Computers are starting to identify objects with near-human levels of accuracy, enabling them to do everything from creating automatic picture captions to driving cars. But now a collection of bizarre optical illusions for these artificial-intelligence systems (AIs) has revealed that machines don’t see the same way we do, which could leave them vulnerable to exploitation.
Image-recognition algorithms learn to recognise objects by training on a large number of images and identifying patterns that mark out a cat from a coffee cup, for example.
Jeff Clune of the University of Wyoming in Laramie and his colleagues wanted to know if they could hook up a particular type of image-recognition algorithm called a deep neural network (DNN) to a second algorithm designed to evolve different pictures.
Such genetic algorithms, working in conjunction with human judgement, have previously created images of apples and faces, so Clune wondered if replacing the human with a DNN, to work alongside the genetic algorithm, would work as well, resulting in a computer that could generate creative pictures by itself.
“We were expecting that we would get the same thing, a lot of very high-quality recognisable images,” Clune says. “Instead we got these rather bizarre images: a cheetah that looks nothing like a cheetah.”
It turned out that the genetic algorithm produced images of seemingly random static that AlexNet declared to be pictures of a variety of animals with more than 99 per cent certainty (see picture). Other images, generated in a different way, look like vaguely evocative abstract art to humans, but fool AlexNet into seeing a baseball, electric guitar or other household object.
The algorithm’s confusion is due to differences in how it sees the world compared with humans, says Clune. While we identify a cheetah by looking for the whole package – the right body shape, patterning and so on – a DNN is only interested in the parts of an object that most distinguish it from others. “It’s almost like these DNNs are huge fans of cubist art,” says Clune. (1)
A computer can be fooled by static.
And see a cheetach where we see nothing but dots…
The humans can be impressed by static.
And see a flower where only dots exist…
The difference cannot be expressed in words.
It can only be experienced.
And this is exactly why it is so fundamental…