The Future Is Here
We may earn a commission from links on this page

Computers Think These Are Real Animals and Objects

We may earn a commission from links on this page.

Even computers can be fooled by optical illusions. While computer vision is rapidly advancing, this set of bizzare images can fool even the best algorithms into thinking that they're real objects.

The images were generated by Jeff Clune fromthe University of Wyoming in Laramie when he applied an image-recognition algorithm called a deep neural network (DNN) to a second algorithm designed to evolve pictures. When the second algorithm is used along with human judgement, the result is images that become clearer and clearer; that didn't happen when a computer took control.


But when these odd images were shown to AlexNet—a computer vision algorithm made by researchers at the University of Toronto who were subsequently hired by Google—it thought they were real objects, from electric guitars to penguins. They look like abstract images to humans, but to computers, they look just like the original objects. New Scientist explains why that's the case:

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.


In other words, the distinctive features that make up these images are the basis for computer vision in the first place—it's just that humans don't interpret the images in that way. Perhaps the best examples are the remote control, which clearly shows the pattern of buttons, and the baseball, which resembles the stitching we're all familiar with.

Of course, it may pay to help computers realize that these aren't actually the real-deal, as they could be used to fool security or AI systems in the future. But for now, it's kinda nice to know that computers fall foul of optical illusions in much the same way as we do. [arXiv via New Scientist]