Countering AI-Generated Fake Video in the Age of Disinformation

Justin Sherman wrote for Technology for Global Security (Tech4GS) about countering deepfake video.

Jordan Peele, director of the Oscar-nominated film Get Out, made headlines in April 2018 with a satirical voice impersonation of former U.S. President Barack Obama. However, Peele didn’t appear in the video — Obama did.

Peele drew public attention to a technique that is at least a year old — using machine learning to modify another person’s mouth movements. While the background scenery of this particular clip is real, President Obama’s words and lip movements are not. The video seems rather innocuous at first, until a formal address on the dangers of fake news turns into a critique of U.S. President Trump and commentary on the film Black Panther. Early on, however, it is not readily apparent that it’s a fake.

The video is startling, but the underlying digital techniques have far more varied — and potentially problematic — applications in the creation of so-called “deepfakes.”