Improvements to machine-intelligence will improve accuracy and speed up police investigations

Against a backdrop of increasing crime and security concerns across Europe, the West Midlands Police force in the UK, along with police in Antwerp, Belgium, have started putting machine-intelligence technology through its deductive paces. The EU-funded VALCRI project has set up trials that make use of three years’ worth of actual but anonymised crime data, equating to 6.5 million records. The VALCRI system performs crime scene analysis by scanning millions of mixed format information sources – such as records, interviews and pictures – within seconds. The system detects suspicious patterns and reconstructs scenes, pointing to promising investigative leads and presenting the findings on interactive touchscreen displays for analysts.

This methodology holds out the hope of removing the laborious and time-consuming aspects of analysts’ jobs, so that they can more proactively pursue lines of enquiry, building cases with more speed and precision. Crucially, the expectation is that the system will also identify connections that might be missed by a human, through error or bias. To counter objections about the replacement of human bias for machine bias, the system renders the process transparent so that reasoning steps can be retraced.

Visual analytics for investigative insights

What makes VALCRI (Visual Analytics for Sense-making in CRiminal Intelligence analysis) especially powerful is that it capitalises on advances in artificial intelligence combined with visual analytics, to enable real-time analytic data interaction. The system also uses facial recognition software to identify individuals from sources such as CCTV.

Recently covering the project, the New Scientist reports project leader Professor William Wong of Middlesex University, London as saying, ‘Everyone thinks policing is about connecting the dots, but that’s the easy bit…The hard part is working out which dots need to be connected.’

Currently, one of the first steps in criminal investigations is to trawl police databases for similar incidents. Links could be made based on criteria such as timing, people involved, location or signature features, such as modus operandi. Another of the project team, researcher Neesha Kodagoda, also quoted in the New Scientist article, states that, ‘An experienced analyst needs 73 individual searches to gather all of this information, before manually putting it into an easily digestible form … VALCRI can do this with a single click.’

However, previously machine intelligence often struggled to make connections between some phenomena which for humans is straight-forward. For example, the descriptive phraseology people often use is subjective and varying, with words like ‘dirty’ or ‘messy’ describing much the same thing. Now an advanced algorithm enables the system to make better conceptual links. VALCRI also integrates machine learning capability which means its analytical faculties improve with increased data and criminal profiles exposure.

Investigators view the analytical results on a touchscreen display, described as a ‘Reasoning Workspace’. This workspace comprises three areas which allow interaction with: the available data, the computational analysis (which can include imported data) and the resultant conclusions based on the evidence assembled. By allowing for the manipulation and organisation of data through selection and dragging, the display is more intuitive than traditionally more static sources, such as lists. In this way the system encourages more imaginative, creative and insightful sense-making and problem solving. Additionally, investigative overviews can be presented in various visual formats such as maps or graphs.

Building the case for deployment in real-time criminal investigations

While there is broad agreement that taking VALCRI to the next stage of development, requires access to non-anonymised data generated during a live investigation, there are also predictably a range of concerns about doing this. From a legal stand-point, use of the system could be challenged in court which could make it counter-productive in criminal prosecutions. Additionally, with transnational investigations, countries differ in their data protection laws making the the sharing of actionable intelligence fraught. And there are, of course, on-going high-profile societal debates around police and security services access and use of personal information.

Yet, in an era of heightened security concerns with pressure mounting for those charged with civil protection, any approach which increases the speed, precision and pre-emptive nature of that response, is highly likely to be on the table for further exploration.

For more information, please visit:
project website