🔗 Share this article

Police forces across the UK effectively campaigned to deploy a facial recognition system acknowledged as biased against females, youths, and individuals from minority ethnic backgrounds, following complaints that a more accurate version generated fewer potential suspects.

The Technology in Practice

British police utilize the national police database to carry out retrospective facial recognition searches. This process involves comparing a reference photograph of a person of interest against a repository of over 19 million mugshots to identify possible hits.

Acknowledged Discrimination

The UK interior ministry admitted last week that the system was biased. This acknowledgment followed a study by the government's National Physical Laboratory determined it misidentified people of Black and Asian heritage and females at much greater frequency than white men. The ministry said it “had acted on the findings”.

“This raises the issue of whether facial recognition only becomes useful if users accept biases in ethnicity and gender. Operational ease is a poor argument for overriding basic freedoms.”

Long-Standing Problem

Internal documents show that this bias has been known about for more than a year. Furthermore, police forces argued to overturn an earlier ruling that was designed to address the problem.

Senior officers were informed of the system's bias in September 2024. The government-ordered laboratory study concluded the system was more likely to produce incorrect matches for photos of females, Black people, and those under 40 years old.

A Policy U-Turn

In reaction, the National Police Chiefs’ Council (NPCC) mandated that the confidence threshold required for potential matches be increased to a point where the bias was greatly diminished.

However, this directive was reversed the following month following complaints from police that the adjusted system was producing a lower number of “investigative leads”. NPCC documents show the higher threshold reduced the proportion of queries resulting in potential matches from 56% to a just under 15%.

Severe Disparities

Although the Home Office and NPCC declined to specify what threshold is currently used, the latest NPL study found the system could produce false positives for Black women almost 100 times more frequently than for Caucasian women at certain settings.

The Home Office commented on these findings: “Our evaluation found that in a specific scenarios the software is more likely to incorrectly include some population segments in its search results.”

Balancing Utility and Fairness

Outlining the effect of the brief increase to the system's confidence threshold, the police records note: “The change significantly reduces the impact of bias across protected characteristics of ethnicity, age and gender but had a significant negative impact on operational effectiveness”. The papers further note that police units argued that “a previously useful tool now delivered outcomes of limited benefit”.

Wider Implementation Proposals

Meanwhile, the government has opened a ten-week consultation on its proposals to expand the use of biometric scanning systems. The minister for police the relevant minister has labeled the tool as the “most significant advance since DNA matching”.

Expert and Oversight Concerns

The chair of a police oversight board, head of the advisory panel for the national policing equality strategy, commented: “We observed very little discussion in equality strategy sessions of the facial recognition rollout even with obvious cross-over with the strategy's goals.

“These revelations demonstrate once again that the anti-racism commitments policing has made via the race action plan are failing to be integrated into wider practice. Independent assessments have cautioned that new technologies are being rolled out in a context where ethnic inequalities, inadequate oversight and poor data collection already persist.

“All deployment of this technology must meet rigorous official guidelines, be subject to external review, and prove it reduces rather than compounds ethnic bias.”

Home Office Response

A Home Office spokesperson stated: “The Home Office takes the conclusions of the report seriously and we have implemented changes. A new algorithm has been externally evaluated and acquired, which has demonstrated no measurable discrimination. It will be tested early next year and will be subject to further assessment.

“Our priority is protecting the public. This gamechanging technology will assist police to apprehend and prosecute offenders. There is human involvement in every step of the procedure and no further action would be pursued without trained officers carefully reviewing the output.”