Explaining Machine Learning to a Child: A Food‑Classification Example

A father is asked by his child, "What is machine learning?" and struggles to answer, prompting a clear explanation drawn from a Quora response by computer‑science PhD Daniel Tunkelang.

The explanation starts with a classification problem: teaching a computer to distinguish tasty (positive) from disgusting (negative) foods using labeled training examples. Each example is described by binary features such as sweet, salty, and vegetable.

With the training data, the computer learns a model—a linear scoring function that assigns weights (e.g., sweet = 3, salty = 1, vegetable = ‑1). The sum of the weighted features yields a score; a threshold (decision boundary) classifies the food as tasty if the score exceeds the threshold, otherwise as disgusting.

The example shows how the chosen weights make all positive examples score ≥ 2 and negatives ≤ 1, illustrating a perfect fit on the training set but warning about over‑fitting, where the model may not perform well on new data.

Generalisation is discussed, noting that simpler models often generalise better (Occam’s razor). An alternative decision‑tree model is presented, using yes/no questions (e.g., "Is it a vegetable?") to reach a classification, emphasizing the need to limit tree depth to avoid over‑fitting.

The article concludes with the hope that the child can grasp these machine‑learning ideas through the food analogy.