At least one police force in the Netherlands is now exploring ways to use A.I. as a means to solve cold cases. In fact, Dutch authorities have already started using the technology in the form of a machine learning algorithm in an 800-member strong division of the police force called "Q." That's according to Q co-founder Roel Wolfert, who says that the A.I. will be invaluable in solving the cases. However, the technology isn't necessarily being used in the way most might assume and, at least for now, it's still in the learning phase. That may be a good thing since the use of A.I. and machine learning in policing has become something of a hot-button issue. Q won't scanning live police camera feeds or even really putting the clues together itself. Instead, it will be examining the more than 30 million pages worth of evidence and documents in order to provide each cold case with a solvability score. The score will be based on current evidence and documents, in addition to the timeframes involved – such as when it was last examined and what new advancements have been made in forensics.
That's bound to be helpful due to the sheer volume of work involved in the process. To qualify as a cold case, cases typically need to be at least 30 years old and carry 12-year minimum sentences. The length of sentencing, if a conviction were to be made, means that a not insignificant percentage of those involve murders. Given the sheer number of cases and the amount of evidence involved, machine learning could allow the police to place their efforts where they will do the most good. What's more, the police think the technology could be used in new cases once it has learned how to analyze evidence sufficiently. Specifically, the team working with Q think that it could help during the crucial initial phases of an investigation to determine which leads are worth following up on.
There are, of course, still plenty of barriers that need to be overcome. The first is the digitizing of the current system for storing cold cases and associated evidence and documents. That's going to take quite some time to accomplish. There's also the problem of "non-forensic" evidence, which covers things such as witness statements and the ability to analyze information with a more human approach using social sciences and social networks. Those could expand on the algorithm significantly but they are typically more abstract than what the technology is used for. Bearing that in mind, it seems as though humans will need to continue augmenting the systems being built to perform these types of tasks for now.