Hot take: a judge in Denver threw out a recidivism tool's score because nobody could say what the inputs actually were | Algorithmic Fairness Debate

Lee Thompson · 2026-10-20T01:40:04.48022+00:00

I sat in on a public hearing in Denver last month about a county using a risk score to help set bail, and the thing that got me was how little anyone could explain. The vendor's rep kept saying the model was "proprietary" when a public defender asked which factors actually moved a person's score up. That bugged me because I keep seeing people argue about whether these tools are biased on race or income, and they skip the step before that, which is whether the inputs are even the right stuff. In this case, one of the listed inputs was basically a person's address history, which to me is just measuring how often they moved, and moving a lot often means you're poor, not dangerous. A local news writeup later said the county paid $85,000 for the contract and got a one page summary of how the score works. So my take is that the fairness fight has to start with "can we see the inputs and the weights," because you can't check bias in a black box. Has anyone here dealt with a city or county that actually published the full factor list for one of these scoring tools, and if so, did it hold up to scrutiny?

Free To Speak

Loading…

Hot take: a judge in Denver threw out a recidivism tool's score because nobody could say what the inputs actually were | Algorithmic Fairness Debate - Free To Speak