An artificial intelligence system created by computer scientists at Rutgers University has picked what it thinks are the most creative and influential paintings in history — and done a pretty good job, too. While determining the quality of an artistic work has generally been an exclusively human field, Ahmed Elgammal and Babak Saleh think they've found a way to objectively determine some aspects of creativity — though they caution that "our goal is not to replace art historians' or artists' role in judging creativity of art products."
Their system looked at a collection of over 62,000 paintings and graded them based on hundreds of artistic concepts, from contrast and proportion to stroke type and subject. Creative paintings, the researchers theorize, will not only show concepts rarely seen before, but will be influential, causing those concepts to be seen more often afterwards. The computer crunched the numbers and produced a creativity measurement for each painting — and it did, in fact, highlight many paintings and artists that critics laud as unusually creative and influential, including Vermeer, Goya, Monet, Mondrian and more.
Elgammal and Saleh say their AI system is another tool critics can use, since it's something of an objective measure. But they caution that it's very limited right now in the criteria it can detect, the library of artworks it has access to, and the definition of creativity it's working with.
Their paper, which will be presented at the International Conference on Computational Creativity at the end of June, can be read for free on Arxiv.