I just finished listening to my second audiobook ever: “On Intelligence” by Jeff Hawkins and Sandra Blakeslee. It is quite an interesting book. I can’t say that I was impressed by any of the ideas – none of them seem novel to me. The discussed a lot of the things that I was actually looking into 6 years ago when I started the license plate recognition project (that actually didn’t work) and then when I was working on my first Ph.D. research topic.
It is difficult to summarize the book in a post, but I’ll try without boring you with the details (and focusing on the ideas I found most interesting): The point he tries to make in this book is that the brain works by hierarchical prediction. A set of inputs in any layer of the brain causes it to try and predict the next set of inputs and pass it down the hierarchy. This has two positive results: you know what happens next so you automatically build a model of how the world works – this lets you plan what to do, where to look, etc.; and also helps the lower levels of the brain learn how to represent something in an invariant fashion. If you move a pen and you expect it to write a letter “a” on a paper, your lower level neurons that actually look at the “a” would associate this “a” to the concept “a” independent on where in the page it appeared.
I know it might sound a little confusing if you haven’t read/heard the book. He spends about 275 pages or 9 hours or talking to explain these points. You shouldn’t expect to understand it all in a paragraph. It’s not a romantic novel or movie I’m talking about here. 🙂
Alright, time to move on and get some cleaning done.