Tool to help courts spot AI-generated evidence is proving hard to build, researcher says

Maura Grossman, a University of Waterloo computer science professor, co-leads a team building a free, open-source tool for courts. It combines several algorithms that vote on whether something is AI-generated.

I don't know that I ever thought it was going to be easy, but it's turned out to be almost daunting how challenging this task is.

— Maura Grossman, researcher

In two real-life tries, the algorithms could not reach enough consensus to help. A social media image is compressed and stripped of its metadata, so it differs from the original by the time it is presented as evidence.

What's next

The two-year project is nearing its halfway point, and the next phase will seek input from law enforcement and policymakers. Grossman says no single tool will be the answer: law, regulation and public education are needed too.