See our Open Letter to Local authorities re use of algorithms for child protection (blog post 7 March). This article shows other dangerous uses of algorithms to target Black communities.
The grim reality of life under Gangs Matrix, London’s controversial predictive policing tool by Peter Yung for Wired. Original article accessed HERE.
2 Apr 2019
AI and machine learning software was meant to make policing fairer and more accountable – but it hasn’t worked out that way
The first time Bill was stopped and searched by police, he was standing outside a friend’s house in south London. “The police pulled up on us, three cars,” he says. “I asked them why they were searching us, and one said, ‘Because I want to’.” Bill was 11. That was nearly a decade ago.
Pupils from different neighbourhoods around the area, such as Brixton, Peckham, and Tulse Hill went to Bill’s secondary school. Some areas were particularly rough, and gangs with minors in them were common. Four boys died in a three-year period back then, he says. One was shot while lying in bed. Another was stabbed in the heart. “One of them was my close bredrin,” says Bill. “I didn’t really think too much of it. I just had to keep it moving, you know?”
From then on, officers took a keen interest in Bill, as they did with many other young black boys from underprivileged backgrounds in the area. When he was 14, police arrested him twice in two weeks – without charge. “That’s that network. That little matrix that they have,” he adds. “They know that people who I know have been arrested for drugs, so they assume that I’m going to have drugs on me.”
Bill had ended up on what is known as the Gangs Matrix, a controversial database created by the Metropolitan Police (the Met) in the aftermath of London’s 2011 riots to purportedly identify and surveil not only those at risk of committing gang-related violence, but also potential victims of it.
Based on a number of variables such as previous offences, patrol logs, social media activity and friendship networks, the matrix relies on a mathematical formula to calculate a “risk score” – red, amber, or green – for each person, in reference to the likelihood they will be involved in gang violence. This intelligence in theory guides an efficient use of police resources and aids court prosecutions.
But critics argue it is one of the most flawed policing initiatives in modern times. In a report last year, human rights charity Amnesty International described it as “a racially biased database criminalising a generation of young black men”, revealing that 35 per cent of those on the matrix had no police intelligence linking them to gang violence and had never been charged with a crime. Sharing certain YouTube videos of grime or drill music, meanwhile, is considered a key indicator of gang affiliation.
The implications of being on the matrix can be chilling, but finding out why you are on it, let alone how to be removed, is extremely difficult. One family received a letter warning they would be evicted from their home if their son didn’t stop his involvement with gangs – but he had been dead for more than a year. A disabled mother’s council-provided car was seized after her son – who acted as her carer and was registered to drive the car – was arrested without charge or further action. In Bill’s case, he was forced out of his mother’s house and put into a residential care home due to being on the matrix. He was later banned from attending the South London Learning Centre.
In November, the Information Commissioner’s Office (ICO), Britain’s watchdog for data use, ruled that the matrix breached data protection rules. The investigation found that it did not clearly distinguish between victims and perpetrators of crime, some boroughs were keeping informal lists of those who were supposed to have been removed from the matrix, and there was “blanket sharing [of data] with third parties” including schools, job centres, and housing associations.
A separate, damning review published the following month by the Mayor of London’s Office, which oversees the Met, found that although there is a need to address violence in the capital, the number of young black people on the matrix was “disproportionate to their likelihood of criminality and victimisation.” The review ordered the force to radically reform the tool within a year. The Met said in a statement at the time that it “does not believe that the Gangs Matrix directly discriminates against any community”.
But according to data obtained by WIRED via a freedom of information request, children as young as 13 are currently listed on the Met’s Gangs Matrix. In total, the list contains some 3,000 people, mostly young, black boys, including 55 children under 16. More than 7,000 individuals have been on the matrix at some point. About 80 per cent of those on the list are described as “African-Caribbean”, 12 per cent are from other ethnic minority backgrounds, and just eight per cent are “white European”. Yet the vast majority are considered to pose little threat of violence by even the police, with 65 per cent currently rated with a green risk score, 30 per cent amber, and five per cent red.
As a result, campaigners are now calling for the tool, the remit of which covers more than eight million people, to be scrapped. “I think it’s deeply problematic,” says Tanya O’Carroll, director of Amnesty’s global technology and human rights programme. “It’s a rudimentary use of data in poorly thought-out ways that ends up being extremely discriminatory to young black boys. The way the matrix works, intelligence about people is essentially hearsay – feeling, not fact.”
The matrix is part of a growing trend of police forces across the UK using open-source intelligence, big data and machine learning as part of its crime-stopping. A report published by Liberty in February revealed that at least 14 forces across the UK – around a third – are already using what has been coined “predictive policing”.
