On Intelligence by Jeff Hawkins

Max developed a great interest in consciousness, brains and artificial intelligence and has begun reading books in the field. He recommended a video showing a lecture by Jeff Hawkins, a computer engineer turned to brain science in California. Since I wanted to show some interest in Max's interests, I checked out Hawkins's book from the library. Sky immediately scooped it up, read it quickly and trashed it. I'm not a good spouse in that I don't just agree with what my husband says, so I took Sky's scorn with a grain of salt. He is well versed in neuro-science, computers and AI, and might have trashed Hawkins just because he trashes almost everyone in those fields.

I am NOT well versed in computer science, electronics, AI and brain science, although I know more than many because I've lived with engineers and worked for brain scientists. But Hawkins's book was obviously a pop read, not a serious study. I also liked his lecture and was excited by his claim that the neo-cortex was used for prediction and that models of intelligence should turn away from behaviorist's models and turn toward something like prediction. That got me thinking and I was really excited to read "On Intelligence" despite Sky's scorn of it.

By page 52, I felt that Hawkins was really astray, so much so that I began to take notes. I no longer felt that the book would be a light-hearted re-examination of an age-old problem, but that Hawkins was on a mission to push his own idea, whether it be true or not. Sometimes this works, but I felt that, in this book, I was being led by the nose and my hackles were up. Why was this? On page 50, Hawkins speaks of a paper called" An Organizing Principle for Cerebral Function" written by Vernon Montcastle in 1978. He speaks of Montcatle's conclusion that if the cortex looks similar everywhere that there is a common function that is performed by all the cortical regions. That seems logical, except for the recent discoveries that although the structure of the cortex is remarkably consistent and the wiring is fairly consistent, at the chemical level, there is much more going on. I also suspect that there may be much more going on at levels which humans have yet to explore due to primitive equipment. But I can go with an idea that the cortex might be performing similar functions throughout the cortical area.

Now, on page 52, Hawkins makes his first error in logical thinking. He suddenly jumps from using the word brain and cortex to the statement: "a single paper and a single idea that united all the diverse and wondrous capabilities of the human mind." Okay. Hawkins lost me at this point. Why? Because he leaped from a discussion of cortical function to a conclusion about something foggy like the human mind. Did he explore a discussion of what is mind? No. Did he talk about cortical function as human mind? No. Did he talk about exciting discoveries in neuro-science about the cortex and about OTHER parts of the brain in humans other species that led scientists to speculate that consciousness and mind is synonymous with the mind? No.

Before Max starts saying "but-but-but" let me go on. At this point in the book, I just noted that my hackles were up. I saw that Hawkins was prepared to lay down an argument for which he made no grounding. So, many people do this. They start with talking about space exploration and jump to talking about FTL drives without any discussion of why it might be possible or not. Okay, the book is just fun, not serious. I wasn't turned off by this point, but cautious, not because of the topic or the conclusion, but that I suspected Hawkins of being a sloppy writer.

On page 70, Hawkins starts his argument for the neocortex being a prediction device. He first makes a definition of memory that is one epistemological style of memory retrieval, which is not unusual, but he bases his further argument on the assumption that all people remember events in an invariable sequential fashion that can only be recalled in the same sequence, which is just wrong. Memory has to do with association. There is temporal association, sequential association, and a myriad of associations that have to do with map-making. The type of memory he speaks of is a characteristic of a type of personality that tends toward and epistemological style that is sequential. He goes on to explain in great detail of how memories MUST be retrieved by running through the sequence in time. He uses fallacious arguments for this such as only being able to remember the alphabet from a-z, which is not only false, but makes him look like an idiot if he can only remember the alphabet if he starts at a.... It is a delightful trick for people to remember it in any form. He goes on and on arguing all sorts of examples for sequential memory from songs to the order of one's house to the invariable sequence of drying off when one gets out of the shower. These are all instances of personality, not universal.

By page 77, he reveals himself as a Platonist. He states that memories are invariant representatives instead of dynamic recognition of patterns in run time. He makes no distinction between recognition memory and synthetic memory, of memetic associations and models.

