UK police facial recognition system shows racial bias in testing
Testing of the UK police national database's retrospective facial recognition system found it produces substantially higher false positive identification rates for black and Asian people than white people, with particularly severe disparities for black women. The Home Office admitted the bias issue, and the Information Commissioner's Office opened an inquiry to consider enforcement action.
What happened
Testing by the National Physical Laboratory of the police national database's retrospective facial recognition technology found differential false positive identification rates across demographic groups. Analysis at a lower setting found the false positive identification rate for white subjects was 0.04%, compared to 4.0% for Asian subjects and 5.5% for black subjects.
The Home Office admitted the technology was 'more likely to incorrectly include some demographic groups in its search results.' Testing also revealed particularly high false positive rates for black women at 9.9% compared to 0.4% for black male subjects.
Reported impact
The Association of Police and Crime Commissioners stated the findings showed 'an inbuilt bias' and that 'technology has been deployed into operational policing without adequate safeguards in place.' They expressed concern that the technology could be expanded nationally to shopping centres, stadiums and transport hubs without safeguards.
The Information Commissioner's Office stated that 'any perception of bias and discrimination can exacerbate mistrust' in the technology and indicated potential enforcement actions including issuing legally binding orders or fines.
Index analysis
The disparate false positive rates across racial groups indicate the facial recognition algorithm exhibits systematic bias against black and Asian people, with black women experiencing rates 247 times higher than white subjects at the tested setting.
Governance controls
The Information Commissioner's Office asked the Home Office for 'urgent clarity on this matter' to assess the situation and consider next steps, which could include enforcement action. The police and crime commissioners questioned why the findings had not been released at an earlier opportunity or shared with affected communities.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.