The magazine MAQ September 2018 MAQ Magazine November 2018 | Page 175

MAQ/November 2018 / 07

The test – and the machine art – comes from Rutgers University's Art and AI Laboratory, led by Professor Ahmed Elgammal. They designed a creative adversarial network (CAN),

training it on 75,753 artworks from abstract expressionism to synthetic cubism. Unlike a GAN, the CAN produces pictures slightly – but not too much – different from its data set.

Groups including artists and art history students have been unable to pick whether picture was created by a human or CAN, and sometimes rated the CAN art more highly than the other contemporary or abstract expressionist examples. The CAN

works were exhibited twice last year, one has been auctioned for $US16,000, and a first-edition series is on sale.