Concept Fable

Machine learning

Conclusion first:It lets computers work not only by rules written by hand, but also by learning patterns from data samples, using the learned patterns to predict, classify, generate, or decide on new inputs.

Machine Learning, ML · Machine Learning

Read the story first. There is no rush to give a definition here; let the question grow on its own first.

Fable

The fruit shop apprentice who could not memorize the rules

The fruit shopkeeper at first used three rules to pick sweet pears: the skin should be yellow, the stem should be curved, and it should smell fragrant. On sunny days this method worked well, but once the rainy season arrived, many pears with darkened skin were mistakenly put into the astringent-fruit basket.

The shopkeeper kept adding rules to the wall—color, spots, weight filled a whole page—and the assistant became even more hesitant. The old farmer brought a batch of pears that had already been tasted and had the apprentice record the appearance, aroma, origin and final sweetness or astringency one by one.

The apprentice stopped memorizing new entries and instead compared these examples over and over, gradually discovering that only several clues together were reliable. He made a judgment on an untasted pear first, then let the guest's tasting correct it, and the wrong parts went back into the records.

By the peak pear season, the fruit shop could handle new batches it had never seen before. Only then did the shopkeeper understand that some judgments are not about hardcoding answers, but about learning patterns from enough examples.

Concept Anchor

This story is about: machine learning

It lets computers work not only by rules written by hand, but also by learning patterns from data samples, using the learned patterns to predict, classify, generate, or decide on new inputs.

Machine learning requires training data, features or representations, a model, an objective function, and evaluation. Its strength is handling problems where rules are hard to write by hand; its weakness is being affected by data quality, distribution shifts, and incorrect objectives.

Story Mapping

Handwritten ripe-peach rules: traditional rule-based programs
Baskets of peaches with results: training data
Color, smell, hardness: input features or representations
Just right, too raw, overripe: labels or target outcomes
New peach judgment: model predictions on unseen samples

Andy Ren Xu's judgment

Machine learning suits problems where rules can't be fully written but samples are sufficient. When implementing in an enterprise, first ask whether there's learnable data and acceptable outcomes, not how new the model is.

supervised learningunsupervised learningreinforcement learningmodel training