Homogeneity Amplification Index (HAI)

GMI-HCM-003Human Capital ManagementIntegration-RequiredPattern O-25: Selection Bias Amplification

Definition

The degree to which AI-screened candidate slates are more demographically and experientially homogeneous than the original applicant pool, measured as the proportional reduction in diversity from pre-screen to post-screen.

The Measurement Gap

AI screening tools are trained on historical hiring data, and historical hiring data reflects historical hiring patterns. When an algorithm learns that certain candidate profiles have been selected more frequently in the past, it amplifies that pattern in future screening, producing slates that are progressively more homogeneous than the applicant pool they were drawn from. This amplification operates independently of any explicit demographic variable because proxy correlations embedded in resume features, educational backgrounds, and career trajectories serve the same narrowing function.

Legacy diversity metrics measure the composition of the hired workforce but cannot detect whether the screening algorithm itself is concentrating the pipeline. Adverse impact ratios compare selection rates across groups but operate at a binary level and miss the subtler homogenization that occurs across multiple dimensions simultaneously. EEO reporting captures outcomes but not the mechanism that produced them. A hiring team can report acceptable diversity numbers overall while the AI screener systematically narrows the pipeline in ways that reduce cognitive diversity, experiential range, and non-traditional talent pathways: all invisible to standard compliance dashboards.

The regulatory implications are direct. NYC Local Law 144 and EEOC guidance require organizations to detect and address disparate impact in automated screening. HAI provides a continuous measurement of the homogenization effect rather than a binary pass/fail assessment.

Decision Supported:

Is our AI screening algorithm making candidate slates more homogeneous than the applicant pool, and by how much?

Formula

Plain language: Compute a diversity index (such as Simpson's or Shannon's) for the applicant pool before AI screening and for the candidate slate after AI screening. HAI is the proportional reduction: how much diversity the screening step removed.

Notation:

HAI = 1 − (Diversity_post_screen / Diversity_pre_screen)

Variables:

Variable Definition Unit
Diversity_pre_screen Diversity index (Simpson's D or Shannon's H) computed across relevant demographic and experiential dimensions for all applicants before AI screening Index
Diversity_post_screen Same diversity index computed for the candidate slate that passed AI screening Index

Proposed formula, pending validation pilots.

At a Glance

Unit
Index (0 to 1)
Trend Direction
Lower; approaching 0
Reporting Frequency
Quarterly
Category
AI Screening & Pipeline Integrity
← Learning Recommendation Diversity Index (LRDI)Screening False Negative Estimation Rate (SFNER) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment