Flesh and Machines: How Robots Will Change Us
One of the things I recall about Brooks' work from grad school, was this idea: that the world is its own best model. What that means is, that instead of trying to model the world on a computer then compute what to do based on the model, you should just do stuff and then see how the input from sensors changes. By acting, you interact directly with the "model" -- the world -- and therefore you can cut down on the computation enormously.
I have the feeling that this notion works well for simple robots, including lower life forms such as insects. Like Genghis, they simple do "simple" stuff based on simple neural computers that hardly warrant the name. But where Brooks' work falls short, as you can see in the review, is where neurons are clumped into serious computers that do model the world. The worst offenders, of course, are humans. The problem is that have no idea how to wire a robot to do that, and a lot of the behaviors we really want from robots rely on it.
AI still has a long, hard road ahead of it. But we will succeed, eventually, simply by virtue of reverse engineering if nothing else.