Training data
The examples an AI system learns from, such as text, images, recordings or records. What is in the data shapes what the system can do and the mistakes it makes.
An AI system learns much like a student working through practice exercises. The training data is that pile of exercises. For a large language model it is mostly text gathered from books, websites and other sources. For a photo tool it might be millions of labelled pictures.
If a system that identifies birds was trained only on photos taken in daylight, it may struggle with pictures taken at dusk. It never saw enough of them to learn.
Gaps and slants in the data carry through into the results, which is one cause of bias. Training data also has a cut-off date, so a model may not know about recent events unless it can look things up. There are ongoing debates about whether people's writing, art and personal information should be used for training without their permission.
Related: Machine learning, Model (AI model), Bias (in AI), Fine-tuning