Systematic errors in AI outputs resulting from prejudiced training data or flawed algorithms, leading to unfair outcomes.
Systematic deviations in AI outputs that unfairly favor or disadvantage particular groups - stemming from skewed datasets, flawed labeling, or mis-specified objectives - and requiring detection, measurement, and mitigation.
A facial-recognition system trained mostly on light-skinned faces shows higher error rates for darker-skinned individuals. The vendor rebalances its training dataset and deploys ongoing bias-monitoring dashboards to ensure equitable performance across all skin tones.
