"We are all potential victims of deepfakes" says leading researcher

The scourge of ‘deepfake’ crime is among the most serious threats facing law and order in today’s world.

Now a global initiative is leading the fighting back, guided by the science of Liverpool John Moores University’s Forensic Research Institute, whose experts have provided evidence for a new ‘road map’ for tackling fake audio and video content.

Global Guideline: Policy Approaches to Deepfake Detection, is published this month by Interpol and offers advice to law enforcement agencies – and governments - in 196 countries.

We spoke to Dr Emily Williams, a postgraduate researcher, who worked with Dr Colin Robinson and LJMU alumni in compiling the global guidelines.

Emily, who are the victims of deepfakes and what crimes are being committed?

“Synthetic media crime can affect all of us - individuals, businesses, public institutions and the justice system itself.

“Common examples include fraud, impersonation, harassment and non-consensual intimate imagery. But there are also more serious ones, including AI-generated child sexual abuse material, which can cause profound harm to children and facilitate their exploitation.

“Fakes can also undermine genuine evidence through the ‘liar's dividend’, which is when someone claims authentic media is fake. That can affect investigations, prosecutions and the ability of victims to obtain justice.”

When you talk about synthetic media, what are we referring to exactly?

"'Synthetic media' is an umbrella term to cover the variations of 'deepfake' content, i.e. completely AI-generated (deepfake), disguised deepfakes (Leinster fakes), partial fakes (shallow fake and hybrid shallow fake) and other forms of editing. Synthetic isn’t always AI-generated, though, it could just be fabricated by other means.”

How big is this threat right now?

“Massive! In a FoI survey, myself and two other PhD researchers at LJMU (Tom Dawson and Maisie Howard) asked UK police forces about the size of their caseloads and West Midlands Police reported 1,078 in a year. Clearly, those figures only tell us about what has been recorded and identified, and that’s just a single force! There may be many more where faking has not been recognised, or where it has not been reported to the police."

What is the strength of your approach to this challenge, which has clearly impressed Interpol?

“It’s about sticking to what you observe. I tend to approach this from the perspective of what we can actually establish, rather than what a particular technology claims to be able to do.

“AI detection tools can be useful, but they don't necessarily provide the answers. My interest is in understanding how we can bring together technology, forensic reasoning and established principles of evidence to make our decisions more reliable and defensible.

“If my work can help stop someone from having their life ruined because they've been wrongly accused of something they didn't do, simply because synthetic media was mistaken for genuine evidence, I feel I'd have done my duty.”

What advice are you offering law enforcement agencies?

“To start taking synthetic media seriously as an evidential issue, rather than treating it as something that can be solved simply by finding the right detection software.

“There is a lot of focus on how realistic AI-generated media is becoming, but I think we need to step back and consider what that actually means for investigations. What can be established about the evidence? Where did it come from? Has it been altered?

"I would also encourage agencies to think about their people. Not everyone will have the technical knowledge to examine it, but they should be able to recognise when something might need investigation and how to escalate.

“Let’s approach this as a forensic and investigative problem, rather than assuming that technology alone will provide the answer.”

Does this contain findings from any particular primary research?

"Yes. The paper draws on wider research into synthetic media and on the work of the AFTR research group at LJMU. Myself and my collaborating authors, Brian Sheil (Brytestar), Darren Wright (Fiosru), Colin Robinson (LJMU) and Grace Robinson (University of Glasgow), have drawn on immense practical experience and knowledge to translate what we've seen into the recommendations included in the Interpol False Facades paper. 

"The paper also connects with the work I carried out on the Leinster Principle with Jason Rodgers, an LJMU alumnus and forensic officer with Police Service Scotland and on our work with Merseyside Police."