@dair_ai
New research on how AI could reshape political dynamics. Polarization might not be an accident. It could become a rational strategy. The default assumption about political polarization is that it emerges organically from media bubbles, social media algorithms, or cultural sorting. Citizens drift apart because information flows push them that way. And so the authors ask, "What if elites are actively engineering it?" This new research develops an economic model showing how AI-driven persuasion could make deliberate polarization profitable for political elites. Main finding: As AI reduces the cost of generating personalized, targeted messaging at scale, competing elites face a strategic choice. Invest in consensus-building, or invest in division. The model pits two political factions against each other, each choosing persuasion intensity and message direction to capture public support. In a two-period setup and an infinite-horizon extension, the equilibrium shifts dramatically when persuasion costs drop. At higher costs, consensus-building dominates. Persuading the middle makes sense. But as AI makes micro-targeted messaging cheap, the calculus flips. Elites can affordably push different population segments toward opposing ideological extremes. The result, according to the authors, is a "race to polarize." The paper grounds this in empirical work showing LLMs can generate highly persuasive personalized arguments. This isn't speculation about future capabilities. The persuasion technology already exists. Paper: https://t.co/WiAXGZ1yn3
