How to Spot AI Fakes: Insights from Oklahoma Researchers Ahead Of Presidential Election

Artificial Intelligence experts are preparing for the possibility that AI-generated content creators might attempt to influence the upcoming U.S. presidential election.

In February, tech companies united in a stand against election misinformation; vowing to restrict the use of their consumer tools to interfere with elections. Weeks later, the director of the Federal Bureau of Investigation addressed its preparations for responding to digitally-savvy foreign influence spurred by evolutions in artificial intelligence.

In Oklahoma, academic researchers are teaching ways to identify fake content. "I talk about triangulating," said Laura Dumin, the AI coordinator for the University of Central Oklahoma. "Can we find the same information from three different actual sources? Not three people reposting the same information."

In two recent high-profile examples, AI was used to create a fake robocall from President Biden. And earlier in March, an image of former President Trump with Black voters was determined to also be fake. "I think it's just going to be important for people to take ownership for their role in the election cycle and their perpetuation of misinformation," said Amanda Keesee, the manager of academic technology for UCO.

Keesee and Dumin also shared advice for visually identifying AI-generated content through remnants, which are small pieces of a photo that indicate it might have been edited. Other tell-tale signs include odd numbers of fingers or other small details, particularly on photos of people.

"We do realize all of these AI fakes, it's getting really hard to detect and it's just going to get harder and harder to detect," Dumin said.