What is Machine Learning?
Machine learning is how computers learn to do a task by studying examples, instead of being told the exact rules.
Explain like I'm new to AI
Normally, software follows rules a human wrote. Machine learning (ML) flips this: you give the computer lots of examples, and it figures out the rules by itself.
Want to detect cats in photos? Instead of describing "a cat has pointy ears, whiskers...", you show the system thousands of labeled cat and non-cat photos, and it learns what makes a cat a cat.
Mental model
Think of ML as learning by example, like a child. Show a toddler enough dogs and cats, and eventually they generalize — they can identify a breed they've never seen. ML does the same: it generalizes from examples to handle new, unseen cases.
How it works
- Collect data — many examples (often with labels, like "spam" / "not spam").
- Choose a model — a flexible math function with adjustable knobs (parameters).
- Train — the model makes predictions, measures how wrong it is (the loss), and adjusts its parameters to do better. Repeat many times.
- Evaluate — test it on data it hasn't seen to check it actually generalized.
- Predict (inference) — use the trained model on new, real inputs.
Real-world example
A bank predicts whether a loan will be repaid. It trains an ML model on years of past loans (each labeled "repaid" or "defaulted"). The model learns which patterns of income, history, and amount predict repayment, then scores new applicants.
Technical explanation
ML comes in three broad flavors:
- Supervised learning — learn from labeled examples (input → correct output). Most common.
- Unsupervised learning — find structure in unlabeled data (e.g. clustering).
- Reinforcement learning — learn by trial and error via rewards.
Training minimizes a loss function using optimization (typically gradient descent), adjusting parameters until predictions match the data well without overfitting (memorizing) the training set.
Common mistakes
Believing more data always beats a better approach. Garbage or biased data produces a garbage model — data quality and representativeness matter as much as quantity.
- Confusing memorizing the training data (overfitting) with truly learning.
- Skipping evaluation on unseen data, then being surprised it fails in production.
When to use it
- The task has patterns in data but no clean rulebook (vision, language, recommendations, forecasting).
- You have (or can get) enough representative examples.
When NOT to use it
- A simple deterministic rule already solves the problem.
- You have too little data, or you need fully explainable, guaranteed outputs.
Alternatives
- Rule-based systems and classic algorithms for well-specified problems.
Quick quiz
Related concepts
- What is Deep Learning? — Deep learning is machine learning using many-layered neural networks that learn features automatically.
- What is a Model? — A model is the trained artifact that stores what an AI system learned and turns new inputs into predictions.
- Training vs Inference — Training is teaching a model from data; inference is using the trained model to make predictions.
- Supervised, Unsupervised, and Reinforcement Learning — Supervised learns from labeled examples, unsupervised finds structure without labels, and RL learns from trial-and-error rewards.
Last reviewed: 2026-08-30 · Written by ByHeart AI · Reviewed by ByHeart AI