Last updated:
The degradation of an AI model's predictive performance over time due to changes in real-world data or the environment.
Drift occurs when the statistical properties of the data the model sees in production no longer match the data it was trained on. AI governance requires post-deployment monitoring to detect drift, as a model that was accurate and fair at launch can become unreliable or biased as external conditions change.
Real world example:
A retail AI trained on 2019 spending patterns experiences model drift during the 2020 pandemic because consumer behavior changed so radically that the model's stock predictions became useless.
