Welcome to Quantum Magazine's podcast.

Each episode we bring you stories about developments in science and mathematics.

I'm Susan Vallett.

When you look at a photograph of a cat, chances are that you can recognize the pictured animal,

whether it's ginger or striped,

and you still know it's a cat if the image is black and white, covered in dust specs, or worn and faded.

The cat might be a blur of motion or obscured by tree branches, and you'd still know what it is.

You've naturally learned to identify a cat in almost any situation. In contrast,

machine vision systems powered by deep neural networks can sometimes even outperform humans at

recognizing a cat under fixed conditions, but images that are even a little novel, noisy, or grainy, can throw off those systems completely.

A research team in Germany has now discovered an unexpected reason why.

Humans pay attention to the shapes of pictured objects.

If we look at cat pictures, we know the rough outline of a cat,

whether it's standing, curled up sleeping, or climbing a tree. But deep learning computer

vision algorithms routinely latch onto the object's textures instead. Researchers presented this

finding at the International Conference on Learning Representations last year.

It highlights the sharp contrast between how humans think and machines process information.

It illustrates how misleading our intuitions can be, too, about what makes artificial intelligences tick.

It may also hint at why our own vision

evolved the way it did. Let's take a look at how deep learning algorithms work. Say we present

a neural network with thousands of images that either contain or don't contain cats. The system

finds patterns in that data.

...