Try writing the rule

Imagine teaching a computer to find dogs in photographs.

You might begin with a simple rule: a dog has four legs. But a dog behind a sofa may show only two, while a photograph of a horse shows four.

Try fur. A hairless dog fails the rule, while a cat passes. Try a tail, floppy ears, or a long nose. Some dogs lack each feature, and other animals have them.

You can usually look at a new photograph and decide in a moment. Yet the knowledge you use is spread across shape, texture, pose, surroundings, and many exceptions. It does not arrive as one clean sentence you can give a computer.

The ability is real. The complete rulebook is missing.