Police using AI risks making discrimination worse, researchers warn
Researchers warn that AI tools used to predict crime hotspots can amplify discrimination if they rely on police records shaped by existing bias. Greater scrutiny and human oversight could help prevent automated predictions from reinforcing patterns that affect minority communities.
The concern is that areas receiving more police patrols may produce more recorded crime, then appear even riskier in later predictions. Sociologist Sarah Brayne’s study of the Los Angeles Police Department found that its data was skewed towards crimes visible in public and policing focused disproportionately on black neighbourhoods. AI researcher Dimitrios Kollias says newer systems can account for historical bias, but adds that increasingly autonomous tools still need meaningful human control.
In short
- Biased police data can skew AI crime predictions.
- More patrols may mean more recorded crime in targeted areas.
- Researchers call for scrutiny and human oversight.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
AI crime predictions can help police allocate limited patrols more efficiently and identify patterns that might otherwise be missed. Supporters argue that systems designed to account for historical bias, combined with careful scrutiny and meaningful human oversight, can improve decisions rather than simply reproduce past practice.
The case against
Police records reflect where officers have patrolled and which offences they have prioritised, so predictions built on those records can send more patrols to already heavily policed communities. Critics argue this feedback loop risks entrenching discrimination, while the focus on recorded crime may obscure harms that receive less police attention; human oversight may not be enough if the underlying data and incentives remain biased.
AI Research Science Technology