Wednesday 29 November 2023

Live Facial Recognition for Law Enforcement: The European Union’s Regulatory Approach Should be Informed by UK Police’s Practice



Asress Adimi Gikay (PhD)

Senior Lecturer in AI, Disruptive Innovation, and Law at Brunel University London

Photo credit: Dirk Ingo Franke, via Wikimedia Commons


Divergent Approaches to Regulating Live Facial Recognition

In what is characterised as an 'Orwellian Nightmare', UK’s Minister of Crime, Rt Hon Chris Philp  MP recently suggested the opportunity for the UK police to search national passport database using facial recognition technology to tackle shoplifting crimes. This occurred as preparations were underway for the AI Safety Summit, an event that took place in London. It has some irony, to the surveillance anxious participants who gathered to discuss AI Safety as London is one of the cities with the largest number of CCTV cameras; but importantly where the police frequently use live facial recognition (LFR) in public spaces. 

Since South Wales Police made the first arrest using LFR over six years ago, UK Police use the technology to locate criminal suspects from crowds. In  LFR, the artificial intelligence(AI) software compares, in real-time, biometric facial images captured by a camera with existing facial templates of persons of interest  in a police-created database known as ‘watchlist’. This is contrasted with retrospective facial recognition system where the facial recognition takes place in the absence of the person of interest based on a video or still image taken from a source (also known as post-system). The UK government has called up on police force to expand the use of the technology, amidst growing concerns that the technology could endanger civil liberties.

By contrast, the European Union’s (EU) upcoming AI Act, in the Parliament’s Compromise Amendments categorically bans the use of LFR. However, the ‘trialogue negotiations’ seem to have led to a compromise where the use of the technology is permitted for specifically listed crimes punishable by at least five years. The EU’s restrictive position seems  intended primarily to appease civil society organisations, 12 of which wrote a letter to the EU Council in 2022 reiterating the need to prohibit the technology in categorical terms. Meanwhile, despite the recent call from 61 MPs and 31 civil society organisations demanding the immediate cessation of the use of the technology by UK police and private companies, efforts to stop the technology have remained unsuccessful in the UK.

In this post, explain why the UK’s approach should inform the regulation of LFR in the EU, by using evidences from the use of the technology by the UK police. I also introduce the theory of incrementalism, a normative framework for regulating novel technologies posing evolving risks whose magnitudes are yet to be known, in my forthcoming Cambridge Law Journal Article—‘Regulating Use by Law Enforcement Authorities of Live Facial Recognition Technology in Public Spaces: An Incremental Approach.’ 

Why the EU’S Approach Should Take an Incremental Approach

Incrementalism calls for regulating the use of the LFR technology by the police, and by extension similar technologies with novel advantages and risks, through progressive adjustment of the existing legal framework in the light of the potential risks and evidence of actual harm. This is different from a regulatory framework that responds to the risk of harm assessed in abstract terms, without considering the context of actual application of the technology, existing safeguards as well as the overall benefit of the technology.

I propose this theory to incorporate four main ingredients: sectoralism; reliance on existing legal frameworks; evidence-based regulation; and flexibility. This post will only explain evidence-based regulation as one of the important elements of the theory. The UK’s prevailing approach to regulating LFR and AI in general reflects certain elements of incrementalism. A measured regulation of facial recognition technology in the EU requires adopting this theory in its entirety or partially.   

Evidence-Based Regulation

The position of the EU on LFR does not appear to be based on thorough assessment of the benefits and risks of the harm of the technology as well as public support and the ability of law enforcement authorities to use it in proportionate manner.  Indeed, these are important factors in choosing the appropriate response to a new regulatory phenomenon. The experience in the UK provides an excellent insight into understanding the issue.

Evidence of Public Support and Benefits of the Technology

In a 2019 UK national survey conducted by Ada Lovelace Institute, 70% of respondents thought police should be permitted to use facial recognition in criminal investigation, with 71% supporting its use on public spaces, if it helps reduce crime. This positive public view aligns with the existing evidence of the effectiveness of the technology in tackling crimes. In 2020 and 2022, the London Metropolitan Police Service identified nine suspects in eight live facial recognition deployments. Earlier in 2018, the technology assisted South Wales Police in, reportedly,  making 450 arrests.  Several deployments in the UK have recently shown the effectiveness of the technology in helping arresting people suspected of committing violent crimes.

