How Do We Stop Deepfakes From Tricking Juries?
Reflecting on the evidence that passes through her Phoenix, Arizona courtroom, superior court judge Pamela Gates says she’s becoming less confident that the average person can sort out the truth.
Say a victim presents a photograph showing bruises on their arm and the defendant argues that the injuries were digitally added to the image. Or perhaps a plaintiff submits an incriminating recording and the defendant protests that while the voice sounds identical to theirs, they never spoke the words.
In an era where anyone can use free generative AI tools to create convincing images, video, and audio, judges like Gates are increasingly worried that courts aren’t equipped to distinguish authentic material from deepfakes.
“You had a better ability to assess [evidence in the past] just using your common sense, the totality of the circumstances, and your ability to verify the authenticity by looking at it,” said Gates, who is chairing an Arizona state court workgroup examining how to handle AI-generated evidence. “That ability to determine based on looking at it is gone.”
The explosion of cheap generative AI systems has prompted some prominent legal scholars to call for changes to rules that have governed court evidence in the U.S. for 50 years. Their proposals, including several that were reviewed by a federal court advisory committee earlier this month, would shift the burden of determining authenticity away from juries and place more responsibility on judges to separate fact from fiction before trials begin.
