Epistemic benefits of language model agents in democratic decision-making

Show whether language model agents can deliver the epistemic benefits associated with democratic procedures, specifically the pooling of normally distributed knowledge and the forced interaction of diverse knowledge sources to surface additional reasons and proposals.

Background

The authors discuss two key epistemic benefits of democratic processes: pooling dispersed knowledge and inducing interactions among diverse knowledge sources, which together contribute to superior collective decisions.

While LLM agents might theoretically provide these benefits, the authors explicitly note that it remains to be demonstrated. Establishing this is essential for evaluating whether delegating participation to AI agents can sustain democracy’s epistemic virtues.