Expectation-Based Models of Novelty for Evaluating Computational Creativity

The most broadly agreed-upon definition of a creative artefact is one that is both novel and valuable. This chapter argues for measuring novelty as the violation of observers’ expectations instead of as the objective differences between artefacts. We discuss why models of novelty based on measuring the differences between known and new artefacts are insufficient, and how models based on expectation can address those insufficiencies. We also describe how expectations can influence the generation of new creative artefacts, showing how unexpected discoveries can focus attention and reformulate goals. We base this on cognitive science research that shows this behaviour during creative design in humans. We present formal notation for a simple model of expectation, and give examples from our implementations in the domains of product design and culinary creativity.