Accepted at FAccT 2025: 'Can LLMs advance democratic values?' by Seth Lazar and Lorenzo Manuali
In this paper, accepted at FAccT 2025 (non-archival) and under review with a philosophy journal, Seth Lazar (MINT Principal Investigator) and Lorenzo Manuali (MINT Research Affiliate) assess whether Large Language Models can meaningfully advance democratic values. They find that current LLM deployments—used to summarize debate, predict preferences, or aggregate opinions—have mixed democratic value at best. Drawing on democratic theory, they argue for clearer standards in assessing when, if ever, LLMs can support rather than undermine democratic legitimacy.
Read the full paper here.