�Ǹ�Թ�

Illustration by Dana Heimes

Public outcry following the deaths of Renée Good and Alex Pretti earlier this year at the hands of immigration agents was swift. So, too, was the flood of content altered and generated by artificial intelligence that purported to show the victims as aggressors.

The idea of media manipulation in support of a government narrative is chilling—especially when two people were killed. But if there is a sliver of a silver lining to these tragedies, Sandra Ristovska sees it in how most Americans rejected deepfakes and A.I. content as they spent time with, and tried to make sense of, footage from police body cameras and bystanders’ smartphones.��

“People heard the official narratives and statements, saw multiple videos from different angles and perspectives, and arrived at their own interpretations,” said Ristovska, an associate professor of media studies and an expert on visual evidence in the age of generative A.I.; she also holds an affiliate position on the faculty of the university’s law school.

Ristovska just completed a yearlong fellowship with the Center for Advanced Study in the Behavioral Sciences, at Stanford University, where she had the opportunity to “dream again, as an adult” as she developed new perspectives on her research work, which has strong legal and ethical overtones. At the same time, she was highly sought by journalists for her perspectives on how courtrooms, newsrooms and classrooms need to rethink our relationship with visual media, including A.I.��

Though we need to develop better verification and detection tools around A.I., “the solution is not just a technical fix,” Ristovska said.��

“There is this kind of long-standing desire to outsource human judgment to technology, which has never played out well for us,” she said. “We are better off looking at evidence and putting together the story, rather than just blindly trusting any one technology.”

In fact, another high-profile case from Minneapolis followed just this course. When Derek Chauvin was tried in connection with George Floyd’s murder, “prosecutors showed video, but at the same time, they were carefully bringing in other witnesses. They had forensic reports. They had multiple videos—from bystanders, police body cameras, dashboard cameras, surveillance cameras,” Ristovska said. “The way they crafted the case shows they very much understood that no single��video speaks for itself.”

Ristovska’s work, which includes her role as director of the college’s Visual Evidence Lab and a planned book, challenges us to be critical thinkers who understand both our own fallibility and the limitations of video and photography.

“The evidentiary challenges A.I. poses are neither entirely unprecedented nor primarily technical,” she said. “In terms of our court system, we need to think about safeguards that empower judges and jurors with research-based instructions that help mitigate some of the challenges with video."

“To me, it’s just as important that we ensure rigorous visual interpretation of the evidence as it is that we worry about whether the image is real��or not.”��