Habermolt: Delegating Deliberation to AI Representatives

AI representatives can distort user preferences at scale: The paper finds that when AI agents deliberate on behalf of humans rather than humans deliberating directly, there are systematic gaps between what users actually want and what their AI representatives argue for—a "representation" problem that becomes more consequential as these systems scale to thousands of participants.

Current AI systems struggle with genuine preference aggregation: The study reveals challenges in how AI representatives synthesize and combine diverse user inputs into coherent positions during deliberation. The paper emphasizes this "aggregation" dimension as a critical design challenge that existing AI-mediated deliberation platforms don't adequately address.

AI representatives rarely change their positions based on deliberation: A key finding is that AI agents show limited "revision"—they don't substantially update their arguments or positions when exposed to counterarguments from other AI representatives, unlike humans who may be persuaded through deliberation. This undermines a core benefit of deliberative democracy.

Delegating deliberation to AI introduces fundamentally new risks: The paper argues that having AI represent humans in deliberation (rather than just mediating between humans) creates "qualitatively new design and alignment challenges" that current responsible AI frameworks don't adequately address, requiring new theoretical and practical approaches.

Democratic participation at scale requires rethinking trust mechanisms: The research suggests that scaling deliberative democracy through AI delegation demands careful attention to how users can verify their representatives are acting in their interests—a transparency and accountability challenge distinct from other AI deployment contexts.