MPs hold first ever debate on live facial recognition

Accuracy issues

On the issue of accuracy and bias, shadow home secretary Chris Philp – who when in government pushed for much greater police use of the technology and called for LFR watchlists to be linked to the UK’s passport database – highlighted a recent study by the National Physical Laboratory (NPL) that found “no statistically significant” racial bias with police LFR systems when used in certain settings.

“When this technology was first introduced about seven years ago, there were reports – accurate reports – that there was racial bias in the way the algorithm at the time operated,” said Philp. “The algorithm has been developed a great deal since those days, and it has been tested definitively by the National Physical Laboratory, which is the nation’s premier testing laboratory.”

He noted that the NPL specifically found that when deploying the Neoface V4 facial recognition software provided by Japanese biometrics firm NEC, both the Met and South Wales Police could achieve “equitable” outcomes across gender and ethnicity by setting the “face-match threshold” to 0.6 (with zero being the lowest similarity and one indicating the highest similarity).

However, other MPs contested this, noting that while the accuracy may well have been improved by police using LFR at that particular threshold, there are no rules in place to stop them lowering that threshold at any time.

“There’s no such thing as no misrepresentations, or people who are not wrongly identified, and it’s also very easy for a police service to lower that number because we have no judicial oversight of it,” said Labour MP Dawn Butler, who later added that even set at 0.6, the LFR software was still less accurate than so-called police “super-spotters”, which are specialist officers trained to identify people quickly in crowds.

“There could be a case where a police service is trying to prove this system they bought is value for money, and you can imagine a police officer not getting many hits at 0.6, lowering it to 0.5 so they can get more hits, which, in turn, will mean more people are misidentified.

“There should be regulation around this issue. Taking away somebody’s liberty is one of the most serious things we can do in society, so we need to think very carefully if we’re going to introduce something that accelerates that.”

Racial bias and trust

Lambeth Labour MP Bell Ribeiro-Addy also argued that even when using this threshold there is room for error, especially if images in a police database or pulled from publicly available sources online are mislabelled.

People of colour are already disproportionately stopped and searched at higher rates, and the use of potentially flawed technology will only serve to increase the rate at which ethnic minorities are stopped, searched and possibly even incorrectly detained

“It’s almost inevitable that images will be mislabelled and innocent people will be subject to needless run-ins with the police… The Metropolitan Police’s own testing of its facial recognition algorithm identified disproportionately higher inaccuracy rates when attempting to identify people of colour and women,” she said, highlighting a 2023 study by civil liberties group Big Brother Watch which found that over 89% of all UK police LFR alerts since the technology was introduced have wrongly identified members of the public as people of interest.

“People of colour are already disproportionately stopped and searched at higher rates, and the use of potentially flawed technology will only serve to increase the rate at which ethnic minorities are stopped, searched and possibly even incorrectly detained.”

She further argued that increasing stops via LFR could further dampen trust in the police, particularly among ethnic minority communities, which are already over-policed and under-served, and that allowing police to collect and check other types of biometric information like DNA or fingerprints from people in the street would not be accepted: “Why should we look at this intrusive, automated biometric software any differently?”

While police and their LFR suppliers claim that people’s biometric data is deleted instantly if they do not match any images contained in the watchlist, independent MP Iqbal Mohammed highlighted how Google’s incognito browser was supposed to be “very private” until it was discovered the company was storing that data in breach of UK data laws. “Companies telling you things are immediately deleted is not always true,” he said.

Other MPs variously claimed that the further roll-out of LFR would “exacerbate existing inequalities and discrimination”, “cause further division and mistrust of the police” and “undermine several of our fundamental rights”, including rights to privacy, freedom of assembly and expression, and non-discrimination.

Judicial oversight and specific legislation

Conservative MP David Davis highlighted the need for judicial oversight of the technology and specific legislation laying out clear rules for its use, arguing that it should not be left to non-statutory guidelines or police discretion.

Multiple MPs also spoke about the future of LFR, including the potential for mission creep and the possibility of linking the UK’s six million CCTV cameras to facial recognition software, arguing there needs to be laws in place before the technology is used even more widely.

“The technology is prone to slippage. Way back when … we introduced automatic number plate recognition [ANPR] to monitor IRA terrorists coming from Liverpool to London. That was its exact purpose, but thereafter it got used for a dozen other things, without any legislative change or any approval by Parliament,” said Davis.

Mohammed also said the lack of specific legislation and judicial oversight created huge room for police misuse and overreach, and further highlighted how it could easily be leveraged to undermine civil liberties around the right to protest. “Facial recognition can deter individuals from participating in protests or public gatherings due to the fear of being monitored or identified,” he said.

There are many, many doubts, so we should probably be thinking about halting the use of this technology until we’ve cleared up those doubts

Even while noting that LFR accuracy had improved greatly since the Met’s initial deployment in 2016 and recognising the operational benefits it can bring police, Philp also emphasised the need for legislative control and judicial oversight of the technology.

“In Croydon, [LFR] has resulted in approximately 200 arrests of people who would not otherwise have been arrested, including for things like Class A drug supply, grievous bodily harm, fraud, domestic burglary… [and] a man who had been wanted for two rapes,” he said. “They’d still be walking free if not for this technology.”

Philp added while it is “not true to say there is a complete vacuum as far as rules and regulations are concerned … there is merit in clarifying at a national level where these guidelines sit”.

He said while he would not want to see the UK go as far as the European Union has with its Artificial Intelligence Act (AIA) – which has banned remote biometric identification in a range of circumstances – on the basis it would allow criminals to go free, he sees using some sort of “regulation-making power” as a more sensible approach than new primary legislation.

One MP – Liberal Democrat Bobby Dean – explicitly called for a complete halt on police deployments of the technology, at least until primary legislation is in place to control its use: “I think it’s clear from this room today, there are many, many doubts, so we should probably be thinking about halting the use of this technology until we’ve cleared up those doubts.”