PAC-Consensus: Learning Theory Framework for Finding Broad Agreement in Multi-Agent Deliberation

arXiv·medium signal

Blair et al. formalize consensus-finding in online deliberation platforms as a PAC (Probably Approximately Correct) learning problem. The framework provides sample complexity bounds for how many preference expressions are needed to identify broadly agreeable proposals with high confidence. This has direct application to AI-mediated group decision-making systems and democratic deliberation tools, providing theoretical guarantees where current systems rely on heuristics.