories in our thinking, as opposed to traditional computer operations. One of the main goals of his efforts is to find a "neocortical algorithm". The book is purported to set a scientific framework to find this algorithm.

I am not convinced of the wisdom of this effort, except that it may allow us to understand human behavior better. There are definitely ethical questions raised if we create such machines: Will people become dumber as they rely on smart machines? Will mankind invent an evolutionary successor who may become our master? Will people be placed out of work? Movies like "The Time Machine", "A.I." and "I Robot" come to my mind. Science fiction writers will find a wealth of juicy information they can use.

Apart from those fears, the intellectual challenges are exciting. Many of us are consumed by a desire to understand how the human mind works, in clear logical terms. Mr. Hawkins contributes unique insights based on his experience as a computer designer. I found many of his viewpoints be somewhat like reading about electrical circuits - which is not to say that this is bad thing. He brings an engineering approach to the understanding of this problem which I can appreciate.

Personally, I believe a more realistic approach would be to create "conscious machines" rather that merely "intelligent machines". This may sound more difficult, but I think you can't separate the two experiences. One can even argue that "intelligence" is merely a subjective attribute of consciousness. A "conscious machine" would have senses and observe it's own behavior - internally saying, "I am doing this now" (writing its stories), categorizing it's stories, and making analogies between stories (predicting). There is a time element involved in the "experience of experience", that I think the book seems to overlook.

The challenge to create a machine like this is a race, akin to the race to invent the first flying machines. In the race to build flying machines, some inventors built contraptions that looked like birds and bats. Studying neurons may shed light on the logic of conscious systems, but building "neuron machines" may not be the way to go. My own bet is on taking a new approach to AI, integrating video game programming methods.

Overall, there are many interesting ideas in this book.


Rating:
Summary: Engaging and informative
Review: This is a light and engaging read on the structure and function of the brain for lamen. The first few chapters debunk historical artificial intelligence efforts. The author then continues on to propose his theory on the function of the neocortex. Which is a hierarchal mesh of pattern recognizing nodes that propagate both forward and backward to predict the world and to react to changing responses.

I enjoyed this book. It was a bit of a struggle in the first few chapters, but after he got into developing his own theories it turned into a page flipper that was hard to put down.

Rating:
Summary: Exciting new theory on intelligence
Review: We often routinely talk about intelligence and we attempt to measure it for for a variety of purposes. But do we know what it is? Jeff Hawkins is one of the first people to present a specific and comprehesensive theory of intelligence with a leading role for the human neocortex. Hawkins starts by stating that Human intelliigence is fundamentally different from what a computer does.

But isn't artifical intelligence (AI) a good metaphor for human intelligence? No, says Hawkins. In AI a computer is taught to solve problems beloning to a specific domain based on a large set of data and rules. In comparison to human intelligence AI systems are very limited. They are only good for the one thing they were designed for. Teaching an AI based system to perform a task like catching a ball is hard because it would require vast amounts of data and complicated algorithms to capture the complex features of the environment. A human would have little difficulty in solving such everyday problems much easier and quicker.

Ok, but aren't neural networks then a good approximation of human intelligence? Although they are indeed an improvement to AI and have made possible some very practical tools they are still very different to human intelligence. Not only are human brains structurally much more complicated, there are clear functional differences too. For instance, in a neural network information flows only one direction while in the human brain there is a constant flow of information in two directions.

Well, isn't the brain then like a parallel computer in which billions of cells are concurrently computing? Is parallel computing what makes human so fast in solving complex problems like catching a ball? No, says the author. He explains that a human being can perform significant tasks within much less time than a second. Neurons are so slow that in that fraction of a second they can only traverse a chain of 100 neurons long. Computers can do nothing useful in so few steps. How can a human accomplish it?

All right, human intelligence is different from what our computers do. What then is it? I'll try to summarize Hawkin's theory.

The neocortex constantly receives sequences of patterns of information, which it stores by creating so-called invariant representations (memories independent of details). These representations allow you to handle variations in the world automatically. For instance, you can still recognize your friends face although she is wearing a new hairstyle.

All memories are stored in the synaptic connections between neurons. Although there is a vast amount of information stored in the neocortex only a few things are atively remembered at one time. This is so because a system, called `autoassociative memory' takes care that only the particular part of the memory is activated which is relevant to the current situation (the patterns that are currently flowing in the brain). On the basis of these activated memory patterns predictions are made -without us being aware of it- about what will happen next. The incoming patterns are compared to and combined with the patterns provided by memory result in your perception of a situation. So, what you perceive is not only based on what your eyes, ears, etc tell you. In fact, theses senses give you fuzzy and partial information. Only when combined with the activated patterns from your memory, you get a consistent perception.

The hierarchical structure of the neocortex plays an important role in perception and learning. Low regions in the structure of the neocortex make low-level predictions (about concreet information like color, time, tone, etc) about what they expect to encounter next, while higher-level regions make higher-level predictions (about more abstract things. Understanding something means that the neocortex' prediction fits with the new sensory input. Whenever neocortex patterns and sensory patterns conflict, there is confusion and your attention is drawn to this error. The error is then sent up to higher neocortex regions to check if the situation can be understood on a higher level. In other words: are there patterns to be found somewhere else in the neocortex, which do fit to the current sensory input?

Learning roughly takes place as follows. During repetitive learning memories of the world first form in higher regions of the cortex but as your learn they are reformed in lower parts of the cortical hierarchy. So, well-learned patterns are represented low in the cortex while new information is sent to higher parts. Slowly but surely the neocortex builds in itself a representation of the world it encounters. Hawkins: "The real world's nested structure is mirrored by the nested structure of your cortex."

This model explains well the efficiency and great speed of the human brain while dealing with complex tasks of a familiar kind. The downside is that we are not seeing and hearing precisely what is happening. When someone is talking we by definition don't fully listen to what he says. Instead, we constantly predict what he will say next and as long as there seems to be a fit between prediction and incoming sensory information our attention remains rather low. Only when he will say something, which is actively conflicting with our prediction, we will pay attention.

The author takes his model one step further by saying that even the motor system is prediction driven. In other words, the human neocortex directs behavior to satisfy its predictions. Hawkins says that doing something is literally the start of how we do it. Remembering, predicting, perceiving and doing are all very intertwined.

I think this is a fascinating and stimulating book. Many questions about intelligence may remain unanswered but I believe this book to be a step forward in our quest to understand intelligence. The author predicts we can soon build intelligence in computersystems by using the principles of the neocortex. He is optimistic about what will happen once we succeed in this. He (reasonably convincing) argues these systems will be useful for humanity and not a threat.

Coert Visser, www.m-cc.nl