Generative Music Evaluation: Why do We Limit to 'Human'?

Objective evaluation of aesthetic subjective judgements is, by very definition, tricky business. This may be the reason many generative music systems merely skip the evaluation part entirely. Typical evaluation systems that do exist are often variations of a Turing-style discrimination test whereby the autonomous system must convince a human interrogator that what it has produced was created by a human. In this paper we propose that this may be selling the computational systems short. With the ever increasing power of computational machines, why should we limit these new intelligent systems to a human level of creativity we barely understand ourselves? We consider that autonomous, statistical evaluations would be superior to the traditional human judgement tests. We describe a number of evolutionary systems that have been applied to compositional tasks and propose that these are the most suitable methods to use in developing autonomous evaluation measures.