In October 2023, the Metropolitan Police identified 149 suspects of retail crimes using retrospective facial recognition. They compared hundreds of CCTV still images provided by retail businesses of their ‘prolific retail offenders’ against custody images. The result is significant as retail business are crucial to the UK economy creating a job for one in ten Londoners. Additionally, these crimes lead to the loss of estimated £1.9 billion in revenue whilst involving rampant abuse of retail workers. 

With the technology garnering some public support, campaigners struggle to present a persuasive evidence of harm of using it to back their push for blanket prohibition or suspension in the UK. As EU members states have not allowed the technology to be used, it is impossible to understand if the technology actually causes harm.  Generally, advocacy groups  highlight the inaccuracy of face recognition systems, especially in identifying women of colour, with Big Brother Watch claiming that Met and South Wales Police facial recognition systems are over 89% inaccurate. However the National Physical Laboratory independently tested two facial recognition systems used by the UK police in 2022. The result showed the software underperformed the most on Black-Female faces, but the discrepancy in accuracy rates across demographics was found to be statistically insignificant. Equally importantly, the inaccuracy of the technology does not inevitably translate into harm due to the existing legal safeguards the UK police adhere to, safeguards that exist or can be implemented in the EU.

Evidence of Safe and Proportionate Use

Despite the technology seeming to be unquestionably inaccurate, there is no reported case of serious harm resulting from the use of the technology in the UK because the UK police use it safely and proportionately. This can be contrasted with the US, where troubling incidents of wrongful arrests using facial recognition systems have been documented. For instance, Nijeer Parks, a Black American misidentified by facial recognition was wrongfully incarcerated for ten days. According to a civil complaint against the Director of Woodridge Police and others, Nijeer Parks voluntarily visited the police station to clear his name upon learning of an arrest warrant issued for him, in what he believed to be a case of mistaken identity. The police subjected him to coercive interrogations and solitary confinement, to secure his confession, whilst ignoring his alibi and the mismatch between DNA and fingerprints found at the crime scene with those of Nijeer Parks’.

These kinds of incidents seem to reflect more of police misconduct than the inherent challenge posed by the technology. Despite the concerns in the US,  in 2021, not less than seventeen state legislatures rejected bills to ban facial recognition. Some states including New Orleans and Virginia, that had previously banned facial recognition have now reversed course, to allow its regulated use.  Legislatures seem to want legal frameworks that strike a balance between benefits of the technology and addressing its risks.

Such a legal framework largely exists in the UK and the EU, as the two jurisdictions have similar human rights regimes. The UK police are obligated to use facial recognition technology in compliance with the Human Rights Act and the Equality Act, the latter imposing equality impact assessment obligation. Equality impact assessment requires the police to proactively tackle the potential discriminatory impact of the technology on specific groups and implement risk mitigation measures. These measures are supplemented by privacy law under the European Human Rights Convention and data protection rules under the Law Enforcement Directive (applicable both in the EU and the UK).

In the UK, the police also adhere to a national document prescribing detailed procedures for deploying live facial recognition technology known as the authorised professional practice. This national code of practice has a binding force. As a result of these comprehensive safety frameworks, the UK police have used facial recognition technology for seven-years without a single instance of wrongful arrest or abuse. As the European Union has advanced legal frameworks on human rights, privacy, data protection, and rule of law in general, it is inconceivable that the result of using facial recognition technology in the EU would be different from that of the UK.

Advocacy groups also highlight privacy intrusion and the expansion of surveillance as further concerns in relation to the use of LFR in public spaces. Nevertheless, these are addressed by legally limiting the duration, purpose and context of the use of the technology in the UK. The police are required to  assess the proportionality of using the technology, especially in places where it could have serious privacy implications such as hospitals and schools. This is mainly because article 8 of the European Human Rights Convention, similarly to its EU counterparts, articles 7, and 52 of the Charter of Fundamental Rights allow interference with privacy right based on the assessment of proportionality and necessity.  The existing legal framework in the UK does not permit surveillance at will. Things are not different in the EU, and in any event, a  loophole that creates a room for excessive surveillance could be addressed by a legislation.

