Bias Feedback Loop

How to Use

1. Watch the cycle -- a golden glow travels clockwise around the four stages, illustrating how bias self-reinforces
2. Click any stage -- the detail panel at the bottom explains that stage's role in perpetuating bias
3. Click "Break the Cycle" -- expands to show four evidence-based mitigation strategies
4. Hover the arrows -- see a description of how each transition works

About

When organizations use machine learning to make decisions about people -- hiring, promotion, development opportunities, performance ratings -- there is a risk that historical biases get encoded into the model and then amplified through a self-reinforcing feedback loop. Biased training data produces biased predictions, which lead to biased decisions, which create biased outcomes that feed back into future training data.

Breaking this cycle requires deliberate intervention at one or more stages: fairness-aware algorithms, human review of model predictions, disparate impact testing across demographic groups, and regular bias audits that examine outcomes over time.