Using algorithms in police investigations

Date of news: 26 March 2020

Self-thinking computers that can help the police to solve murder cases is a vision of the future that is becoming more and more likely. The police are already using algorithms to analyse large amounts of data. This smart technology is particularly useful when it comes to police investigations.

Anyone who has seen films like The Matrix, I Robot or Minority Report, will not be unfamiliar with the idea of an information-driven computer that controls everything and everyone. Whether this is something that also lies ahead for the police, Business Intelligence Expert Oscar Wijsman of the Police Service Centre (PDC) dares not say, but he hopes that it is not the case. Wijsman is an expert in the field of big data, analytics and data science. He is involved in the Future-proofing Investigation and Prosecution programme (TOV), which was launched in 2016 in the interests of renewing the police investigation process. Wijsman finds the idea of using artificial intelligence for police work quite intriguing. “We’ve been using algorithms in our investigations for at least ten years and these are continually being developed further in the direction of artificial intelligence.”

Imagine feeding an algorithm a large amount of data about suspects in a criminal network or a specific area. The computer then makes connections between people and criminal markets and recognises patterns. If you keep feeding the algorithm data, you’ll end up with a multitude of networks with millions of links and connections, which is something that the human brain can no longer oversee. The computer consequently maintains this overview and notifies you when a notable change occurs. It will even give you a perspective on when to take action, such as the interventions that can be carried out.

Text mining and image recognition

This is still certainly a future prospect, but it is no longer elusive, says Wijsman. For years the police have used text mining, which offers the option of searching large amounts of text for certain keywords, correlations, persons and patterns. An algorithm simplifies this type of work, because it is essentially a mathematical formula that allows you to rapidly search and filter information. This produces a particular result, which is dependent on the criteria that the algorithm has been given. Among other things, text mining is now being used to scan information from different criminal investigations, registrations and data from smartphones. The police are also investigating whether algorithms can be used to analyse cold cases, for example, to make connections between cases and to find promising evidence that can then be re-examined.

Although this technique has long been used, the work with algorithms continues to develop. Wijsman: “We’re working on converting wiretapped conversations from speech to text. This text is then analysed on the basis of connections with the other information that we have. This isn’t such an easy task, because criminals often use coded language that the computer finds hard to identify. Accents, background noise and jargon also need to be taken into account. The trick is to get the computer to understand this. I think that this technique will be operational in one to two years’ time.” Image recognition has also been used for some time to catch criminals. The Amsterdam Operational Information Efficiency Team has developed an algorithm that makes it possible to read specific images on a phone. Photos that contain an IS flag, or weapons or drugs can consequently be recognised much faster, especially when thousands of images are stored in the phone’s memory. “The technology is ready, but we still occasionally struggle with the legal aspect. The fact is that police are only allowed to view information, about a suspect for example, that’s relevant to their investigation. The image files may contain photos of medical records. We’re not allowed to see these images, let alone analyse them. We also need to ensure that we stay within the legal framework.”

Self-learning algorithms

The algorithms that are now being used in police investigations are called ‘weak artificial intelligence’, because they focus on the execution of one single task. For example, if you tell the computer to “search through all the images and find a pistol” this will yield a result. But if you also wanted to search for guns or pocketknives, you would have to retrain the algorithm to recognise these particular objects. This type of algorithm teaches itself to recognise new objects by practising on a huge number of new images. “And it’s not always the case that the computer itself decides what it will learn; that’s something that we determine,” explains Wijsman. “The next step will be what we call ‘strong artificial intelligence’, because it involves a form of intelligence that more closely resembles the workings of human intelligence. This means that the computer is potentially capable of executing general tasks, and understanding and thinking through complex problems. In this case, the computer learns without us giving any explicit indication. The results are not predictable like they are in the case of weak artificial intelligence. This is a form of artificial intelligence that’s still in its infancy.”

If the technology is available, why aren’t the police using these self-learning algorithms to take immediate action? Jan ter Mors, Programme Director of Intelligence, explains: “We’re currently using a strategy that contains a number of general perspectives on algorithms and artificial intelligence. For example, we never allow automated decisions to be made by algorithms; decisions are always made by people. But because the world around us is changing and data is becoming increasingly important for investigations, we’re collaborating with a number of portfolios—such as business operations and digitisation and cybercrime—on a new strategy that focuses more specifically on the use and treatment of artificial intelligence.”

An algorithm for each murder

More and more data are available to help detectives solve murders. For example, there are a variety of sensors that record images of us or register our assets, there are databases that are filled with our data and we are increasingly able to trace evidence from a crime scene. How are you supposed to select the persons of interest in a murder investigation when you have this much information? This is the exact question that investigative psychologist Daan Sutmuller has endeavoured to answer in his PhD research.

“Murder investigations do not seem to employ a generic algorithm for answering this question,” Sutmuller explains. “This is probably because there is a wide range of murders, while the features of a crime scene can overlap. For example, stab wounds can point to both a relational murder and a psychotic perpetrator. Because every murder case is unique, it has its own algorithm. This is why I’d like to produce a software package that contains a library in which the elements of evidence are assessed. This will allow an analyst to build an algorithm for a particular case. By using the building blocks from the library and adapting them to the particulars of the case, the algorithm should reveal the persons of interest. The results should point the detective in a certain direction. This may require searching the victim’s social network or investigating the people who used their phone near the crime scene. The algorithm should support the detectives in making their choices.”

