At least two global “robot art” competitions happening at the moment—the RobotArt contest and Dartmouth College’s Neukom Institute Prizes in Computational Art—the world is perhaps getting nearer to devising a Turing test for creativity. Progresses in machine-learning—a key aspect of artificial intelligence (AI)—are in the throes of making human labour redundant for a host of jobs, including driving and financial services advisory. But can a truly “creative” machine ever emerge? An essay by art historian Martin Gayford, in the MIT Technology Review, examines who far in approaching human creativity AI already is.
Google’s Brain AI project has already come up with a programme that can demonstrate a rudimentary form of “imagination”—it creates digital images by autonomously transposing objects absent in a photograph whose shape resembles that of the subject. In humans, this capability to imagine one thing as something else is perhaps what could explain why mankind came up with constellations. Similarly, AARON, a programme which has been in development since 1973, has now emerged with a better sense of colour than the artist who has collaborated in the development of the programme from the start—of course, it is the artist who gave AARON the set of rules by which it determines the colour weightage for a particular piece it creates. Nevertheless, it is clear that these advances in machine-learning make the complex interplay between factors that inspire human painters seem less daunting to programme.