Show it instead

If people cannot write all the rules, there is another thing to try.

Show the machine many photographs marked dog or not a dog. Let it guess. Compare the guess with the answer, adjust the machine slightly, and repeat.

Nobody writes a complete definition of what every dog looks like. The machine develops its own adjustable pattern from the examples.

This is machine learning: improving a task through examples or feedback instead of receiving every rule in advance.

In this example, people still choose the task, collect the photographs, provide the labels, and decide whether the result is good enough. Human judgement moves from writing every rule to shaping the learning process.