Computer vision is one of the trickier aspects of artificial intelligence; it's one of the things that we humans are really good at, but that machines have a lot of trouble with. If we see a duck then we can immediately identify it as a duck, no matter what angle we see it from, whether it's swimming or in flight, or if it's partially obscured.
For a computer such a feat of identification is near-impossible, but there are great advances being made with computer vision; for a demonstration of what a deep neural network can 'see', head over to this site and try uploading some photos. It's basic and prone to error, but it's heading in the right direction.
Identifying the subjects in a photo is all very well, but how does a computer cope when presented with art? Thankfully there's a site devoted to that very question. Novice Art Blogger is a project by artist and researcher Matthew Plummer-Fernandez, in which he's exposing a deep neural network to art for the first time.
Using state-of-the-art deep learning algorithms, the system looks at each piece of art and then tries to describe what it sees, and the results are strangely fascinating; its attempts to try to make sense of an image and provide a literal breakdown of what's in it can be unintentionally hilarious, and often it'll see things hidden within noise that we'd never spot, or perhaps ask for some cocoa.
In its current state, Novice Art Blogger comes across a little like a toddler trying to explain artwork, and some of the comparisons it comes out with are delightfully daft and not a little surreal. As it learns, though, it should hopefully become a lot more adept at its task; could it become the first artificial art critic a few years down the line?
Words: Jim McCauley
Jim McCauley is a writer, editor and occasional podcaster, and is available for children's parties.
Like this? Read these: