AI That Sees
Close your eyes for a second. Now open them. In less than a blink, your brain identified everything around you — the room, the furniture, the faces of people you know. Your brain is incredibly good at seeing and recognizing things. Now here is the challenge AI engineers set for themselves: can we teach a computer to recognize things the same way? Spoiler — they made a lot of progress! Let's see how.
How Computers Learn to See
A camera captures a photo as millions of tiny colored dots called pixels. To you, those dots look like a face. To the computer, they start as just a grid of numbers — a red dot here, a blue dot there. AI learns to find patterns in those numbers. Engineers show the AI thousands, sometimes millions, of labeled photos. This cat is labeled cat. This dog is labeled dog. This stop sign is labeled stop sign. After seeing so many examples, the AI learns which combinations of pixel patterns belong to which object. This kind of AI is called computer vision — vision meaning sight, just like when you use your eyes.
Computer vision is AI trained to find patterns in images. It does not see the way your eyes and brain do. It finds mathematical patterns in pixels and matches them to things it has learned to recognize.
Have you ever taken a photo on a phone and the camera drew a little square around everyone's face automatically? That is computer vision finding faces in the image — so the camera can make sure the faces are in focus and well lit. Some photo apps let you search your pictures by typing dog. The AI scans every photo you have ever taken, finds the ones with a dog shape in them, and shows only those. It recognized your dog — not because it knows your dog, but because it found patterns that match its training for what dogs look like. Face unlock on phones, which you learned about in lesson two, uses this same idea. The AI trained on thousands of faces so it can tell yours apart from a stranger's.
Computer vision is also used in hospitals to help doctors spot unusual patterns in medical scans. It is used in stores to notice when a shelf is running low on products. It is used in farms to spot sick plants in a field. Everywhere people need to look at lots of images quickly, AI vision is being put to work.
Match each computer vision task to a real-world example.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
Computer vision can be wrong sometimes. If the AI was mostly trained on certain kinds of faces or objects, it might not recognize others as well. Engineers work hard on fairness — making sure the AI is trained on a wide variety of examples so it works well for everyone. Remember: AI sees patterns, not meaning. It does not understand what it is looking at the way you do. A very clever matching of patterns is still just patterns.
With a grown-up's help, open a photo app or online photo tool that has a search feature. Search for a person, a pet, or an object. See how many photos it finds — and notice if it ever gets one wrong. Spotting mistakes is a great way to understand what AI can and cannot do.
What does computer vision AI use to recognize objects?
Why might a computer vision AI sometimes make mistakes?
Pixel Detectives
- You will need: a printed photo (any photo), a magnifying glass if you have one, and crayons or colored pencils.
- Look very closely at one small part of the photo — a face, a tree, or a shirt.
- Draw a four-by-four grid of squares on paper. Each square is one pixel.
- Fill in each square with the color you see in that tiny part of the photo.
- Step back and look at your drawing. Does it remind you of the thing you zoomed in on?
- This is exactly what a computer sees when it looks at an image — a grid of colored values. Now you know why AI needs millions of examples to learn what things look like!