A phone taking a photo of a fluffy grey cat and showing the word cat on its screen.A phone taking a photo of a fluffy grey cat and showing the word cat on its screen.
1 / 4
0:00
0:00

How Computers Learn from Examples · Episode 0 of 3

Why does a computer mix up a cat and a fox?

Discover how a computer learns animals from training pictures, and why the patterns of a cat and a fox can trick it into calling a fox a cat.

18Views

Episode text

Why does a computer mix up a cat and a fox?

Have you ever seen a phone look at a photo of an animal and tell you its name? Picture a soft, fluffy cat in that photo, with warm fur you could almost stroke. The phone looks, thinks for a moment, and says cat.

Now picture a fox in a photo instead. The tiny computer inside the phone looks at the fox and says cat too! Why does a computer mix up a cat and a fox?

A computer has no eyes like yours, and it has never stroked a cat. To a computer, a picture is thousands of tiny dots of color. It needs help to learn which dots make a cat and which dots make a fox.

So the people who build it give the computer training, which means showing it lots of pictures with the right name on each one. They show it thousands of cats with the word cat, and thousands of foxes with the word fox. Training is how the computer learns from examples.

During training, the computer looks for a pattern, which means a look that comes back again and again in many pictures. In the cat pictures, it finds pointy ears, whiskers, soft fur and a long tail. The computer keeps each pattern as a clue for later.

Then the computer gets a brand new picture. It checks the picture for every pattern it learned and picks the name whose patterns fit best. That is how it names an animal in a photo it has never seen before.

Here comes the tricky part. Foxes have pointy ears, whiskers, soft fur and a long tail too! Cats and foxes share most of the patterns the computer knows. Which pattern do you think could still tell the computer it is looking at a fox?

A long, pointed nose! When a fox curls up asleep and tucks its nose under its tail, that clue is hidden. Then the fox shows only patterns that cats have too, so the computer picks the name cat. It does not know what a fox really is; it only matches patterns.

The computer also learns only from its own training pictures. Imagine the people who chose them showed foxes running through snowy woods, but never a fox curled up asleep. Then the computer has no pattern for a sleeping fox, and it cannot tell that picture is a fox.

Training, pattern. A computer mixes up a cat and a fox because the training people give it teaches it only patterns, and cats and foxes share most of those patterns.

Goodnight, clever one, and the next time you meet a cat or a fox, enjoy every little difference your own eyes can spot.

Technology & Transport

See all
Technology & Transport
How Computers Learn from Examples
Part of "How Computers Learn from Examples"

How Computers Learn from Examples

0 of 3 done
In this series