The UK’s data protection regulator has pressed the Home Office for “urgent clarity” over police use of facial recognition technologies that have been found to exhibit evidence of racial bias.

Recent testing by the National Physical Laboratory (NPL), commissioned by the Home Office, found that facial recognition systems applied to the Police National Database misidentified racial minorities at higher rates than white people, with the ‘false positive identification rate’ (FPIR) much higher for Asian (4.0%) and black subjects (5.5%), than white (0.04%).  

Analysts from the NPL found that the algorithm used by the police for Retrospective Facial Recognition, which compares still images captured from the likes of CCTV against a reference image database, was more likely to incorrectly include these demographic groups in its results in a “limited set of circumstances”.

The report has sparked fierce backlash, coming at the same time as the government outlined plans to ramp up the use of FR tech and biometrics by law enforcement, touting its adoption by the Metropolitan Police, which claims to have made 1,300 arrests using facial recognition.

But while the government’s policing minister, Sarah Jones, hailed the technology as “the biggest breakthrough for catching criminals since DNA matching”, others have pointed to the harm which will inevitably come from inaccurate profiling and wrongful arrest.

The ICO was quick to make this point, with deputy commissioner Emily Keaney agreeing that while FR technology can play a valuable role, “public confidence in its use is paramount, and any perception of bias and discrimination can exacerbate mistrust.”

Expressing the watchdog’s disappointment that it had not already been made aware of the issue, Keaney said the ICO had “asked the Home Office for urgent clarity” as it considers next steps. 

Others expressed stronger condemnation. In a statement, the Association of Police and Crime Commissioners said that while there has so far been no evidence of an adverse impact in any individual case, that was “ more by luck than design”.

“The report from the National Physical Laboratory evaluating the accuracy and equitability of the Police National Database’s (PND) retrospective facial recognition technology tool sheds light on a concerning in-built bias,” said the APCC.

“The language is technical but, behind the detail, it seems clear that technology has been deployed into operational policing without adequate safeguards in place. We recognise the importance of embracing emerging technologies in order to combat crime, but governance and accountability on behalf of communities is not a “nice to have”. 

“These technologies are increasingly invasive and sophisticated. If they are to gain the trust and support of the public, particularly those groups historically mistrustful of the police, then full transparency is vital.”

Recommended reading

• UK Police Using Passports For ‘Secretive’ Facial Recognition Searches
• UK Expands Live Facial Recognition in Police Van Rollout
• Ada Lovelace Institute Condemns UK’s Facial Recognition Governance

The Home Office, for its part, acknowledged that the technology is “more likely” to incorrectly include some groups in its search results, but said that a new algorithm has already been procured, which shows no statistically significant bias, and is due to be operationally tested next year.

The Home Office also stressed that retrospective facial recognition is only one piece of intelligence used as part of wider investigations, and that manual safeguards in police training and operational practice mean all potential matches returned from the PND must be visually assessed, a safeguard in place before the NPL’s testing.

“While we want to remove bias in facial recognition systems entirely, policing are confident that established training, processes and operational practices protect the public from the identified bias,” said the government.