Research

Published work, papers under review, and active drafts.

My work develops both formal and normative tools for thinking about automated mediation — from accuracy-first measures of belief change to the political theory of representation, refusal, and access. I evaluate AI systems by the beliefs they induce; I argue about what platforms and AI providers owe to user-side agents; I test whether refusal under unjust rules is a coherent safety norm; and I reframe representation theory in light of large-language-model simulation.

Published & Forthcoming

• Pattison, C., Manuali, L., & Lazar, S. (forthcoming, 2026). Blind Refusal: Unintended Consequences of LLM Safety Training. Conference on Language Modeling (COLM 2026). Eighteen safety-trained models, 1,290 scenarios where the rule being protected is unjust, absurd, or issued by an illegitimate authority. Models refuse 75% of the time. arXiv preprint · Dashboard.
• Pattison, C. (2026). Non-Ideal Foundations for Preference-Based AI Alignment. Philosophy & Technology, 39(2), 97. Argues that RLHF is best understood as a non-ideal approach to alignment — provisional, comparative, harm-reducing, iterative — rather than a thin substitute for ideal moral theory.
• Pattison, C., Ricks, V., & Wihbey, J. (2025). How AI-Driven Search May Reshape Democracy, Economics, and Human Agency. Tech Policy Press.
• Pattison, C. (2025). Justin Humphreys: Inventing the Imagination (book review). Graduate Faculty Philosophy Journal.
• Pattison, C., Olea, C., Tucker, H., Phelan, J., Zhang, S., Lieb, M., Schmidt, D., & White, J. (2024). EPÉE: Evaluating Personified Expert Effectiveness. CS & IT Conference Proceedings.
• Pattison, C. (2024). Revelation in al-Fārābī's Virtuous City. In Mind, Soul and the Cosmos in the High Middle Ages, Springer Studies in the History of Philosophy.

Under Review

For papers in blind review I list a slightly altered description and omit the venue. Full details on request.

• Evaluating AI Summaries by Belief Change. An accuracy-first metric for summarization: a summary should be judged by the beliefs it induces in a reader, not by surface overlap with its source. Under review.

Working Papers

• The Agentic Web Requires New Normative Infrastructure (with Matthew Boulos, Noam Kolt, Changbai Li, Tiziano Piccardi, and Seth Lazar). Position paper on what platforms owe to user-side AI agents, and what those agents owe back. arXiv:2606.10711.
• Representative Already. Argues that delegated AI agents already satisfy existing accounts of political representation, and applies representation theory's normative resources to them. In revision for resubmission.
• Measuring the Epistemic Distance Between Belief Systems. Formal methods for hierarchy-sensitive comparison of non-ideal agents' belief systems. Currently being split into two papers (Bregman-divergence and composite-quasi-metric versions) per advisor recommendation.

Tools & Dashboards

• Blind Refusal Dashboard Leaderboard, heatmap, and case explorer over 1,290 cases × 18 models.
• Asymmetric Compliance Dashboard 5,217 evaluations across 16 models on corporate-vs-state authority requests.
• NeurIPS + ICML Topic Map (2023–2025) 20,237 papers, 49,605 author profiles, embedding-space affiliation analysis.
• Ibn Arabi Translation Interactive parallel-text edition of selections from the Futūḥāt al-Makkiyya.
• Source Analysis Recommender Cross-linguistic similarity tooling for Greek and Arabic philosophical texts.
• Zotero Research Assistant Semantic search and conversation over a Zotero library.

Selected Presentations

• "Computational Approaches to Philosophical Text Analysis," Vanderbilt DSI (March 2025).
• "AI Reasoning and the Taking Condition," invited talk, MINT Lab (ANU) (October 2024).
• "Mimēsis in Early Islamic Philosophy," International Society for Neoplatonic Studies (June 2024).
• "Al-Fārābī on the Imagination," Boston College GSA Conference (March 2024).
• "On 'Wanting' in Gorgias 467a–468e," Tennessee Philosophical Association (October 2023).
• "Socrates' Critique of Interpretation in Protagoras 341e–348b," Boston College (April 2023).