Supply Chain LabAI Supplier Bias AuditLab Home
Responsible AI · Procurement

AI Supplier Bias Audit

Train a supplier-ranking model, then audit it. Proxy features lift model value but smuggle in bias; mitigation narrows the fairness gap at a value cost, and an audit makes the decision defensible. Balance value, fairness and compliance.

Round 0 / 20
Value vs fairness gap

Decision-Logic Coach

Round-by-round

Scoreboard & export

Par is the default-policy score. Tune the controls to beat it, export your run, or save a leaderboard best.
You (cum.)
0
Par (default)
0
Delta
0

Debrief

About this game

AI Supplier Bias Audit - value that survives scrutiny

The concept

A supplier-ranking model can be more accurate yet unfair. Proxy features correlate with protected attributes, boosting apparent value while creating disparate impact against groups of suppliers.

How it works

Each round you choose whether to use proxy features, how much bias mitigation to apply, and whether to audit. The fairness gap is measured against the four-fifths rule (≤ 20%); mitigation trades a little value for fairness, and auditing raises compliance.

Winning strategy

Keep the fairness gap within the four-fifths rule - drop proxies or raise mitigation until it clears - and keep audit mode on. Defensible value beats raw value that fails review.

disparate impact ratio ≥ 0.8 (four-fifths rule) ⇔ fairness gap ≤ 20%
Responsible AIProxy featuresDisparate impactFour-fifths ruleBias mitigationAuditability