The biggest terrorist threats against Sweden are judged to come from people who act on their own. Finding such people is difficult, but the tool’s algorithms can offer a helping hand.
The AI-based tool has been named Dechefr and is based on several years of research on communication threat assessments. The purpose of the tool is to be able to detect potential lone perpetrators of violence by identifying different types of warning behavior in, for example, emails or posts in discussion forums.
Lisa Kaati is a lecturer at the Department of Computer and System Science at Stockholm University and one of the researchers who helped develop the tool.
– People do not have the opportunity to read through all forms of communication that could be problematic today. Therefore, we need a tool that can help us sort out information, she says, adding that a manual assessment must always be made afterwards.
Among other things, looking for emotional expressions
Warning behaviors that the tool can flag include if a person expresses a lot of anger, is very fixated on, for example, a certain topic or talks a lot about former perpetrators of violence.
– Just because you have one, two or three of the warning behaviors does not mean that you are a person who will commit an act of violence, but it is more about an overall picture, says Lisa Kaati.
Used in the United States
The security police do not want to answer SVT’s questions about whether the authority uses Dechefr or any similar tool in the hunt for potential perpetrators of violence, but according to Lisa Kaati, the tool is unique.
Today, it is used by companies and police authorities who work at various schools in the United States, she says:
– In the US, it is more natural to make threat and risk assessments of individuals. It is more part of a structured work, but I think there is great potential for this tool in Sweden.