An example of AI as (animal) psychology. Which, incidentally, demonstrates that AI is not concerned solely with modelling or implementing theorem-proving, game-playing, and other instances of ``rational thought''. Nor is it concerned solely with human abilities.
This example is from Intelligence as Adaptive Behaviour by Beer (RSL M91.E00889; PSY KH:B 039). Beer builds a computer-simulated artificial cockroach in order to understand the neural mechanisms underlying some of its basic behviours: walking, wandering, recoil from objects, feeding. But his main interest is the mechanisms underlying the choice between behaviours (page 144). In some cases, one controller may supress or enable another directly by inhibitory or excitatory connections. However (page 152), matters aren't always so simple. One important point of Beer's work is that we can't understand the neural basis of such behaviours without also understanding the dynamics of the body (page 169).
In the work I talked about, Beer starts with Pearson's work on the neural basis of cockroach walking (page 73). Pearson proposed a model for how each leg is controlled, and a suggestion about how control of different legs is coupled. Beer built a computer implementation of that model and found that it correctly predicted fast (tripod) gaits, but not slow ones (page 78). So Beer incorporated an idea - phase-locking - from a more general study on insects. This did predict the slow cockroach gaits, though not those of some other insects (page 81). In summary, Beer implemented an existing model, discovered that it was insufficient, and demonstrated some plausible additions that made it more predictive. Implementing a theory on the computer forces one to consider details that might otherwise be neglected.