Fair to everyone
A New Bias
Independent audits of how AI models treat people differently, all else equal.
Visit anewbias.comThe problem
AI now helps decide who gets hired, housed, treated, and lent money. Change one thing about a person, like race, age, or religion, keep every other fact the same, and the answer can change. Few people measure it in a way anyone can repeat.
What I built
- A counterfactual test runner: the same case sent to many AI models, with one of twelve demographic traits changed at a time.
- Judging and scoring with fixed thresholds that separate a real gap from noise.
- Published, reproducible reports across six areas, from criminal justice and healthcare to lending and education.
- A dashboard for comparing models side by side.
Built with
- Python
- SQLite
- LLM APIs
- Next.js
