Defending against a tsunami of deepfakes

But side effects should be controlled and directed in a beneficial way to adopt new technologies for the good of society.

Photographic “deepfakes” of American superstar Taylor Swift went viral on X, formerly Twitter, over the weekend. Deepfakes refer to explicit videos, audio or images concocted to make people appear to do or say things they didn’t, with the help of AI tools. X, with a policy to not allow synthetic or manipulated media on its platform, took action to remove the fake images of Swift more than 17 hours after they were posted and viewed over 47 million times.

A week earlier, a doctored phone message sounding like President Joe Biden circulated, telling voters in New Hampshire not to waste their vote in the Democratic primary. The bogus message spread quickly, shortly before the primary voting period. Deepfakes can be easily produced with generative AI technology. We must build protection against a deepfake tsunami.

The Korean legislature last December passed a revision to the Public Office Election Act outlawing the use of deepfakes for campaigning 90 days prior to election day. The law on sexual violence also has grounds to punish posting and circulation of deepfake-based explicit materials. Whether or not that will be sufficient to sanction fabricated content, that can be made within minutes, is questionable.

Obscene materials and fake campaign contents generated with AI-enabled deepfake tools can go viral on social media in seconds. But removing and blocking the materials or regulative actions from authorities can come too late. Deepfakes overruled facts in last year’s elections in Turkey and Slovakia.

Fortunately, the tech industry is quickly moving to filter deepfakes. Intel has introduced AI-enabled real-time deepfake detectors. Big-tech giants like Google and Microsoft are using AI tools to digitally watermark fabricated images to help contents consumers better identify misinformation. Platform operators must employ these tools to combat the spread.

X announced it would be creating a team to monitor sexually explicit materials. The platform is grappling with the embarrassment after Elon Musk sacked contents inspecting staff when it acquired Twitter in October 2022. Korean platform operators also must take heed to the danger.

But reactions must be contained. Governments are vying for leadership in AI technology and application. Deepfake technology itself is not evil. The late Song Hae — the legendary host of the National Singing Contest from 1988 until his death in 2022 — was able to “appear” on a TV drama all thanks to deepfake technology. The technology is being actively employed in broadcasting, entertainment and games. Synergy would enlarge when the technology is met with augmented reality and virtual reality capabilities. But side effects should be controlled and directed in a beneficial way to adopt new technologies for the good of society.