Serendip is an independent site partnering with faculty at multiple colleges and universities around the world. Happy exploring!

Jeff Hawkins' On Intelligence

Sam Beebout's picture

 I found Hawkins’ arguments in On Intelligence exciting because they overlapped so much with the themes we have discussed in the course. The subtitle of On Intelligence is: How A New Understanding of the Brain Will Lead to the Creation of Truly Intelligent Machines.  As the title gives away, Hawkins’ project is to define intelligence as an engineered process. Hawkins comes from an unconventional context for studying the brain because his primary area of expertise is computer technology and engineering. Hawkins is the mind behind the program Graffiti that allows people to write on a screen with a stylus as well as the popular palm pilot Treo.  Hawkins was motivated to write this book because of his frustration that the fields of computer science and neuroscience have not been combined because both fields are too self-righteous. Computer scientists, Hawkins explains, aren’t interested in understanding how the brain works because they see it as an inefficient system, an evolutionary fluke that works, but maybe doesn’t work as quickly or effectively as we know computers can.  Hawkins retorts, though, that computer scientists have not yet been able to design, nor are they even close to designing, an intelligent computer.  He asks, “Why can a six-year-old hop gracefully from rock to rock in a streambed while the most advanced robots of our time are lumbering zombies?” (2) While we can design computers that may solve a math equation instantaneously or instantly translate Spanish to English, computer science has not been able to replicate the brain’s system.
    I was instantly intrigued by Hawkins’ discussion about computer scientists misinterpreting the brain’s system when attempting to design intelligent machines. Hawkins’ criticism of neurobiologists is that they are each too isolated in their study of an individual region of the brain to think of the brain holistically. However, I knew where Hawkins’ was going with his discussion about the brain’s overall operation because this is the story we have been working with all semester.  We have discussed that a boy can jump from rock to rock gracefully because our brain works through feedback loops, or reafferent loops that allow us to constantly adjust our response to our environment.  Hawkins takes this one step further and asks how we do this so efficiently, and what about our nervous system’s design allows this to happen? Hawkins asks this because he knows that the brain’s architecture and engineering is important to understanding intelligence.  He explains that the neuron is extremely slow compared to a computer transistor. In fact, he says, “a basic computer operation is five million times faster than the basic operation in your brain”.  The brain’s intelligence, then, must come from intelligent design.
    To uncover the brain’s intelligent design, Hawkins turns to the neocortex, which he is biased to, he admits.  He justifies his bias for the neocortex by explaining that the neocortex is the root of the brain’s intelligent activity, not the brain’s function as a whole.  He explains that he is not trying to build humans, but trying to build an intelligent machine. The neocortex’s design, he argues, is the root of understanding how this intelligence happens.  Hawkins’ discussion of how the neocortex functions also heavily overlapped our theories in the course because Hawkins bases his theory of intelligence on the fact that the neocortex, while responsible for many processes, is a single structure with a single design. “The different parts of the neocortex, whether they are responsible for vision, hearing, touch, or language” Hawkins explains, “all work on the same principles” (6).  We have discussed this phenomenon in great detail in the class, and it is the basis of our metaphor that the brain is a storyteller.  Our five senses all work in a similar process, which is to recognize the edges of things to get enough information and to make reasonable guesses to fill in the rest. We also know that our senses rely on one another, that what we hear is usually implicated in what we see, for example.  And we know that regions of the brain responsible for one sense are easily adaptable. For example, a person who becomes blind can see from a series of patterns that are input on the tongue. A blind person who reads Braille is using the region of their brain responsible for sight even though they are reading by touching their fingers to the pattern.  We know that these examples are possible because we have discussed we are only made up of patterns and signals and circuitry in the brain.  An action potential in the brain is an action potential in the brain, and it doesn’t matter whether that action potential is coming in from the tongue or the eye, our brain will recognize this pattern and adapt a way to interpret it.
    Hawkins explains all of this and makes me realize how amazing the neocortex’s adaptability is.  How is it that a single structure with a single design serve so many functions? The answer, Hawkins thinks, is in grasping a better understanding of the role and operation of memory in the brain.  Indeed, he finds that memory operates in a similar process, metaphorically speaking, to our senses. Our memories are eventually stored in the neocortex, and when they are they are stored in sequences.  We can see evidence of the fact that memories are stored in sequences by thinking about how we remember an event, the whole thing comes flashing back as a whole story.  Similarly, if we are trying to sing a melody we can only do so by singing the whole thing. To identify something by touching it we must move our hand and feel enough of a pattern, enough of a sequence to remember it, but what happens when we remember things? Certainly if I think of a flower, an image of a flower pops into my head, but how? What determines what this flower looks like if all flowers look different?  Hawkins talks about how we develop “invariant representations” that enable us to categorize things. As Hawkins puts it, “memories are stored in a form that captures the essence of relationships, not the details of the moment” (82).  This is the same process of our senses, which operate by capturing the essence of what is going on.  It is no coincidence that our sensory system and our memory operate in the same way because they are invaluably interconnected. The connection between memory and our actions in the world is at the root of our intelligence.  We are able to do what we do—leap from rock to rock, identify a flower as a flower—not because our brains operate faster, but because our brains our actions happen through the lens of our memory.  In essence, the process of filling in details when we see or touch something is a process of prediction.  We are making our best guess based on their relation to our stored sequences. Hawkins explains the brain’s ability to serve so many functions with a single structure is based on the way the neocortex evolved. It got larger and more sophisticated in the types of memories it could store, and then it began to interact more with the motor system of the “old brain”, everything not neocortex.  
    Hawkins final theory is uncannily like our boxes within boxes model, his chart even looks like it with the hierarchy of boxes and the share of information going back and forth between everything. He suggests that an invariate object is not just one box found in one region of the brain, but an interconnected group of boxes on this hierarchy. He suggests that objects become invariable to us as they moved up in this webbed hierarchy.  We stop recognizing things based on literal characteristics and start recognizing them based on memory and higher levels of interpretation. For example when I see a flower I am no longer identifying it with my senses, but through the lens of everything else. I have created a place for it, which is really a process for it, in my mind.  This is how the neocortex works, this hierarchical web of boxes constantly sending information back and forth to one another. His conclusion, it seems, is with the hippocampus, which we did not spend a great deal of time talking about. He realizes, after praising the neocortex throughout his book, that the hippocampus might be that top box in the web of boxes. It is not the I-function because we know that the I-function is really the product of the whole process, a dynamic, always changing thing grounded in some semblance of connectedness.  The hippocampus’ role is fascinating, however, because the hippocampus is the site where new memories are formed.  Hawkins asked why we need the hippocampus? If we take in new information through our cortex, why can’t we just store memories there. The trick is that we take in new information through the cortex, send it all the way up to the hippocampus to understand it, and then send in back down. The hippocampus and the cortex are constantly interacting in a process that allows us to do the majority of what we do through memory, through our ability to predict “normal” and interpret changes to it (170-2). I really enjoyed Hawkins’ book, and I would recommend is as secondary reading for this course from the beginning because he provides so many helpful accessible examples and has a way of making big, profound concepts seem understandable, and then almost obvious.

Hawkins, Jeff. On Intelligence. Holt Paperbacks: New York, 2004.