Statistical Evaluation of Process-Centric Computational Creativity

We adopt a process-centric approach to computational creativity, based on a model of people’s innate ability to process analogical comparisons. A three-phase model of analogical reasoning is adapted to function as an analogy generating machine. It is supplied with two distinct knowledge-bases containing many domain descriptions, with the aim of generating novel analogies – potentially even creative ones. However, because our approach to computational creativity does not have the usual ”inspiring set”, evaluating its output can not be performed by comparison to this inspiring set. Our generic approach to evaluating process-centric computational creativity uses a number of nonparametric statistical techniques. After the creative artefacts are generated, human raters assess these artefacts for the qualities of creativity (quality, novelty etc). We describe the results of two experiments that were conducted on these two collections of domains. The analogies generated on the two collections are analysed and difference in the two result sets are assessed. We argue that true creativity can only be assessed after the creative artefacts are generated. Evaluating creativity only by reference to the inspiring set runs the risk of overlooking creative artefacts that differ from the inspiring set and may under-estimate a model’s creativity.