Additionally, the face recognition software used by the police does not retain biometric data of individuals unless  positively matched, and personal data generated during its use is automatically deleted within a short span of time. These are intentionally built-in features of the software aimed at lessening the privacy and data protection impacts of the technology. Last, LFR is currently used in public spaces where people are unlikely to engage in private activities that should outweigh the public's interest in tackling violent crimes, although again, specific deployments need to observe the requirements of necessity and proportionality. 

Academics and advocacy groups often express doubt about the clarity of the legal basis for using facial recognition or the existence of mechanisms for redress for harms caused by the use of the technology. But this is not based on sound legal analysis. The current law does allow the police to gather information for fighting crimes including using new technological tools.  Furthermore, the police can be liable to pay compensation for harms, if they wrongfully detain or interrogate someone following misidentification using facial recognition technology under civil liability law. This is not to suggest that the existing law is without any loophole, but that addressing any legal gap requires delicate balancing rather than unnecessarily restrictive measures.

The Need to Reverse Course in the EU

The EU Commission’s Initial Draft of the AI Act permitted the limited use of LFR for law enforcement purpose.  First, by way of exception it allowed the use of the technology for narrowly defined, specific, and legitimate purposes [Art. 5(1)(d)]. These purposes are:

(i) the targeted searches for specific potential victims of crime, including missing children;

(ii) the prevention of a specific, substantial and imminent threat to the life or physical safety of natural persons or a terrorist attack; and

(iii) the detection, localisation, identification or prosecution of a perpetrator or suspect of a crime with a maximum sentence of at least three years that would allow for issuing a European Arrest Warrant. 

Second, the relevant law enforcement authority must demonstrate that the use of the technology is justifiable against (a) the seriousness, probability and scale of the harm caused in the absence of the use of the technology;(b) the seriousness, probability and scale of consequences of the use of the technology for the rights and freedoms of all persons concerned; and (c) the compliance of the technology's use with necessary and proportionate safeguards and conditions in relation to the temporal, geographic and personal limitations[Art. 5(2)-3]. The authority proposing to use the technology bears the burden of justification.

Third, the relevant law enforcement authority must obtain prior express authorisation from a judicial or a recognised independent administrative body of the Member State in which the technology is to be used, issued upon a reasoned request. If duly justified by urgency, the police may apply for authorisation during or after use [Art. 5(3)].

The Parliament’s Compromise Amendments categorically banned LFR used either by private companies or law enforcement authorities.  As this post has demonstrated, the evidence in the UK as well the recent reversal of bans in the US clearly indicate that the EU’s position is not based on concrete evidence. As mentioned earlier, the ‘trialogue negotiations’ seem to have led to a compromise where the use of the technology is permitted for specific crimes punishable by at least five years. This is unnecessarily restrictive as host of crimes including money laundering, financial fraud and other offences are not envisioned to be among the crimes for which the technology can be used.

A Call for Measured Regulation

The thinking behind the EU’s approach seems to be highly influenced by campaigners who depict the use of the technology as 'Orwellian' to induce public  fear, regardless of the context in which it is used. It appears to be a knee-jerk reaction, rather than an evidence-based response. The UK’s current practice and legal framework certainly have some loopholes to close. For instance, the technology could potentially be used for all crimes today, regardless of the seriousness of the crimes in question. But addressing these kinds of details does not entail suspending the use of the technology or entirely prohibiting it. Neither does it requires unnecessarily restricting it. Any legislative effort that aims to strike a delicate balance between the societal benefits and risks of the technology should take an incremental approach, that allows for timely response to evolving risk based on actual evidence of harm than conjecture. Starting with the strictest regulatory framework, uninformed by evidence could needlessly deny society of the benefits of the technology.


No comments:

Post a Comment