In an AI risk management context, what does 'concept drift' refer to?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal — concept drift means the relationship between inputs and outcome changes after training, so patterns your model learned no longer describe current reality. Loan-default model trained before recession can fail after recession because rules changed, not just data.
Full Explanation
Concept drift refers specifically to the change in the underlying relationship between input features and the target variable after model training, meaning the patterns the model learned no longer accurately describe current reality. For example, a loan-default model trained before recession may have learned patterns from a pre-recession economy that no longer hold post-recession. The gradual shift in the statistical distribution of input features used by the model is data drift, not concept drift, because the input distribution changes without the input-to-target relationship changing. The degradation of model performance caused by software bugs introduced during redeployment is implementation or deployment issue, not concept drift. The divergence between model predictions and human expert judgment over time is not standard machine-learning definition of concept drift.