Model or data drift: Which contributing factor best describes this cause?

Gradual performance loss as real-world data changes against a static model is model or data drift.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. Model or data drift, where changes in real-world input or relationships gradually erode the performance of an unchanged deployed model. ✓

   Correct: gradual degradation from changing real-world data against a static model is the definition of model or data drift.
2. Brittleness, where a model fails abruptly when inputs stray outside the narrow conditions it was trained to handle.

   Plausible but wrong: brittleness describes sudden failure on small input changes, not the gradual, time-based degradation described here.
3. Insufficient testing, where inadequate pre-deployment evaluation lets defects reach production that better test coverage would have caught earlier.

   Plausible but wrong: insufficient testing concerns defects present at release, not performance loss driven by later real-world change.
4. Lack of quality data, where flawed or unrepresentative training data causes the model to perform poorly from the very start of its operation.

   Plausible but wrong: poor training data causes weakness from the outset, whereas here performance declined over time from shifting inputs.

How to remember it

Gradual performance loss as real-world data changes against a static model is model or data drift.

More Understanding How to Govern AI Development questions

• Recording datasets: Which practice best satisfies all three needs at once? →
• The likelihood that a harm will occur and the severity: What two dimensions does this matrix combine to help →
• Causes such as drift: What is the primary reason this collaboration is emphasized? →
• All 135 Understanding How to Govern AI Development questions →