Learn AI · Level 2 · How AI works · step 5 of 16 · word 2 of 5
What is data labelling in AI?
Data labelling is the work of tagging examples with the right answer, such as marking a photo "cat" or a payment "fraud", so a model can learn from them. Much of it is done by people.
In plain English
Supervised learning needs an answer key. Someone has to go through thousands of examples and write down the right answer for each: draw boxes round cars, mark emails as spam, rate which chatbot reply is better. Those tags are labels.
An example
Chatbots need labels too. In RLHF, people rank pairs of answers so the model learns which ones humans prefer. Labelling is now big business: in 2025 Meta invested in Scale AI, a labelling company, in a deal that valued it at over $29 billion.
Watch out
Labels can be wrong. A 2021 MIT study found errors in at least 3.3% of the labels in ten famous test datasets. A model trained or tested on bad labels learns, or is scored on, the mistakes. More in how a machine learning model is built.