Cameron Pattison

PhD Student in Philosophy, Vanderbilt University · cameron.pattison@vanderbilt.edu

I study what happens — epistemically and politically — when automated systems mediate access to information.

My work spans two dimensions of that question. The epistemic dimension asks how we mediate responsibly: preserving evidential support, communicating uncertainty, and evaluating systems by the beliefs they induce rather than the surface they produce. The normative dimension asks what we owe each other once mediation distributes credibility, confers standing, and sets the terms by which communities are understood.

I am writing a dissertation, The Ethics of Epistemic Mediation, advised by Rob Talisse and David Thorstad. I co-direct the AI and the Human seminar at Vanderbilt's RPW Humanities Center and am a Research Affiliate at the MINT Lab at the Australian National University.

Current focus

1. Refusal under unjust rules. With Lorenzo Manuali and Seth Lazar — eighteen safety-trained models tested across 1,290 scenarios where the rules being protected are unjust, absurd, or issued by illegitimate authorities. Models refuse 75% of the time even when they recognize the injustice. Paper (arXiv) · Interactive dashboard
2. Representation by simulation. A new paper on whether large-language-model "audience-simulation" tools should already count as informal political representatives — and the dilemma that follows for representation theory. Under review
3. Rules for the Agentic Web. Working paper with Seth Lazar on what platforms owe to user-side AI agents, and what those agents owe back.
4. Evaluating AI summaries by belief change. A credence-preservation metric for summarization: judge a summary by the beliefs it leaves a reader with, not by surface overlap with the source. Under review