AI can detect abusive messages 21 times faster than humans

New digital forensic research has found that artificial intelligence (AI) models can accurately identify threatening and abusive messages up to 21 times faster than humans, which could help significantly increase the speed of police investigations.

Virtually every crime has a digital element, and the NPCC estimates there is a backlog of 25,000 digital devices waiting to be examined as part of live investigations in England, Wales and Northern Ireland.

The 12-month study was conducted by digital forensic experts from the Forensic Capability Network (FCN) – the national body for forensic science funded by the National Police Chiefs’ Council’s (NPCC) – and AI experts from the University of Warwick.

By analysing synthetic or ‘dummy’ data using customised natural language processing (NLP) models on secure servers, the project aimed to identify the effectiveness of AI at identifying threatening and abusive language related to violence against women and girls (VAWG).

The FCN’s models were able to comb vast amounts of text-based material to quickly find words and phrases that could be relevant to a criminal investigation, even when the abuse was subtle or used slang and colloquial phrases which are more difficult to detect.

In one test, the model took just over one minute to identify three aggressive and emotive phrases within 456 messages, around 21 times faster than an average human investigator.

The project also successfully demonstrated that AI could help protect the privacy of VAWG victims, as it was trained to home in on relevant messages and ignore material irrelevant to the investigation.

Researchers suggest the technology could be used in future as an additional tool to flag potential items of interest to investigators, and AI wouldn’t make investigative decisions itself.

Image

The NPCC’s Digital Forensic Lead, Northumbria Police Deputy Chief Constable Jayne Meir, said:

“These findings are very encouraging. Analysing huge amounts of data has become a crucial part of modern policing, and the volume of data is only growing. So, we must explore the most efficient and streamlined models for doing this investigative work, ones which achieve the best quality while alleviating the burden on police resources.

“Our police investigators and digital forensic specialists will always be the ones to make decisions. But if we can help them analyse evidence faster with technology, then we should absolutely explore that.”

Based on the project’s success, the FCN intends to test the NLP models on a wider range of offences beyond VAWG, including high volume crimes such as drug offences.

The FCN’s lead scientist Simon Cullen, who led the project, said:

“We’re some way off taking these models from a test environment into operational policing. But we’ve shown in theory that carefully customised AI models can operate in the background, and flag useful information that could be relevant to an investigation. Then, it’s up to a human to decide what should happen next. In terms of other use cases for this technology, we think the impact could be even more effective in tackling volume crime.”

The research was funded by the Office of the Police Chief Scientific Adviser (OPCSA) through a science, technology, analysis and research (STAR) grant.