UC Berkeley expert warns AI is making it harder to trust media

In short: As AI-generated fakes spread beyond photos to audio, video, and text, UC Berkeley researcher Hany Farid says proving what is real now needs both technical checks and broader rules and labeling systems.

What's going on

Hany Farid, a digital forensics expert at UC Berkeley, says generative AI is making it harder for people to trust what they see and hear online. Generative AI is software that can create new content, like a photo or a voice clip, based on patterns it learned from lots of examples.

Farid is known for studying how to tell real media from synthetic media. That can include looking for physical clues in images, like whether shadows and reflections make sense, or whether lighting matches the scene. It can also include statistical clues that suggest an image was made or altered by a computer.

But Farid warns that these clues are not enough on their own. If you do not find clear signs of editing, it does not prove something is real. He also points to the “liar’s dividend,” meaning once fakes are common, people who get caught on real audio or video can claim it is fake and some audiences will believe them.

Farid is also pushing for more than detective work. He supports content-authenticity efforts like the Content Authenticity Initiative and C2PA-style standards, which aim to attach a kind of history label to media (like a receipt that shows where it came from and what changed).

What to watch

Farid and others say detection tools have a limited shelf life because AI generators keep improving. Watch for platforms and governments to decide whether to require clearer labels, stronger source tracking, and penalties for harmful deepfakes.

Source: NYTimes