AI model drift

noun

AI model drift occurs when a trained AI model’s predictions become less accurate because the data it encounters in the real world differs from the data it was trained on. This happens because user behavior, language trends, or external factors evolve, making the model’s assumptions outdated.

For example, a language model trained in 2020 may struggle with new slang or cultural references in 2024. Similarly, an image generator might produce outdated fashion styles if its training data doesn’t include recent trends. Model drift forces developers to retrain or fine-tune models periodically to maintain performance.

In practice, this means companies must monitor their AI systems continuously. If left unchecked, drift can lead to poor user experiences, incorrect recommendations, or even safety risks. For instance, a medical AI diagnosing X-rays might miss new virus strains if its training data is too old. Typically, retraining costs time and resources, but ignoring drift costs more in the long run.

A chatbot trained on 2021 data starts giving outdated answers about smartphone models in 2024.

See also: model fine-tuning

Découvrez nos autres services

ElpisIA, c’est tout un univers — explorez nos autres services.