Art, Science, and Artificial Creativity

Art and Science, often contrasted as e.g. Snow’s (1959) “two cultures”, both share methods with design. AI models are trained with the broadly scientific aim that a single model can be fitted to the data and the data adequately represent the world, but this is unlikely to be possible in many cases, and I will make a case that that this approach is fundamentally wrong for use in design. Creativity is frequently defined (e.g. Newell, Shaw and Simon 1958) as consisting of both novelty and value; the race toward a better, larger model and minimising error is fundamentally limited in both these criteria. I will contrast how each is typically approached in generative AI with how they are achieved in typical human creative situations, using several cases from my own work in both art and engineering as examples to illustrate mechanisms that produce creative leaps. As a result, where much research is focused on computation on representations, more is needed on the way in which the world can be represented in different ways, and how imperfect representations may be used. Methods of design have traditionally done this, and so can inform use and development of AI.