Mat Hayward - stock.adobe.com
“You are being watched,” proclaims the opening narration to the television series Person of Interest. “The government has a secret system; a machine that spies on you every hour of every day. I designed the machine to detect acts of terror, but it sees everything. Violent crimes involving ordinary people…”
A few years ago, it could have been argued that the premise of Person of Interest was pure fancy, but now it is edging closer to reality. The system being developed by West Midlands Police, called National Data Analytics Solution (NDAS), will use predictive analytics to identify potential perpetrators of certain types of crime.
This is not the first time that predictive analytics has been used by a UK police force. In 2017, Durham Constabulary began using the Harm Assessment Risk Tool (HART) algorithm to inform custodial decisions. “Durham police has rolled out an AI [artificial intelligence] system that predicts whether someone should be held in custody, by predicting someone’s risk of offending,” says Stylianos Kampakis, CEO of Tesseract Academy. “The system has met with some criticisms, as to whether it reinforces bias.”
Recent years have witnessed police budgets come under increasing pressure, but at the same time, demand for policing has increased.
Digital technologies present a series of opportunities that the Home Office is exploring. “We are working to tackle the devastating consequences through our Serious Violence Strategy,” said a Home Office spokesperson. “As part of our efforts, we are working with a number of police forces and partner agencies to explore innovative approaches to violent crime.”
As part of this initiative, West Midlands Police is developing an algorithm to harness its vast data stores. After considering hundreds of potential use cases for NDAS, the force chose three specific areas to target:
- Gun and knife crime
- Modern-day slavery
- Workforce wellbeing
Recent years have seen a rise in gun and knife crime. Much of this has been driven by gangs and organised crime groups and the resulting fallout.
Using internal police records, the NDAS system will review the data of people who have already been convicted of gun or knife crime, to build analytical models of their behaviour in order to understand their shared characteristics.
The NDAS uses data about crime recording, criminal convictions, custody, criminal intelligence and incidents attended by police.
These common denominators will be used to form key predictive indicators that the NDAS can use to search police records for offenders who share characteristics but have not yet committed a violent crime. “We are looking to predict those individuals who are deemed most likely to escalate their offending behaviour to become perpetrators of gun and knife crime,” says West Midlands Police superintendent Iain Donnelly, police lead for the NDAS project.
The police will use these findings to advise social services, or another non-police partner, to intervene. “This is not Minority Report,” says Donnelly. “This is about our understanding those individuals who represent the greatest threat to communities and intervene early to divert and support them and their families in the most appropriate way.”
The second use case for the NDAS system will be tackling modern-day slavery. This will use similar datasets as the violent crime use case, as the police may be called out to an incident where a dispute is taking place, which, under greater scrutiny, could reveal organised crime group activity, with people being held against their will.
As modern-day slavery is a cross-border threat, this will enable police to use the full force of their shared criminal intelligence apparatus to identify organised crime networks, as well as identifying key indicators of modern-day slavery.
The final use case will look at the police force itself. Because of the nature of their work, every police force in the country is affected by long-term stress-related sickness. This may be due to workload or exposure to traumatic incidents.
The NDAS system will review existing cases where officers have needed to take long-term leave because of stress, to assess what their shared key indicators are. Using this information, NDAS can identify officers who are potentially suffering high levels of stress, so they can be given the help at an early stage.
NDAS is currently a proof of concept. It is not intended as an ongoing operational capability, but as a test-case scenario to assess what can be achieved.
It was originally envisioned as an on-premise system, but this evolved to it being in the cloud, using Amazon Web Services (AWS), as this allows the system to be scaled easily. “If we need to lift and shift this in the future, it would be much easier to do it in a cloud environment than if we were to dismantle an expensive on-premise solution,” says Donnelly.
At the time of writing, the AWS cloud platform had been developed, pen-tested and deployed. Focusing initially on data from West Midlands Police, the initial data science and network analysis provided some insights that give confidence that the NDAS system will be able to identify individuals who are at risk of committing violent offences. Alongside this, early work using NDAS has successfully identified modern-day slavery networks in historic data.