On page 91, he begins to re-define his terms. He hammers down his argument that memory is prediction, that when one goes through a sequence of events such as walking through one's darkened house, one is predicting what will come next. One predicts that when one turns on the stove, the burner will get hot. On and on, until he "proves" that what the neocortex does is predict. At this point, I was appalled. When I get up in the middle of the night and stumble over a chair, I say "I don't remember this chair being here!" I don't say "I didn't predict that this chair would be here!" He seems to think that memory, even compulsive memory like remembering a tune when one hears the first notes, is prediction. By page 102, not only is he saying that prediction allows humans to think better, but that they developed better motor control than other animals, which is just not true at all. He seems to think that bigger is better, and goes on about how the cortex and the way it is wired accounts for all the differences in humans and other animals. He says with a flourish that "with humans the cortex has taken over most of our motor behavior. Instead of just making predictions based on the behavior of the old brain, the human neocortex directs behavior to satisfy its predictions." This is an interesting statement, not only in that he glossed over much of what he needed to prove, but that he jumps to the conclusion that the neocortex is directing behavior to satisfy predictions. If, by this, he means that you will pull back your hand if you put it on your car and your car is so cold that you know your hand will freeze, this hardly seems like higher level cognition.

But, I'm not putting down the book at this point. I'll concede him some hand waving and some assumptions that just because he thinks he remembers in such and such a way, it is the way everyone does and that the neocortex is reponsible.

Hawkins goes on to talk specifically about the cortex. I had a quibble about his "invariant saccades" when it is obvious that saccades are trained and a huge part of what goes on in infancy on through all kinds of specialized training. My saccade of a face will be very different from a "normal" saccade of a face because I am looking as an artist, not as a man or as a woman or as a doctor. He seems to think that we have invariant memories of people, when many people have worked on just why this cannot be the case. We have a recognition memory that often persists in the light of changes or misidentification, and it can be highly trained in different ways. I think by the huge amount of information flowing both ways in recognition, that it's obvious that the cortex is not only pulling up old memories, but laying down new associations all the time. In our highly social society we must do this to survive. If we had a brain stuffed with invariants, there would be less adaptability for the countless changes of association that we must make all the time. Invariants or stereotypes are less common that people want them to be.

On page 125, he begins to talk about word models. Here, he makes several logical and philosophical errors. He assumes that hierarchy among the cortical structure is assembling parts, not nested classes. So he uses connotations of hierarchy in a way that promotes errors in his model. He thinks that each object is composed of smaller parts in a hierarchy, that a faces are at the top, composed of eyes, noses, mouths, etc. Now a hierarchy is thus: car-Ford-Mustang-Mustang LX- Mustang LX 1984. A collection of objects is: car-engine-pistons-rings. He seems to think that without this hierarchy in place, all would be confusion, when really all that we seem to do is lump things together on the fly to make up a map of something we can also identify on the fly. We do not think, "this is my room, it consists of these objects:...." When asked where we are, we say, "oh, in my room." Only when asked "what is your room," do people then make some confusing attempt to define their room, usually by the color or the objects within it, often to the fun of someone who wants them to "prove" that they are in their room, which is impossible to do. In dreams, we feel compelled to think "oh, that is Max" when the person looks nothing like Max, but we think for some reason that we need a Max character. If pressed, we then try to remember something about Max, Maxness, such as someone tall. But more often than not, Max will become someone else or vanish entirely. There is no "Max" in a dream--even in lucid dreaming is it notoriously difficult to build up a realistic image of Max. If the brain had collections of neurons dedicated to "Max" there might be some Platonic invariant we could recall better. Some things are easy to recall and seem to be more so with repetition, as if the neurons are trained to do so. But recognition clearly does not need to be so hard-wired as Hawkins seems to think.

By page 146, he convinced me of the opposite of his hypothesis that the neocortex was responsible for mind and for intelligence. But, because of reading this book, I am fascinated by the thalamus and the role played by the hippocampus in consciousness since it seems that both these parts of the brain are involved with higher functions. Without parts of the neocortex, we cannot remember things or how to do things; without the thalamus, we are vegetables. Hm. Which is essential for mind?

On page 154, I found Hawkins descriptions of the recognition of a musical note in the process of the cortex to be way, way too involved. The brain is not a computer. Hawkins himself said that the cortex was very slow compared to a computer. His process of retrieval is way too involved and, again, invalidates some of his modeling of how the brain might work, not on the cell level, but on the mentation level. He's trying to figure out what goes on mentally when the cells excite and he's not getting very far.

