The United States Department of Justice has failed to convince a group of US lawmakers that state and local police agencies aren’t awarded federal grants to buy AI-based “policing” tools known to be inaccurate, if not prone to exacerbating biases long observed in US police forces.
Seven members of Congress wrote in a letter to the DOJ, first obtained by WIRED, that the information they pried loose from the agency had only served to inflame their concerns about the DOJ’s police grant program. Nothing in its responses so far, the lawmakers said, indicates the government has bothered to investigate whether departments awarded grants bought discriminatory policing software.
“We urge you to halt all Department of Justice grants for predictive policing systems until the DOJ can ensure that grant recipients will not use such systems in ways that have a discriminatory impact,” the letter reads. The Justice Department previously acknowledged that it had not kept track of whether police departments were using the funding, awarded under the Edward Byrne Memorial Justice Assistance Grant Program, to purchase so-called “predictive” policing tools.
Led by Senator Ron Wyden, a Democrat of Oregon, the lawmakers say the DOJ is required by law to “periodically review” whether grant recipients comply with Title VI of the nation’s Civil Rights Act. The DOJ is patently forbidden, they explain, from funding programs shown to discriminate on the basis of race, ethnicity, or national origin, whether that outcome is intentional or not.
Independent investigations in the press have found that popular “predictive” policing tools trained on historical crime data often replicate long-held biases, offering law enforcement, at best, a veneer of scientific legitimacy, while perpetuating the over-policing of predominantly Black and Latino neighborhoods. An October headline from The Markup states bluntly: “Predictive Policing Software Terrible At Predicting Crimes.” The story recounts how researchers at the publication recently examined 23,631 police crime predictions—and found them accurate roughly 1 percent of the time.
“Predictive policing systems rely on historical data distorted by falsified crime reports and disproportionate arrests of people of color,” Wyden and the other lawmakers wrote, predicting—as many researchers have—that the technology only serves to create “dangerous” feedback loops. “Biased predictions are used to justify disproportionate stops and arrests in minority neighborhoods,” further biasing statistics on where crimes occur.
Senators Jeffrey Merkley, Ed Markey, Alex Padilla, Peter Welch, and John Fetterman also cosigned the letter, as did Representative Yvette Clarke.
The lawmakers have requested that an upcoming presidential report on policing and artificial intelligence investigate the use of predictive policing tools in the US. “The report should assess the accuracy and precision of predictive policing models across protected classes, their interpretability, and their validity,” to include, they added, “any limits on assessing their risks posed by a lack of transparency from the companies developing them.”
Should the DOJ wish to continue funding the technology after this assessment, the lawmakers say, it should at least establish “evidence standards” to determine which predictive models are discriminatory—and then reject funding for all those that fail to live up to them.