Control mechanisms

In addition to future-proof policies, there should be control mechanisms that monitor the work of the police and ensure legal quality. It is for this reason that Wijsman is involved in the Big Data Quality Framework, a method that must be used by data scientists and developers when building an algorithm. It is a kind of step-by-step plan for checking which problem you want to solve, what the algorithm is for, what data you are using and whether your data or algorithm contains prejudices, because this can yield incorrect results. What also must be explainable is how an algorithm works or how a certain outcome has been reached. “This method is particularly intended to help us explain to the judge exactly what we are doing,” says Wijsman. “Because if the police can’t explain how artificial intelligence was used to arrest a suspect, we won’t have a leg to stand on in court.”

Bas Testerink, who is a researcher at the National Police Lab AI, is in the process of automating the explainability of algorithms. The lab is a collaboration between the police and several academics from the field of artificial intelligence. Testerink: “I want to create a kind of building block that detectives can use in the future to assist them in their work.

This building block must be filled with specific information for each case. The algorithm then calculates and justifies an action perspective and suggests it to the detectives. What’s more, the algorithm must provide a contradiction, so that the detectives don’t end up with a tunnel vision. The idea is that the system stores the working method and logic itself, so that these can automatically be explained to the judge.”

‘Beware of data hunger’

Human rights treaties protect the right to privacy. But under certain conditions, these treaties also allow the police to infringe on the privacy of suspects, otherwise nothing can be accomplished in an investigation. The degree of freedom that the police are consequently given depends on the seriousness of the crime and the degree of privacy violation. This is the theory of Frederik Zuiderveen Borgesius, Professor of ICT and Private Law at Radboud University.

“For example, some politicians believe that following an attack, the police should use all possible means to find and punish the perpetrators. The seriousness of the offence ensures that the ends justifies the means.” According to Zuiderveen Borgesius, this is an example of what happened after the terrorist attacks in Madrid in 2004 and in London in 2005. “There was a massive outcry, both among the citizens and in political circles. The Data Retention Directive was subsequently adopted in the EU, which meant that telecom companies were required to store all metadata, such as phone records, for a limited period of time. This also included data from customers who were not suspected of anything. Politicians argued that this information could always prove to be useful for tracking down terrorists or criminals. The guideline was controversial on account of its privacy violation aspect. In 2014, the Court of Justice of the European Union declared the directive invalid because it violated the right to privacy.”

This so-called ‘data hunger’ among politicians is something of which we should be wary, says Borgesius. The professor believes that algorithms can help to advance the cause of police investigations by automating certain processes. Despite the fact that each suggestion made by an algorithm is always checked by a human, Borgesius still has a number of concerns. “For one thing, people often find anything that is reported by a computer to be very convincing, as if it were the truth, even though the suggestion made by a computer isn’t necessarily correct. The second reason is that employees like to play by the rules. So, if an algorithm suggests that ‘Peter is acting very suspiciously’, several detectives will act on this information. But if a detective has doubts about this claim, and subsequently ignores it and Peter ultimately commits a crime, the detective will obviously be afraid of getting into serious trouble. Some employees think that if you always follow the rules, you can’t get into trouble. And this type of attitude can actually cause problems. You should always use your common sense.”

Machines enhance humans

Wijsman believes that even though the computer will take over more and more of our work in the future, people will always remain central to the investigation process. “It’s actually a good thing that the computer is doing more work for us, because it means that we can work much faster and more accurately. Because the world increasingly revolves around data, there is an increase in the amount of data that is available within an investigation. The human brain will never be capable of processing and remembering so much information, but a computer will. Effective collaboration between man and machine is vital.” Wijsman once again refers to the example of mapping criminal networks: “If your detective work involves mapping out a network, that’s manageable. But more and more information is being made available by such instances as municipalities, our own registrations, the Tax and Customs Administration, and the Chamber of Commerce. If it also turns out that one of the suspects is active in another criminal network, things start to become very complicated. Eventually, it’s too much for one person to manage. The computer can combine all of the available data and can see connections that would be overlooked by us. Information that may initially seem irrelevant may subsequently prove to be valuable if it contains a pattern or connection. The computer is not only capable of providing an overview and making all the connections, but it is also able to report suspicious changes within a network and consequently make suggestions. So, for example, if you’re pursuing a person of interest in a weapons network, you’ll also know whether that person is involved in drug trafficking and extortion. Eventually we’ll be able to carry out interventions on the basis of the computer’s suggestions because we’ll see that something is potentially going to happen. As a result, we’ll even be able to prevent crime.”

Future vision

Ideally, the police investigation process would involve using all of the available information, which is something that is not happening at the moment. “Detectives are currently only able to follow a number of leads at a crime scene, because they’re only allowed to submit a few pieces of evidence to the Netherlands Forensic Institute (NFI). In the future we’ll hopefully be able to process all the information in the computer, which will enable us to submit the most promising evidence to the NFI. This would also mean that you wouldn’t have to select as much information yourself, because you’d be able to save all of the information. And this means that you wouldn’t overlook any possible evidence.”

Storage of all of this information requires a new system that collectively encompasses all of the police’s big data environments. The Police Data Platform, which is currently in development, should offer a solution. “This means that police officers in every region can collaborate more effectively and that we will be able to see more coherence in the information that we have. Once we’ve reached that point, we’ll get ready to start using artificial intelligence.”

Text: Romy Donk. This article previously appeared in the Dutch National Police magazine ‘Blauw’. Image by Robynne Hu via Unsplash.