On page 166, he describes how a child learns to read and doesn't even remember his own discussion of recognition memory that he made early in the book. He says that a child learns the alphabet and then tries to sound out three letter words. A child only does this when all other ways fail. Most children learn to read at a higher level. They know language extremely well by this time and they sit with an adult or sibling who reads to them. Max never looked at his name and said, mmmm aaaa ks, mmaaks, mmmaks--MAX! like we see parodied on TV. He had seen MAX so often with the word associated with it, that he just new that the lump figure MAX meant the word max, even when it was on the air conditioner/heater in my car with "min" which he did not recognize or sound out. My friend Sam learned to read at 5 because he quickly learned that the words were "talking" and knew how the cadences of words went in books. When he got to a word, he only read the first couple of letters, if that, and recognized the word by the association with the remembered sentence. When he came on unfamiliar words, then his older brother, who could not read as well, had to sound out the word phonetically, like Hawkins describes. Still, they persisted at the morpheme level with "in-form-a-tion" rather than at the phonetic level with "iii-nnnn--fff--ooo--rrr--mmm--ay--sh--uu-nn", oh, "information." If learning reading took place as Hawkins suggests, then sign language and reading pictographs and glyphs would be impossible.

A child doesn't learn to walk like a robot learns to walk. There is no trial and error. You cannot explain to someone the steps to throw a ball or beat batter, you must show them, sometimes by manhandling them and going through the motion with them. The motion is learned, like reading, not in this hierarchy of commands which forms Hawkins brain model, but at a wholistic level where the association can be repeated as a package deal. Everything points to the hierarchical model being contrived and used in abstract thought, not in learning or thinking.

In chapter 7 on Consciousness and Creativity, Hawkins attempts to put his cortical theory into an explanation of the functions of the mind. First of all he offended me with his heavy anthropomorphizing of the behavior of other beings, like "the tree is predicting where it will find water and minerals based on the experience of its ancestors." Oh boy. I know what he's trying to say, but he's trying to use his own jargon and it gets pretty silly. Then he tries to make imagination and creativity into prediction, but again, he gets very sloppy. He uses Shakespeare as an example of creativity and quotes "There's daggers in men's smiles" as analogy (an example of literary genius) in that daggers are analogous to ill intent and men's smiles are analogous to deceit, and forgets that all primates bar their teeth to intimidate other primates and that their eye-teeth (way more prominent in other mammals) are SHAPED like daggers. Sheesh. But this is nit-picking. I only include it to whine about Hawkins being generally sloppy and not respecting his reader. I was pretty irritated by this point.

But, my real beef came with him saying that creativity is merely messing with people's predictions. He elaborates when he says that shuffling his Scrabble letters is creative and that creative problem solving is backing off the problem. Since, he's decided that prediction is recognition memory, when the brain recognizes new words in a reshuffling of the Scrabble letters, that's making a new prediction and being creative. AGH!!
Prediction is not recognition. Prediction is saying what a pattern will be based on an amalgamation of past patterns that one evaluates given current conditions. Creativity is the creation of a NEW pattern out of old data. It is not the recognition of a word when the pattern emerges but the creation of a NEW word or, better yet, a new way of glyphing words altogether, like shorthand. He even admits, that in his own creation of handwriting recognition in his Graffitti invention, he did not do handwriting recognition, but made people learn a new glyph system that the computer would recognize! By this time, I'm trying not to hate this man, who doesn't even listen to his own arguments.

Okay, now he jumps to consciousness being in the cortex. I have no idea why, because he's not explained ANYTHING being in the cortex but trained recognition modules that assist in memory. He tries to explain the differences between self-awareness and what he calls "qualia" or feelings being associated with different sensations are independent from input. He uses qualia to support his Platonic model. Then he goes on to make imagination a form of prediction and I gave up.

By this time, I knew that his chapter on the Future of Intelligence was going to be idle speculation. I was intrigued enough by his idea that the cortex and brain was not just a behaviorists' box and might be a prediction machine, but I wish that I had not read the book, because I know that he thinks the brain is merely memory. He falls into the category of "more ram" equals intelligence, which is fallacious and silly and no different from all the other AI guys who think that the brain is a device.

I would like to see some studies done. One group of people can memorize a long text. The other group can memorize the same text as a song. Does one group of brains light up in a different area when the text is remembered? Does the song group or the verbal group have to remember from start to finish or can they jump in at any point. If so, do their brains light up differently?

But Sky is right. Without a non-invasive, portable device that reads what is going on in the brain, it will remain a black box and intelligent guys like Hawkins can talk about what goes on inside without having to know what goes on. We know a lot about neurons, but not a lot about the brain in action. I think his book clearly demonstrates that there is a long way to go.