Scientific American ran a short article today entitled "Computers Can't Show You the Monet" that collected many of the random ideas circulating through my mind into one coherent stream. It in turn is a summary of an article in the journal Computers & Graphics. In the study, non-art-expert humans and computers tried to place 275 paintings into 11 artistic periods. The computers looked a brush patterns, colors, the way the patterns were laid down on the canvas, just about any quantifiable quality of the paintings. The net result - humans did immensely better than the computers.
This result is interesting to me because one might assume that with sufficient time and training, a heuristic programming approach would become proficient in sorting the art. But there is also a nagging suspicion that a fundamental underlying difference that may pose an insurmountable hurdle to objective machine classification: humans classify by emotional reaction.
The article concluded with a wonderful summary:
Computer algorithms judged the art by obvious and quantifiable parameters, such as the way the paint was laid on the canvas, or the color composition. But humans classified art based on complex psychological evaluation. We ask questions such as, who is in the image? And, what emotions are being portrayed in the scene? This kind of analysis is crucial for correctly identifying art—because even non-expert people were right two thirds of the time, far better than their computer competitors. And that makes sense: ultimately, art is about our emotional reaction to a Starry Night or a Girl With A Pearl Earring. But to a computer it’s all just brushstrokes.
So what are your thoughts?