liveBook · Manning

Disparate impact

As explained in AI Governance

Disparate impact is a measure of differences in outcome rates between groups, often expressed as a ratio.

Disparate impact measures differences in outcome rates between groups, often as a ratio. It can help check whether an AI system’s decisions produce unfair disparities. section 6.2.4

For employment decisions, the US EEOC uses 0.8 as a threshold; the appropriate value depends on the domain. AI Fairness 360, Fairlearn, and LangFair can compute metrics for compliance dashboards. Measurement does not replace human review or an appeals process. section 6.2.4

How AI Governance covers it

How deep it goes
Defines the ratio and gives a domain-specific threshold, while noting that the right value varies by domain.
Examples
Uses employment decisions to illustrate a 0.8 threshold and names three metric toolkits.
What you can do after
Evaluate group outcome disparities with AI Fairness 360, Fairlearn, or LangFair.
How it connects
Connects fairness testing with compliance dashboards, human review, appeals, and fine-tuning.

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.