I just finished On Intelligence
by Jeff Hawkins and Sandra Blakeslee. While I enjoyed it a lot, I don't have a great deal to say about it right now. My thoughts will have to work themselves out slowly.
The central tenet of the book is that "prediction" can be said to occur on basically all neurological levels and that it's the basis of understanding. To build mental models of parts of the world, what we need is to be able to predict how they will behave in our experience. I agree and I feel that this is a pretty good place to locate an intersection between evolutionary theory and pragmatism. More thoughts on this subject will have to wait for another time.
I'd recommend the book for light reading. It could easily be read in a day. I took several days to finish it because I got stuck on a section in the "How The Cortex Works" chapter. This part involves a hierarchy of cortical columns ("higher" and "lower" regions), each of which is made up of a confusingly connected series of layers (which are also referred to with the "higher" and "lower" terminology). I didn't like the diagrams in the book, so I drew my own diagram of the structure of a cortical column, which after an iteration or two ended up looking something like the following:
It's missing a few details, but imagine trying to get a mental picture of that from a description and you can see why this section threw me for a loop.
Later I'll write about another thing that threw me off, but I have to meditate on it for a while first.
Things I liked:
- The Memory-Prediction framework. I haven't really written much about it but I anticipate it'll be coming up in my thought and discussions of other things soon.
- Learning about the Neocortex and general brain structure.
- Mountcastle's theory about a common cortical algorithm.
- Invariant representations. Add one more thing to the list of reasons to suspect cognition has a sort of metaphor as a basis.
- Treatment of sensory data as abstract patterns & the hierarchical structuring of pattern recognition.
- I was forced to think about how neural processing works & finally got a glimpse of what Hebbian learning is all about.
- The section near the end on differences between animal and human thought points in the right direction. It also contains an E. Coli thought experiment that's very similar to one from Stuart Kauffman's Investigations. I wonder if that's coincidence or influence.
Things I didn't:
- Though I think prediction of a sort is a huge part of cognitive function, the theory at hand seems try to render very important and complex aspects of intelligence as small problems to be worked out later. In the early part of the book the author spends quite some time explaining why intelligence has nothing to do with outward behavior. I suppose this is why the explanation of the brain's generation of outward behavior is so inadequate. I'll need to hear a lot more about how the memory-prediction model generates outward behavior to be convinced that it does. Also, I'd like to hear more about abstract reasoning. Maybe I've missed something here, but it seems like there's quite a bit that needs to be elaborated on before we can call the memory-prediction framework a total theory of intelligence.
- Since it's for popular audiences, it doesn't really get into details much. This is a problem with many of the books I've been reading lately, and I guess it's somewhat hypocritical to complain, as I've been specifically selecting books that I could read quickly.
- It seems that one of the reasons it doesn't get into details is that the theory presented in the book is very hypothetical. There are too many "we don't really know much about this" or "some neuroscientists would disagree but trust me" moments. I would like concrete models that I could play with but maybe they're just not available at this stage.
- I haven't really been able to find much discussion of the theory in question in other places. Though I like the theory I don't have any opinions about it other scientists, so I don't really know how to evaluate it as a claim to truth. As is pretty clear at the end of the book, there's quite a bit of work to be done before this is possible.
- It sets of my teleological and anthropomorphic interpretation detectors quite a bit, though when developing a semantic account of the function of brain structures for popular audiences, maybe there's no getting around this.
- It kind of loses focus near the end and starts going on with weird and questionable sections on creativity and consciousness. I suppose these sections are designed to suppress the horrific flow of reader mail there would be without them.
P.S. Invariably the "Things I didn't" list is longer than the "Things I liked" list, even though I definitely liked this book as a whole. I try to remain very critical and often the reasons to remain critical of a book are easier to describe than the impact it had on my thought processes as a whole. This is particularly evident in this post.
Labels: book review