Investigation finds US predictive policing tool PredPol, used countrywide, often perpetuates biases, directing police to poor, Black, and Latino neighborhoods
PredPol grew from a LAPD/UCLA research project into a tool used across more than 60 jurisdictions. The new investigation shifts attention from the promise of data-led deployment to the geographic consequences of its recommendations.
PredPol's deployment recommendations and the patrol decisions made from them face sharper scrutiny from the jurisdictions that use the tool, particularly where they concentrate police activity in poor, Black, and Latino neighborhoods.
For PredPol, the investigation directly weakens the claim that historical crime data can guide neutral police-resource allocation.
Second-order effects
Police departments considering replacement predictive-policing programs inherit the same credibility problem: changing vendors does not address bias where training data and deployment practices reproduce it.
LAPD's new initiative is likely to receive more scrutiny because related reporting connected it to discontinued PredPol and Operation Laser surveillance programs.
Third-order effects
If investigations continue to find unequal deployment outcomes, public-safety AI governance will increasingly evaluate systems by where police are sent, rather than treating technical predictive accuracy as sufficient.
The pattern puts pressure on the predictive-policing market to justify whether data-driven allocation can be separated from historically uneven enforcement.
The trend: Predictive policing is moving from an efficiency-focused technology story toward outcome-based scrutiny of bias, accountability, and public-safety AI governance.
This predictive policing software (PredPol) has been in use for a decade, and The Markup/Gizmodo bombshell investigation found that the it was disproportionately predicting crimes in Black and Latino neighborhoods. https://themarkup.org/... https://twitter.com/... https://twitte…
We observed a pattern: In the majority of jurisdictions, more predictions occurred in the block groups that contained a higher proportion of low-income residents and Black and Latino residents. https://twitter.com/...
We ranked the block groups based on how many predictions each received and then looked at the race and household income of the people who lived there. You can read more about the disparate impact analysis in our methodology: https://themarkup.org/...
Take Jacksonville, TX. PredPol predicted that “assault” would occur an average of 5 times a day at the Sweet Union Apt public housing community. The red squares indicate it was one of the most targeted locations in the city. You can look at this map here: https://markup-public-da…
Another example was the Buena Vista low-income housing complex in Elgin, Ill. Also the red square. Six times as many Black people live in the neighborhood where Buena Vista is located compared to the city average. More about Elgin here https://markup-public-data.s3.us-east- 1.ama…
We used 5.9 million predictions from 38 police departments and analyzed the distribution of these predictions for each jurisdiction at the geographic level of a census block group.
New story from @themarkup and @gizmodo. This is a first-of-its-kind analysis of the data generated by PredPol. Let's dive into our findings https://themarkup.org/...
When we asked PredPol CEO Brian MacDonald whether he was concerned about the race disparities we found, he didn't address the question directly, but said the software mirrored reported crime rates. /8
We limited our analysis to U.S. city and county law enforcement agencies for which we had at least six months' worth of data. This left us with 38 jurisdictions. We then matched the predictions to Census data to determine the demographics of the targeted areas. /5
NEW: @Gizmodo/@themarkup analyzed millions of crime predictions a predictive policing firm left exposed on its servers. We found the algorithm spared White neighborhoods and relentlessly targeted Black & Latino ones - with devastating consequences for some https://themarkup.org/.…
Police across the U.S. use software from a company called PredPol (recently renamed @Geolitica_PS ) that says it predicts future crime without racial bias. But we found it rarely predicted crime in White areas & disproportionately predicted it in Black & Latino areas. /2 https://…
Critics have long suspected that predictive policing software was racially biased. Today, we have the answer: @themarkup & @gizmodo analyzed 5.9 million algorithmic crime predictions. We found they disproportionately target Black & Latino areas. /1 https://themarkup.org/...
In other words, the whiter the neighborhood, the fewer crime predictions. The same trend proved true for income: the wealthier the neighborhood, the fewer predictions. /3 https://twitter.com/...