DEMOCRACY VS DEEPFAKES

GENERATIVE AI IS NEITHER AN APOCALYPSE FOR INDIAN ELECTIONS, NOR 'JUST NOISE'. TO COUNTER IT, WE MUST STRENGTHEN OUR INSTITUTIONAL CAPACITY AND SOCIAL NORMS

- By Renée DiResta

In India’s 2024 general election, more than 50 million AI-generated voice calls reportedly went out in two months. Deceased politicians were synthetically resurrected to boost successors. A politician appeared in campaign content speaking four languages he doesn’t know. Some of this was outreach—campaigns scaling across 22 languages. Some of it, like faked content of an actor criticising the prime minister, was manipulation. The technology enables both.

Since 2018, analysts, technologists and journalists have warned that generative AI will upend elections, erode trust and create a flood of sophisticated disinformation. Each upcoming election renews the fear. When catastrophe doesn’t arrive, a predictable backlash follows: it was an overhyped panic! Maybe deepfakes and bots are just noise.

Both reactions miss the point. The technology itself isn’t the variable that matters. The information environment it lands in is. One early concern about deepfakes, dating to 2018, was the ‘liar’s dividend’: if convincing fabrications exist, real evidence can be dismissed as fake. By 2020, when my colleagues and I at the Stanford Internet Observatory got early access to GPT-3, it was immediately clear how valuable it would be for propagandists: mimicry of diverse linguistic registers, inexpensive persuasive material produced in bulk. In 2022, we started writing about ways to mitigate the risks; others did too. By 2024, AI companies had begun regularly disclosing that adversarial actors were using their tools.

So why are we still having the same conversation, and feeling unprepared?

Because we keep treating this as a technology problem when it’s actually a trust problem—and have done very little to address the underlying conditions that make synthetic media dangerous in the first place.