PredPol perpetuates racial, ethnic, income bias

An investigation by The Markup and Gizmodo discovered potential bias issues with crime prediction system PredPol.

Analysis of a huge volume of crime predictions left on an unsecured server showed PredPol (now renamed Geolitica) often avoided White and middle-to-upper-income residents neighbourhoods, and targeted Black and Latino neighbourhoods. 

The findings suggest PredPol technology is resulting in so-called 'feedback loops' in which lower-income, ethnic zones are treated as surveillance hotspots and lead to disproportionately higher numbers of arrests of minority populations.

Investigations, assessments, audits 🧐

• The MarkUp (2021). Crime Prediction Software Promised to Be Free of Biases. New Data Shows It Perpetuates Them

• The MarkUp (2021). How We Determined Crime Prediction Software Disproportionately Targeted Low-Income, Black, and Latino Neighborhoods

• The MarkUp (2021). PredPol investigation data

News, commentary, analysis 🗞️

• https://pluralistic.net/2021/12/02/empirical-facewash/

• https://iapp.org/news/a/analysis-shows-bias-with-ai-powered-crime-prediction-software/

• https://thenextweb.com/news/crime-prediction-software-free-of-biases-data-shows-perpetuates-them-syndication

• https://www.reddit.com/r/technology/comments/r78w77/crime_prediction_software_promised_to_be_free_of/

• https://www.thecanary.co/investigations/2021/12/03/foi-requests-reveal-which-police-forces-use-crime-prediction-software/