One of the challenges in creating the NDAS system is that there has been no standard format for recording data among the UK’s police forces. Although not insurmountable, this has added an extra level of processing to adapt data extracts into a shared format.
After this comes the challenge of injecting data into the platform. Ideally, this would be done using a VPN connection, but many partnering police forces are reluctant to share data using such techniques. “They would rather give you the data in a physical device, but we are talking about terabytes of data,” says Donnelly. “That would not be an option we would consider for the end-state solution.”
Read more about technology and policing
- The Scottish Police Authority has approved Police Scotland’s 10-year digital transformation plan, which aims to bring the country’s police forces into the 21st century.
- How Avon and Somerset Constabulary developed a predictive analytics engine to tackle domestic violence and child abuse.
- Met Police trials analytics to fight gang crime.
Another challenge is to determine in the short term just how successful the NDAS system is, by proving something was prevented from happening. But over time, statistics will show whether the police are targeting the right people.
But as the NDAS system learns from historic offenders, there is a danger that it may fail to adapt to how crime changes in the future. “We need to learn from previous examples so we can make predictions in the future,” says Loubna Bouarfa, founder and CEO of Okra Technologies. “Yet there is a risk that the patterns of crime will change in the future, which predictive tools may fail to account for.”
Another significant challenge in the project has been acquiring the data from multiple police partners, because of legislative concerns about cloud storage for sensitive police data. Donnelly says: “In truth, the 43-force structure of policing is unhelpful. Every information security and data protection professional in each force has a different view of what is required by the legislation.”
To mitigate this, West Midlands Police is liaising with the Information Commissioner’s Office to ensure compliance with data protection legislation, with the goal of having a more comprehensive dataset before the initial NDAS project ends in March 2019.
One of the key points is to ensure that this is not a metaphorical black box – that results are not automatically acted upon without human consideration. Section 49 of the Data Protection Act 2018 outlines the restrictions on automatic decision-making, whereby a significant decision cannot be based solely on automatic computing.
“There is a difference between an algorithm advising a human who then acts on a prediction, and that action happening automatically without human intervention,” says Renzo Marchini, a privacy lawyer at Fieldfisher.
Historically, there has been significant mistrust when predictive analytics has been proposed in law enforcement. However, the NDAS system will not be scanning the general population. Instead, it will use existing records to identify individuals who are already known to the police.
“The people we are going to be interested in are those who have already set out on a criminal journey,” says Donnelly. “What we want to do is to try to identify that top tier of individuals who, if we do not do anything, we are highly confident that they will go on to harm others.”
To reassure the public, West Midlands Police approached the Alan Turing Institute Data Ethics Group for advisory scrutiny of the NDAS system. As well as providing a series of ethical issues that needed to be addressed, the group’s advisory report said: “We are impressed by the serious attempt to developing an ethical as well as legally compliant national analytics capability for law enforcement.”
Future of predictive policing
The next stage of NDAS, subject to Home Office approval, will be to expand the system’s scope and data processing.
Rather than new data being added in the current ad-hoc manner, data would be fed to the platform continuously, enabling NDAS to provide more pertinent information.
There is also the potential to expand the scope of NDAS to look for key indicators of other crimes, such as domestic homicides. “Very often there are warning signs,” says Donnelly, “but the problem is that the sheer volume of domestic incidents is so great, it is very hard to spot this stuff in the white noise of demand. In the future, we might look at those precursor incidents for warning flags in the data.”
Predictive policing is not intended to replace police officers, but as a support tool. “I think predictive policing will augment decision-making in the future,” says Bouarfa. “By gathering evidence from different sources, and providing that to the police, it allows them to make better-informed decisions.”
With the digitisation of police data and the rise of machine learning, it has always been a case of when, rather than if, predictive policing will finally happen. “We have a responsibility to do something, rather than sitting back and keeping our fingers crossed,” says Donnelly.