Researchers warn AI policing could deepen discrimination

Researchers warn that the growing use of AI in policing risks increasing discrimination against minority groups.[1] Studies suggest that crime-hotspot predictions may rely on biased data, which can magnify existing prejudice and create a feedback loop.[1] AI systems are trained on data that already contains human bias, and that bias can lead to self-fulfilling prophecies, according to the article.[1] The article says the feedback problem predates AI, but AI may be accelerating it.[1] That distinction matters: the concern is not that AI created biased policing, but that it may intensify an existing dynamic.[1]

In the example described, more crime is recorded in an area because more officers are there to see it.[1] Meanwhile, areas outside the flagged zones can appear to have less crime because fewer people are looking there.[1] The article says crime figures partly record where police have chosen to look rather than offering a full list of crimes.[1] If those figures are then used to predict where officers should go, the resulting data may reinforce the original pattern.[1]

Experts are calling for AI use in policing to be reviewed with greater human oversight.[1] Dr Carolyn McKay says that oversight should include checking police data for bias and removing it.[1] The article also argues that the public functions of state policing ought to be subject to human oversight and scrutiny.[1] For teams assessing such tools, these recommendations point to scrutiny of both the data and the decisions made with it.[1]

The warning appears in an essay that won the 2026 Young Science Writer Award, chosen each year by the Association of British Science Writers.[1] The award-winning essay was written by Adrian Derek Dredge, who was 16 and a pupil at Great Western Academy in Swindon.[2] Dredge celebrated the award at a ceremony at London's Science Museum on September 16.[2] The essay describes biased policing data as data collected from police departments with a history of racism or bias.[2] It warns that an AI system may treat biased data as a pattern to reproduce, and that AI-generated output used as new input can create a dangerous feedback loop.[2]