Thursday, 27 November 2014

I don't know much about robots, but I know what I like

Is it art?
I've always had mixed feelings about the Turing test. This is (a variant on) the mechanism proposed by Alan Turing (you know, the one who looks like Benedict Cumberbatch) to decide if computers could be considered to be intelligent. As I've pointed out previously, the way the test is administered is far too lax. And part of the problem is the requirement of a judge to decide if the entity he or she is communicating with is a person. This is inevitably a subjective decision, and highly dependent on the quality of the dialogue the judge uses.

Now, though, we've got a whole new level of silliness, with a Georgia Institute of Technology professor suggesting that in testing for machine intelligence we should also 'ask a machine to create a convincing poem, story or painting.' What remarkable twaddle. Take the 'art' aspect. We can't agree on which humans can create a convincing painting, so how could we possibly use this as a test? By the standards of modern art, any random collection of paint marks on a canvas could be considered a 'convincing painting' - it purely depends on what those judging persuade themselves is valid and/or meaningful and important. There is no standard against which to measure what the computer produces.

Let's be clear - I am not saying this because I think that art that doesn't require skill and craft is worthless (although I do think this). Merely saying that there is no metric that could be possibly be used. What, for instance, if the computer produced the image shown here. If this had been done by, say, Mark Rothko, it would be classed as a convincing painting. As it happens I did it pretty randomly on an iPad in 2 minutes - so it's not classed as a convincing painting. The metric is not the nature of the artwork itself, but who produced it. Modern art is essentially a celebrity phenomenon. And that means the process is bound to fail.

No comments:

Post a Comment