From Twitter:

This is your mind on media.

Musings on the effects of media on cognition.

“On Intelligence” by Jeff Hawkins

[Amazon link: “On Intelligence] I have to say first that I bought this book with a great deal of skepticism. I found it highly unlikely that a product designer would have anything other than hyperbole and over-generalizations (i.e. Malcolm Gladwell, the “Tipping Point” or “Blink”) regarding something as complex as machine intelligence or the human brain. Boy was I wrong. After reading Hawkin’s introduction and discovering that he was and is far more than a product designer I began to relax my initial resistance to the book. Overall the book is concise and well written. I tore through it in two days which is highly unusual for me. I tend to plod through nonfiction, taking notes, looking up references. But there just wasn’t any time for that with Hawkins’ book because I continually got a sense that Hawkins had the secret to machine intelligence hidden somewhere in the back of this simple book; a simple theory; a real program maybe. And he does, sort of. There are no programs, no mechanics, no and-here’s-how-to-do-it in Java. Oh well. But he does lay out a more thorough hypothesis of how the brain functions than I have yet encountered in my own study of artificial intelligence.

And on that note, probably one of the more interesting sections in the book is the section that describes, in essence, here’s why the other guys haven’t figured it out yet. In short, the neuroscientists haven’t figured it out yet because they’re too busy looking at all the individual parts and sections and collecting data from MRIs. They can’t see the forest for the trees. The AI guys haven’t gotten it yet because they’ve all but discarded the idea of modeling anything on the brain. The brain is a kludge and we can do better. And the connectivists got stuck in the application realm of the science and realy just haven’t gotten around to tackling anything much bigger, or holistic for that matter. Mind you, Hawkins is modest throughout. He doesn’t discard anyone’s work nor claim that his way is the only way. He presents these descriptions of other fields to help clarify where we’ve been versus where we could explore.

Without ever mentioning it specifically, Hawkins is taking an emergence theory approach toward intelligence, a self-organizing system of small parts doing simple things that creates complex behavior. And without doubt, a lot of people are going to have real problems with that approach. It begs questions like “Does it eliminate our capacity for free will?” and “If intelligence is all that simple then how come we don’t see more of it?” He lightly deals with these concerns but it is clearly not the drive of the book — as well it shouldn’t be. The drive of the book is forming a hypothesis that can help us to create intelligence in forms other than the one that has naturally evolved. He doesn’t even attempt to explain the entirety of the brain, only the neocortex — that part of the human brain that has evolved beyond all other animals. He grants other animals like rats, monkeys and dolphins with kinds or levels of intelligence, but points out that ours is a unique kind of intelligence, revolving around an ability to create abstractions (something he calls invariant memories), concrete memories, and predictions.

Specifically, Hawkins calls his hypothesis the memory-prediciton framework of intelligence. His argument is that what we call intelligence arises out of our ability to remember the past and predict things in the world based on those memories. He eloquently points out that many theories of intelligence deal with the concept in terms of behavior and outcomes but that his framework would account for someone lying still, doing nothing, and yet still being intelligent. The mind is a kind of giant dynamic feedback engine that is constantly adjusting its behavior to adapt to an ever-changing world, finding patterns and consistencies and alerting itself to situations that don’t fit patterns or are different. And the best thing about this framework (in my mind) is that it accounts for the fact that humans, with all their intelligence, still make mistakes — from stumbling and catching ourselves to wandering off our usual paths and getting lost but being able to find our way back.

I won’t go into all the great examples of intelligence and how it can be viewed in Hawkins’ framework — that’s one of the primary reasons to read the book! I highly recommend it to anyone with an interest in artificial intelligence and the mind. Even with a very small understanding of neurons in the mind (mainly farmed from Scientific American) I was able to completely understand the more biological parts or the mechanics of the book and Hawkins’ description of the human mind.

I only have two criticisms of the book. I felt it owed a nod to Marvin Minsky’s “Society of the Mind” or Daniel Dennett’s “Consciousness Explained” which I feel both discuss the human mind as an emergent system (without the mechanics that Hakwins goes into). Besides that, I felt that he too easily dismisses the idea that intelligent machines could pose any kind of threat to society or humans. Many, many others, I’m sure, will deal with this issue (as will I in several essays on my blog.)

Besides those two criticisms, read this book. Even if you disagree with the memory-prediction framework as a plausbile explanation for human intelligence, it remains to be an intriguing tactic for developing some kind of machine intelligence.

« Previously: