AI Swarms Threaten Democracy

A Science policy forum paper warned that large-scale AI personas could manipulate online discussions to create seemingly spontaneous consensus.

• If a single operator manages numerous AI accounts, they can infiltrate communities by changing language and tone like humans.
• Messages are adjusted with immediate feedback, optimizing persuasiveness through millions of small-scale experiments.
• UBC's Kevin Leyton-Brown pointed out that deepfakes and fake news networks have already influenced elections in the US, Taiwan, Indonesia, and India.
• The massive content from pro-Russia networks also raises the risk of contaminating training data for future AI models.
• Researchers said declining trust in unfamiliar accounts could work to the advantage of celebrities and existing influencers, and the next election could become a real-world test bed for this technology.