It outlined two main strands: “predictive mapping”, which identifies areas where crime will likely occur, and “individual risk assessment”, which predicts how likely an individual is to commit crime. However, the first has led to over-policing of certain communities and the second facilitates racial profiling, critics argue, while the broader issue of predictive policing is legally ambiguous, lacking accountability and not proven to be effective.
The Gangs Matrix does not exactly implement artificial intelligence or machine learning, unlike the tools of many others forces, says Hannah Couchman, a policy and campaigns officer at Liberty who authored the report. But there is still the concept of “pre-criminality” – being investigated by police without reasonable grounds, she adds. It comes into conflict with age concepts such as innocence until proven guilty and probable cause.
“We are really seeing a pattern of police forces rolling out these technologies without sufficient protection,” says Couchman.
Kent Police, the first UK police force to try to use computer algorithms to predict crime, ended its five-year deal with the US company PredPol last March, citing difficulties in proving that the technology could reduce crime.
South Wales Police and the Met are testing Automated Facial Recognition (AFR) – despite high rates of incorrect identifications, particularly for women and black people. Avon and Somerset police have started using a broad mapping program to assess the likelihood of things such as being a victim of stalking and taking stress-related sick leave.
Despite the significant problems, however, the UK’s national coordination body for law enforcement rejects any criticism. “For many years police forces have looked to be innovative in their use of technology to protect the public and prevent harm and we continue to develop new approaches to achieve these aims,” says Jon Drake, intelligence lead for the National Police Chiefs’ Council.
But while police use of big data may be inevitable, a significant concern is the lack of accountability in these systems, whose processes remain a mystery to even the officers tasked with deploying them – and the experts that have built them. Automation bias, the hesitancy to overrule computers’ automated decisions, is also a significant problem, says Nick Jennings, a professor at Imperial College London.
“We generate so much data from so many devices to make decisions, we can’t possibly do it ourselves and we need computer support,” he adds. “But even for those that build AI programs, it’s not easy to understand the logic behind the most powerful machines. So we need to be really careful about the kind of information that we feed in, and what biases it may have.”
In January, the British government acknowledged those challenges by launching the Centre for Data Ethics and Innovation, the first body of its kind in the world, which will provide advice on “data-enabled technologies”. The Nuffield Foundation’s new £5 million Ada Lovelace Institute will play a similar role, considering the ethical questions raised by big data, algorithms and AI. West Midlands police is also believed to be introducing an ethics committee as part of its predictive analytics.
However, a lack of transparency continues to plague the field. It has now emerged that a secret new system called the Concern Hub, headed by a central team at Scotland Yard that will liaise the Met and hubs in each of Greater London’s 32 boroughs, has already been undergoing an unpublicised trial in the capital.
A spokesperson for the Met says that the Concern Hub is “a new multi-agency diversion initiative” set to launch in south-east London in April, with a wider rollout across the city in the coming months. The aim is “to safeguard young people at significant risk of becoming involved in violence, drugs, or gang activity.”
These secretive approaches to policing could compromise trust, says David Lammy, the Labour MP for Tottenham in north London. “It is very worrying that another similar tool is being created without sufficient consultation with communities.” Jude Lanchin, an associate in criminal law for Bindmans who has worked on a number of cases involving the Matrix, calls the new methods just as opaque as ever.
Lawyers with significant experience of working on cases where defendants were described as “gang nominals”, a term used by the Gangs Matrix to denote association with – but not necessary membership of – gangs, also lambast the lack of transparency by police prosecutors in the courts.
“The phrase ‘gang nominal’ is cited as evidence without an explanation as to what that intelligence is based on,” says Suzanne O’Connell, a youth courts specialist for Tuckers Solicitors who first worked on a case involving the matrix in 2012. She estimates that 80 per cent of her court cases involve a reference to the term. “This whole cloud of secrecy is very difficult to challenge,” she adds.
The troubling rise of murder in London is, however, an issue that can’t be ignored. There were 135 people murdered or unlawfully killed in London last year alone, the highest in a decade. More than 40 per cent were men under 30. Reports of stabbings in the capital come and go almost daily.
But the focus on gangs has always been flawed. The Conservative government initially blamed gangs for the 2011 riots, but later reviews found the causes behind the deadly period of looting, arson and civil unrest to be far broader. A study by the Mayor’s Office for Policing and Crime also estimated that more than 80 per cent of all knife-crime incidents resulting in injury to a victim under 25 were deemed to be non-gang related.
Put simply, the causes behind violence in London are deep-seated may be going far beyond criminal groups and the limits of technology. “It reflects the bias in society,” thinks Katrina Ffrench, chief executive of the campaign group Stopwatch. “I don’t expect it to not be racist, why would you?”