Scientists warn AI swarms mimic real users to manipulate public opinion, posing risks to democracy. Learn about synthetic consensus and detection methods.
© RusPhotoBank
An international team of scientists has raised the alarm about a new threat: so-called "AI swarms" that can create the illusion of widespread public opinion. In a Science journal publication, the researchers note that combining large language models with multi-agent systems enables coordinated networks of AI personas. These personas act like real users and can subtly influence digital environments.
Unlike the primitive bots of the past, modern AI agents can maintain consistent "personalities," remember previous interactions, and adapt their communication style based on audience reactions. Such digital profiles can operate in sync, promoting the same agenda while mimicking independent viewpoints. This creates a "synthetic consensus" effect, where artificially generated activity is perceived as broad public support for a particular position.
The researchers argue these mechanisms pose a serious threat to democratic processes and public dialogue. A single malicious actor using AI can impersonate thousands of voices, influencing trends, pressuring public figures, and spreading disinformation. Moreover, mass production of artificial content can distort training data for other AI systems, amplifying manipulation effects.
Experts emphasize that traditional moderation of individual posts is ineffective against such networks. Instead, they call for detecting statistically anomalous coordination, tracking content origins, and applying behavioral science methods to analyze collective AI agent actions. Proposed measures include implementing privacy-preserving verification mechanisms, creating a distributed system for monitoring AI influence, and reducing financial incentives for inauthentic online activity.