Detecting deepfakes, in a world where even reality is suspect

In February, a Tomahawk missile struck an Iranian elementary school, resulting in over 150 deaths, mostly children. President Trump speculated that it might have been launched by another country, emphasizing that the missile is generic and commonly sold. When Iran provided video evidence, Hany Farid, a renowned expert on deepfakes from Dartmouth College, was approached by media to assess the authenticity of the footage.

Farid faced challenges due to the poor quality of the grainy, shaky video, and the small size of the missile. However, he could analyze certain aspects scientifically, such as the speed of sound, light, and the missile's dimensions, leading him to conclude that the video was authentic and the claims were true.

Farid expressed frustration that one cannot confidently present evidence in court to convict someone, questioning the judge's authority to accept AI-generated evidence. This issue was illustrated by a case involving Alex Pretti, who was killed by an ICE agent in Minnesota. A blurry photo of Pretti showed an enhanced image with a gun in his hand, which was not present in the original footage.

Farid highlighted that AI has learned to create convincing enhancements based on common patterns, such as the presence of guns around armed individuals.

Approximately half of the content widely viewed on the internet is fake, labeled as "AI slop," which includes manipulated footage of airplanes, toddlers, and Elvis. AI is being utilized maliciously for various purposes, including creating child sexual abuse material, non-consensual intimate imagery, state-sponsored disinformation campaigns, and even aiding in job interviews by concealing identities and geolocations. Many Fortune 500 companies have unknowingly hired individuals using AI-assisted techniques.

Farid emphasized that the issue of deepfakes extends beyond entertainment, as they are weaponized in serious ways, from creating harmful content to manipulating job interviews. He expressed concern over the increasing prevalence of AI-generated content, with estimates suggesting that half of the online material is fake. The constant exposure to such content takes a toll on Farid, who struggles with nightmares and anxiety due to the disturbing nature of the material he encounters daily.

To cope with the stress, he returned to his rural property in Vermont, where he focuses on mowing meadows and cutting wood to escape the disturbing images. However, the pressure and social media scrutiny continue to be a source of concern for Farid.