AI Trust Glossary · Canonical Definition
Bias (AI Bias)
Systematic errors in AI output that result from prejudiced assumptions in training data or model design - causing the model to consistently favor or disfavor certain groups or outcomes.
Explanation
Bias is directional, not random. A biased hiring model systematically disfavors candidates from certain demographics. Bias enters through training data (historical inequalities encoded as features), model architecture, or evaluation metrics that do not measure what matters.
Why it matters
Biased AI agents cause real harm, undermine public trust, and expose deploying organizations to legal liability under anti-discrimination laws and the EU AI Act. Detecting bias requires specific measurement techniques beyond standard accuracy metrics.
How Borealis uses it
Bias evaluation is incorporated into the audit process for high-risk agent categories. Agents in hiring, lending, and healthcare require evidence of bias testing before certification. Bias findings affect constraint adherence and behavioral consistency scores.
See also