Developing and evaluating cognitive architectures with behavioural tests

Abstract

We present a methodology for developing and evaluating cognitive architectures based on behavioural tests and suitable optimisation algorithms. Behavioural tests are used to clarify those aspects of an architecture’s implementation which are critical to that theory. By fitting the performance of the architecture to observed behaviour, values for the architecture’s parameters can be automatically obtained, and information can be derived about how components of the architecture relate to performance. Finally, with an appropriate optimisation algorithm, different cognitive architectures can be evaluated, and their performances compared on multiple tasks.