‘Responsible data use in criminal investigations? We can learn from the rules on use of force’

In developing a data-driven approach, we can learn a great deal from the rules that apply to use of force. This was what Ruud Staijen, programme director for data-driven policing, concluded at the close of the Police Dialogue at the Netherlands Police Academy’s Apeldoorn campus last Tuesday. ‘The use of force is subject to clear rules, training requirements, and verification procedures. We need something like that for data too.’

‘Use of force is an excellent example we can learn from: the police handle it really well. We undergo use-of-force training every three months. And there’s a consensus: When do I use a firearm? When do I use my police baton? When should I use pepper spray? I think that with data too we will soon be able to say that a certain instance of data use is proportionate or not’, Ruud suggests.

Proportional and responsible

The challenge is for us to reach a consensus about what constitutes proportional and responsible use of data. Ruud: ‘We need to discuss in advance what we do and do not do. What can we explain? This is not just a legal question, but a societal one. What do we entrust to the police? And what are we not prepared to give up, where privacy is concerned. That’s the balance we need to be continuously seeking together.’

Data-driven policing presents us with challenges. That’s one thing everyone in the Police Academy’s conference auditorium can agree on. In a lively dialogue, experts from the Police Academy, operational policing, and the criminal law system exchanged a broad range of perspectives. ‘We live in a world where we constantly leave digital traces. That gives us access to a wealth of information. But it also entails serious risks’, researcher Wendy Schreurs points out.

Self-confirmation or fact-finding?

‘We make use of smart algorithms that instantly discern patterns that human beings don’t pick up. But these sorts of systems aren’t infallible. They may strengthen pre-existing ideas instead of revealing new facts’, Wendy states in her argument. In her view, there is a danger that data-driven investigations may lead to self-confirmation instead of fact-finding.

As Wendy sees it, the police have not yet sufficiently clarified for themselves what a data-driven approach means for good policing. The legislation lags behind the technological developments. For this reason, it is important for us to constantly remain critical and ensure we do not overstep the ethical and legal boundaries. ‘The heart of good criminal investigation work is not just data or algorithms: human judgement about the output of the technology is also crucial’, Wendy concludes.

Bed or dead?

In the course of the Dialogue, Sophia van Toledo of National Investigations and Special Operations talked about the investigation into the encrypted message service Encrochat. In this investigation, the police gained access to millions of chat messages that criminals were exchanging. ‘We could read along, live, with conversations about murders, drug deals, and liquidations’, Sophia explained. ‘We would never have been able to process all that data without algorithms: there’s much more data coming in than you could ever cope with.’

The police, in collaboration with the National Forensic Institute, developed a tool that uses artificial intelligence to recognize suspicious words. That helped them pick out the most important messages. ‘We trained a model on words that had come up in earlier intercepted chat messages’, Sophia explained. ‘Words like ‘spotting’ or ‘sleeping’ could point to an imminent liquidation.’

But human interpretation by criminal investigators is still essential. ‘Someone needs to “sleep”. But does that mean he needs to go to bed, or to be eliminated? Algorithms don’t do context. AI helps with filtering data, but you need an investigator to draw the conclusions.’

Rules still in their infancy

‘Rules about the use of data in police investigations are still in their infancy’, says Frederick Leentfaar, a lawyer specialised in privacy issues. For that reason, he thinks it is extremely important that data use is properly delineated in the preliminary phase. That prevents differences of opinion later in court. Moreover, as Yolanda van Setten, deputy justice and legal adviser, adds, ‘the evidence must also be replicable’. And that is not straightforward with AI: ‘A self-learning system is constantly changing. If you can’t explain how a conclusion came about, the court cannot verify it.’

Another problem with AI is that judges often do not fully understand the technology used. ‘They are not exact scientists’, Yolanda points out. And more generally, too, sufficient digital knowledge is lacking in the criminal justice system. Public prosecutors, defence lawyers, and investigating magistrates all often struggle to understand the technology underlying data-driven investigations. More joint training would be a good first step, someone in the audience suggests. The idea is also put forward that the Police Academy might be able to fulfil a role in this for the entire criminal justice system.

What the Dialogue reveals above all is that there is a great deal that is not yet clear. And that we must never lose sight of the ethical sides of the use of algorithms in police investigations. ‘It’s a question of finding the right balance between security and the type of society we want to live in’, explains ethicist Mijchanou Kowalczyk. ‘If your name turns up in a certain dataset, that may automatically earmark you as a suspect, even if you are completely innocent’. Ruud Staijen agrees: ‘Basically, it’s all still a bit up in the air. Data-driven policing has huge potential, but it does also entail risks. It’s vital to get the balance right.’