•  50
    Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback
    with Vincent Conitzer, Rachel Freedman, Jobst Heitzig, Wesley H. Holliday, Bob M. Jacobs, Nathan Lambert, Eric Pacuit, Stuart Russell, Hailey Schoelkopf, Emanuel Tewolde, and William S. Zwicker
    Proceedings of the Forty-First International Conference on Machine Learning. forthcoming.
    Foundation models such as GPT-4 are fine-tuned to avoid unsafe or otherwise problematic behavior, such as helping to commit crimes or producing racist text. One approach to fine-tuning, called reinforcement learning from human feedback, learns from humans' expressed preferences over multiple outputs. Another approach is constitutional AI, in which the input from humans is a list of high-level principles. But how do we deal with potentially diverging input from humans? How can we aggregate the in…Read more
  •  73
    Probing the quantitative–qualitative divide in probabilistic reasoning
    with Duligur Ibeling, Thomas Icard, and Krzysztof Mierzewski
    Annals of Pure and Applied Logic 175 (9): 103339. 2024.
  •  399
    Is Causal Reasoning Harder Than Probabilistic Reasoning?
    with Duligur Ibeling and Thomas Icard
    Review of Symbolic Logic 17 (1): 106-131. 2024.
    Many tasks in statistical and causal inference can be construed as problems of entailment in a suitable formal language. We ask whether those problems are more difficult, from a computational perspective, for causal probabilistic languages than for pure probabilistic (or “associational”) languages. Despite several senses in which causal reasoning is indeed more complex—both expressively and inferentially—we show that causal entailment (or satisfiability) problems can be systematically and robust…Read more