A researcher focused on aiding courts in identifying evidence that has been created or altered using artificial intelligence is facing challenges in developing a practical tool for real-world applications. Maura Grossman, who co-directs a team of researchers in Canada, is working towards creating an open-source and user-friendly tool intended for use by courts and other stakeholders within the justice system.
As the two-year project approaches its midway point, Grossman reveals that the team has successfully established a system that utilizes advanced algorithms. These algorithms function collectively, acting as a committee that votes on whether a particular piece of evidence is AI-generated. However, during Grossman's attempts to apply this tool in real-life scenarios, the algorithms were unable to achieve a consensus that would render them useful.
She describes the situation as a "wicked technical problem," highlighting the significant challenges her team must overcome in the upcoming phases of the project. Grossman, who is also a computer science professor at the University of Waterloo and teaches at York University’s Osgoode Hall Law School, emphasizes that their next steps will involve engaging with law enforcement and policymakers. While initially optimistic, she acknowledges the daunting nature of the task at hand.
Grossman notes, "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." She adds that the team remains hopeful but has become more realistic in their expectations. They are now recognizing that a singular tool may not be the ultimate solution. Instead, addressing the issue will likely require collaborative efforts from legal frameworks, regulatory measures, and enhanced public education on discernment of real versus fake content.
A significant obstacle in the project arises from the way digital evidence is often altered. For instance, when an image is shared on social media, it typically undergoes compression and loses its metadata. Consequently, the shared version sufficiently differs from the original file, complicating its use as evidence. Grossman states, “It’s not going to have that information that the original copy had, and chances are the original copy is gone three years later by the time you sue me for saying something horrible about you or whatever I did.”
Additionally, she explains that as files are emailed and shared among multiple people, the data becomes increasingly distorted. This deterioration exacerbates the challenge of establishing the authenticity of digital records over time. The research initiative is one of two inaugural projects funded through the Canadian Institute for Advanced Research’s Canadian AI Safety Institute program, which was launched by the federal government in 2024 as part of its broader AI safety strategy.
The research team comprises technologists and legal scholars affiliated with universities across Ontario and British Columbia, complemented by an advisory board that spans North America, including judges and organizations aimed at supporting self-represented litigants. Upon the project’s announcement last year, Grossman observed that courts were ill-prepared for the impacts of AI. She criticized existing commercial tools, labeling them as lacking in transparency and reliability, especially within the judicial context where the stakes are high.
Moreover, Grossman reiterated that her team never anticipated their tool to act as a comprehensive solution, stressing that any technology would still necessitate human involvement and critical analysis. "You’re going to need somebody to be able to testify why they think what they think and all of that," she explained. She also pointed out the importance of addressing policy considerations, suggesting that certain AI-generated materials might need to be prohibited entirely instead of solely focusing on detection after the fact.
