Police adoption of artificial intelligence tools risks entrenching and amplifying discriminatory practices, researchers have warned. They say many of these systems are trained on historical policing records that already reflect bias, and that using those tools without careful controls could replicate and worsen those patterns.
How the risk emerges
AI systems learn patterns from the data they are given. When training data contains disproportionate stops, arrests or other interactions involving minority groups, the model can internalise those patterns as predictive signals. Researchers caution that this means the tools may recommend actions that mirror past bias rather than provide neutral, objective assessments.
Potential consequences and concerns
Without robust oversight, the use of biased models by police could lead to decisions that further disadvantage affected communities. The researchers emphasise that such outcomes would not simply reflect technical error but would be rooted in social and historical factors embedded in records used to build the systems.
Calls for oversight and transparency
The experts urge authorities to put in place governance measures around procurement, testing and deployment of AI tools in policing. They say independent audits, transparent reporting of how systems are trained and used, and safeguards to prevent biased outcomes are necessary to reduce the risk of harm.
Broader debate
Their warning adds to a wider debate over the role of automated tools in public services, where supporters argue AI can improve efficiency while critics warn it can codify inequality. The researchers' findings underline the need for policymakers and law enforcement agencies to weigh the risks carefully and to ensure that technology does not substitute for accountability or equitable decision-making.