Analysing Psychological Data by Evolving Computational Models

We present a system to represent and discover computational models to capture data in psychology. The system uses a Theory Representation Language to define the space of possible models. This space is then searched using genetic programming (GP), to discover models which best fit the experimental data. The aim of our semi-automated system is to analyse psychological data and develop explanations of underlying processes. Some of the challenges include: capturing the psychological experiment and data in a way suitable for modelling, controlling the kinds of models that the GP system may develop, and interpreting the final results. We discuss our current approach to all three challenges, and provide results from two different examples, including delayed-match-to-sample and visual attention.