Why is this important: Generative AI has great potential for misuse. Fraud and cyberattacks on financial systems spring to mind. But new research suggests the primary category of misuse is to influence political opinion with falsehoods, which has been an issue in past election campaigns and is expected to become even more prevalent this election cycle.
A new study by Google DeepMind has found that AI-generated political content is far more likely to be a misuse of technology than a cyberattack. DeepMind based its conclusion on an analysis of GenAI misuse cases reported between January 2023 and March 2024. In fact, a video about Joe Biden was circulating last year despite being declared a deepfake.
As the election campaign heats up, we're likely to see more examples of these forms of manipulation. The study found that public opinion manipulation was the most common misuse of GenAI features, accounting for 27% of all reported cases. Bad actors can use a variety of tactics to distort the public's perception of political reality, including impersonating public figures, creating fake media, and using synthetic digital personas to mimic grassroots support or opposition (aka astroturfing).
Bad actors can easily manipulate legitimate videos to make a political candidate appear visibly old and unfit to lead. Though more complex, skilled AI artists can create videos from scratch that paint an opponent in a negative light.
The report noted that a less common but emerging trend is the private use of AI-generated media by political candidates and their supporters to build a positive public image. In one example, the Philadelphia sheriff used generative AI to create positive news articles on his campaign website.
Political actors are also using generative AI for targeted political outreach, such as faithfully simulating politicians’ voices to appeal to voters in their native language, or deploying AI-powered campaign robocallers to have customized conversations with voters about important issues.
These tactics may sound familiar, as they have been used in political campaigns long before the advent of generative AI. The difference is that recent rapid advances in AI models have given new power and democratized access to these ancient tactics.
