I'm interested in how intelligence works, in people and in animals. One of the most interesting models I know about is behavior-based reasoning. This is the theory that what we think of as intelligence is not a single thread of thought, but is the result of many simple components which are all working at the same time. This sounds kind of like neural networks / parallel distributed processing, but the difference is behavior-based AI uses units of intelligence which are not just homogenous equations, but are recognizable behaviors linked in particular ways.
Another interesting theory about intelligence is that most (or even all!) of it never uses any form of variable-based logic or representation. This is called reactive reasoning, because it involves acting without planning, at least not by the agent. In recent years, the term "reactive" has come to be employed for any action triggered by reacting to the environment rather than deliberation, or cognitive assesment.
The research that served as a starting point to mine combines both of these approaches. Reactive, behavior-based intelligence is composed of many interacting reactive behaviors. One of the best documents of this approach is Intelligence Without Reason (MIT AI Lab Memo 1293, April 1991) which is by Professor Rodney Brooks. See also my related research web page, or Maja Mataric's Behavior Based Control page.
The reactive, behavior-based approach is famous for producing the first robots that can reliably interact with the real world at realistic, animal-like speeds. I personally have also used it for creating an AI music program, the Reactive Accompanist. However, it has been over a decade since the approach was first introduced, but it is still not clear whether it can lead to robots capable of more sophisticated behavior. In fact, no new purely behavior-based robots have come out that could do significantly more complex things than the early projects such as Herbert and Polly. My research is concerned with understanding why this is, and trying to find ways to extend the capabilities of such systems without losing their advantages.