AI learns patterns from examples — Technology, 7–10 years
Some computer programs find patterns in many examples instead of receiving every rule one by one. This can be useful, but it does not make the machine understand everything like a person.
Learning a pattern
In this kind of AI, people provide examples and the computer looks for repeated clues. After training, it uses those clues on a new case, such as deciding whether a picture contains a cat, even though it has never seen that exact picture before.
Why learn from examples?
Some tasks are easy to show but hard to explain with exact instructions. It is difficult to write every rule for recognising a face or a handwritten number. Examples let a program discover useful clues, although people still choose the task and check the results.
Sorting red and blue apples
Give a program four labelled examples: small red, large red, small blue and large blue apples. It notices colour, not size, if all red examples are labelled red and all blue ones blue. A new medium red apple is then predicted as red.
It can be confidently wrong
A neat answer can feel trustworthy, especially when a computer gives it quickly. But the program may have learned a poor clue, or its examples may be incomplete or unfair. Confidence is not proof that the prediction is correct.
Recognising and recommending
AI pattern systems can help sort photographs, suggest a song or turn speech into written words. They work best when the examples match the real situation, so people should be able to question important results rather than accept every answer automatically.
Keep exploring
Other languages
Loading MyLeoNes™…