Robots beat art experts at their own game

Robots have made connections between paintings that have taken experts years to realise. David DeHetre via Flickr

Researchers at Rutgers University in New Jersey have created a computer programme designed to analyse and identify paintings. When shown an image of an unfamiliar work, the algorithm can correctly name the artist 60% of the time and identify the style or movement 45% of the time. To train the programme, the computer scientists Babak Saleh and Ahmed Elgammal created a database of 80,000 paintings by more than 1,000 artists spanning 15 centuries. The programme can draw connections between works that have taken historians years to realise. “The ultimate goal of our research is to develop a machine that is able to make aesthetic-related semantic-level judgments,” the researchers say. Still, art historians need not worry that they will be usurped by machines any time soon. The computer had difficulty distinguishing between artists and those they influenced, such as the American Impressionist Childe Hassam and the French Impressionist Claude Monet.