Examples
Worked examples
- Is an instance
A Gender Shades-style audit of a face-classification model reporting accuracy by skin-tone and gender.
- Is an instance
A statutorily required NYC Local Law 144 bias audit of an employment-screening tool.
Counter-examples
Looks similar, but isn't
- Not an instance
An aggregate accuracy report with no sub-group breakdown.
- Not an instance
A model card section listing potential biases without measurement.
Editorial commentary
Bias audits measure fairness metrics (equal-opportunity, demographic parity, calibration parity, predictive-parity gaps) and qualitative failure-mode analyses across pre-specified groups. The choice of metric is itself ethically loaded (Chouldechova's impossibility theorems show that distinct fairness metrics generally cannot be simultaneously satisfied). NYC's Local Law 144 and several US-state laws now mandate bias audits for automated employment decision tools.
References
- Buolamwini, Gebru, 'Gender Shades' (FAccT 2018); Chouldechova, 'Fair prediction with disparate impact' (Big Data, 2017); NYC Local Law 144.
Also known as
fairness audit · algorithmic bias audit
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Bias audit (model)"
vocab-term-identifier="https://casrai.org/dictionary/term/bias-audit-model" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"name": "Bias audit (model)",
"identifier": "https://casrai.org/dictionary/term/bias-audit-model",
"description": "An audit specifically focused on disparate model performance across demographic, geographic, or contextual sub-groups, including testing for direct, proxy, and intersectional disparities.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-and-ml-research-outputs/",
"url": "https://casrai.org/dictionary/term/bias-audit-model",
"sameAs": [
"fairness audit",
"algorithmic bias audit"
],
"license": "https://creativecommons.org/licenses/by/4.0/"
}
