Daniel Bisig, what is artificial creativity?
We speak of artificial creativity when human-made systems transcend predictability. Similar to artificial intelligence, there are two types of artificial creativity: weak and strong. The weak type solves a specific problem, while the strong one is a generalist who solves a wide variety of problems. A weak artificially creative machine does something new, albeit unawares. Examples are swarm simulations, chaotic systems and neural networks. A strong artificially creative system reacts to feedback, is inspired and develops its own criteria.
Strong artificial creativity has yet to be explored in-depth. Questions such as how a machine can be made to develop its own values still need to be answered. One theory says that human creativity has an evolutionary background. Transposed to a computer, for example, this means: it must be afraid of falling down and look forward to being powered up. It needs intrinsic motivation.
Machine learning does not engender artificial creativity per se. A computer that learns to compose like Mozart merely imitates his compositional techniques. Among machine learning methods, reinforcement learning seems to be the best way of overcoming such limitations. Machines learn according to the trial-and-error method. My colleague Tatsuo Unemi and I are investigating this in our “Robalz” project. Our aim is to find out how robots acquire aesthetic behaviour and selection criteria through partner selection. This kind of learning behaviour might lead to creativity on a par with human creativity. I suspect and hope that the results of strong artificial creativity will differ significantly from those of human creativity.