Supervised learning means teaching an AI with examples that already have the right answer attached. The model guesses, checks its guess against the answer, and adjusts, until it can give good answers for new examples it was not trained on.
Think of flashcards with the answer on the back. You look at the front, guess, flip the card and see if you were right. After enough cards, you can answer new ones without flipping. In supervised learning, the AI is doing the flashcards, and the answers on the back are called labels.
In a sentence: “Our spam filter was built with supervised learning: people marked thousands of emails as spam or not spam, and it learned from those answers.”
What the word means
The plain point: “supervised” means every example comes with the answer, like a teacher marking work.
Google’s introduction to machine learning opens with exactly that: “Supervised learning uses labeled data to train models that predict outcomes for new, unseen data.”1
Three words do the work:
- Label: the right answer for one example. Google’s glossary calls it “the ‘answer’ or ‘result’ portion of an example.”2
- Labelled data: a set of examples that all carry labels. It is one kind of training data.
- Loss: how far the model’s guess was from the label. Training keeps shrinking it, which is how the model’s parameters get set.
A simple example
The plain point: spam filters are the classic case.
People mark thousands of emails as “spam” or “not spam”. Those marks are the labels. The model studies the emails and their labels, then sorts new emails on its own. Google names spam and rain forecasts as typical jobs.3
Who adds the labels: data annotation
The plain point: labels are usually added by people, and that work has a name.
Adding labels is called data labelling or data annotation. IBM describes it as tagging raw files so a model knows what they show.4
You may have done some yourself. IBM points to reCAPTCHA, the “prove you are human” puzzles that asked people to pick out the photos with cars in them.5
For big AI systems, some of this work comes from paid crowd workers. Anthropic, for one, lists “Data that our users or crowd workers provide” among its training data.6 How an AI is actually built covers the people behind this work.
A supervised model is only as good as its labels. Careless or slanted labelling teaches careless or slanted answers.
Often confused with
- Unsupervised learning: the model gets no answers and looks for patterns by itself. Google defines it as learning that “finds patterns in unlabeled data.”2
- Learning from human feedback: chatbots are also tuned using people’s judgements of their answers, which is a different method.7 One side effect, a habit of agreeing with you, is covered in how AI influences you back.
- Fine-tuning: extra training on your own examples, often with answers attached, on top of a finished model. See fine-tuning meaning.
- Memorising: a model that learns its flashcards too exactly fails on new ones. That is overfitting.
Questions people ask
What is the difference between supervised and unsupervised learning?
Supervised learning learns from examples that come with the right answer attached; unsupervised learning looks for patterns in examples that have none. Google’s glossary describes unsupervised learning as finding patterns “in a dataset, typically an unlabeled dataset”.8 Unsupervised learning meaning covers the other side.
What is an example of supervised learning?
A spam filter is the classic one. Google names jobs like “identifying spam or predicting precipitation”: in both, the model learns from past examples whose answers are already known.3
Do chatbots like ChatGPT use supervised learning?
For part of their training, yes. In OpenAI’s 2022 paper on teaching GPT-3 to follow instructions, people wrote example answers, the model was fine-tuned on them “using supervised learning”, and a later stage used people’s rankings, called reinforcement learning from human feedback.9
Learn more on this site
- How machines learn, from the start: A short history of AI.
- Fairness and labels: AI bias meaning.
- The basics: the AI glossary and AI explained for complete beginners.
- Related: Reinforcement learning meaning.
- Related: Unsupervised learning meaning.
- Related: What Is Sentiment Analysis? Meaning in Plain English.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What makes learning ‘supervised’?
- A person watches the AI the whole time it runs
- Every example comes with the right answer attached
- The AI learns only from rewards and penalties
Show the answer
Answer: Every example comes with the right answer attached. The page says supervised means every example carries its answer, like flashcards with the answer on the back.
2. In the spam filter example, what are the labels?
- People’s marks of ‘spam’ or ‘not spam’
- The subject lines of the emails
- The number of emails in the inbox
Show the answer
Answer: People’s marks of ‘spam’ or ‘not spam’. The page says people mark emails as spam or not spam, and those marks are the labels the model learns from.
3. What does the page’s takeaway warn about?
- Labels make a model forget older lessons
- More labels always slow a model down
- A model is only as good as its labels
Show the answer
Answer: A model is only as good as its labels. The page says careless or slanted labelling teaches careless or slanted answers.
Related terms: Scaling laws · Singularity (technological singularity) · Superintelligence (ASI) · Synthetic data · Training data · Transformer
- 1Google for Developers, Supervised Learning, Introduction to Machine Learning, last updated 25 Aug 2025, read at source 1 Oct 2026“Supervised learning uses labeled data to train models that predict outcomes for new, unseen data.” ↑
- 2Google for Developers, Machine Learning Glossary: ML Fundamentals, read at source 1 Oct 2026.↑a↑b
- 3Google for Developers, Supervised Learning, last updated 25 Aug 2025, read at source 1 Oct 2026supervised learning can be used for tasks like “identifying spam or predicting precipitation.” ↑asupervised learning can be used for tasks like “identifying spam or predicting precipitation.” ↑b
- 4IBM, What Is Data Labeling?, published 28 Sep 2021, updated 23 Jan 2026, read at source 1 Oct 2026it “involves identifying raw data, such as images, text files or videos and assigning one or more labels to specify its context.” ↑
- 5IBM, What Is Data Labeling?, read at source 1 Oct 2026users were asked to “identify all the photos containing a car to prove that they were human”. ↑
- 6Anthropic Privacy Center, How do you use personal data in model training?, 16 Mar 2026, read at source 1 Oct 2026.↑
- 7Sharma et al., Towards Understanding Sycophancy in Language Models, arXiv, 20 Oct 2023, read at source 1 Oct 2026“Human feedback is commonly utilized to finetune AI assistants.” ↑
- 8Google for Developers, Machine Learning Glossary: ML Fundamentals, undated glossary, read at source 12 Oct 2026unsupervised machine learning is “Training a model to find patterns in a dataset, typically an unlabeled dataset.” ↑
- 9Ouyang et al. (OpenAI), Training language models to follow instructions with human feedback, arXiv:2203.02155, submitted 4 Mar 2022, read at source 12 Oct 2026“we collect a dataset of labeler demonstrations of the desired model behavior, which we use to fine-tune GPT-3 using supervised learning”, then “reinforcement learning from human feedback” ↑
Sharma et al., Towards Understanding Sycophancy in Language Models, arXiv, 20 Oct 2023, read at source 1 Oct 2026.
The comparison on this page (flashcards with the answer on the back) is this site’s own way of explaining, not a claim from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK