Why Britain’s police forces are taking to AI

HARRY SCHONE’s job is to work out how the police can use artificial intelligence. At Police Scotland’s headquarters, a glass cube in Glasgow’s East End, he has assembled a team of coders and engineers who stand out among strait-laced colleagues. They are working on an array of schemes: a program that transcribes evidence; a model that helps shift-planners deploy officers; a tool that matches reported thefts with ads on resale websites.

Across the rich world police forces are struggling. Many have had their funding cut. In Britain sluggish response times, low clear-up rates and a series of scandals have left public confidence in policing near a record low: 51% of people think the cops are doing a good job, down from 75% in 2000.

What if there were a silver bullet? Policing—perhaps more than any other public service—could be transformed by AI. Whether that opportunity will be grasped, however, is open to question.

Policing is an old craft that has often resisted change. In the 20th century police chiefs opposed the introduction of motor cars (horses were just fine) and radios (officers would get lazy). In the 21st the rank-and-file resisted computers, preferring to write their case notes by hand. But at its core policing is about intelligence, and involves processing vast troves of information. That makes it a good test case of AI. “People underestimate how much this could transform our service,” says Superintendent Lewis Lincoln-Gordon of the National Police Chiefs’ Council.

Take transcription. Police Scotland employs 40 typists. Most forces still operate a typing pool to transcribe interviews and produce evidence for court. Officers, trained to fight crime, spend a lot of time filling in forms. In England and Wales half a million officer hours each year are wasted on unnecessary paperwork.

Mr Schone’s team has built a transcription tool, similar to those embedded in most videoconferencing software. They had to make it themselves, for data-security reasons, but the proliferation of open-source models has made that simple with decent coders. It is not yet as accurate as a typist, who will still be better for the most sensitive documents. But it can be applied at an almost infinitely larger scale, constrained only by computing power, freeing officers and speeding investigations.

Technology should also transform how evidence is gathered. The established model—someone calls up and is told to come into the station—is antiquated. London’s Metropolitan Police has recently introduced a chatbot for reporting crime. Some forces are making it easier for people to quickly submit evidence, such as footage from a dashcam or security camera.

Visit the headquarters of Essex Police in Chelmsford, and you will find a dozen headphone-clad officers on video calls. This is the rapid-video-response team, which handles domestic-abuse cases. “Many victims don’t want to come to the station or have a police car turn up on their road,” says Sergeant Stacey Rothwell. The call is automatically transcribed and put into a case file. Some perpetrators have been arrested within two hours.

As the mountain of evidence grows, AI can extract what is useful. In another Police Scotland project, advanced analytics is being used to scour data on sexual abuse of children. Facial recognition, powered by AI, has been piloted by the Met and South Wales police, and has proved helpful. One force used cameras to identify paedophiles trying to attend a Taylor Swift concert. But adoption has been slow. In August the Home Office announced the roll-out of ten facial-recognition vans, to be shared across all of England and Wales.

AI’s potential to improve policing is large, but three obstacles will get in the way. One is money. In England and Wales the police budget will rise by 1.7% a year over the rest of the parliament, not enough to keep pace with staffing costs (this year’s pay award was 4.2%). So capital budgets have been raided. The central pot for new technologies, including AI, will be halved next year. More widely, police chiefs’ hands have been tied by ministers’ obsession with getting bobbies on the beat.

Second, as in the past, is inertia. As employees of the crown, police officers cannot be made redundant. Nor, because of collective bargaining, can their administrative roles be easily adapted. Most forces have not prioritised digital skills. And, as Mr Schone puts it, “young people just don’t think about policing as a tech career.” Police Scotland, which is the second-largest force in the country, has some pulling power. Most forces in England and Wales are far smaller. The Met, Britain’s largest, would struggle to offer the salaries needed to attract tech talent in the capital.

Third is the risk of public opposition. People think AI can improve policing and support its use. But there could be a backlash if they get the sense that machines are making decisions. To uphold Britain’s model of policing by consent, police leaders will need to explain how their methods are changing. That has not always been a strength: the outcry over facial recognition has in part been a failure of persuasion.

All this will slow progress, despite the obvious benefits. Yet as Mr Schone notes, AI is already transforming criminality. The police can ill afford to stand still. ■