About

Thomas Hofweber is a professor in the Department of Philosophy at the University of North Carolina at Chapel Hill, holding the title of William R. Kenan Jr. Distinguished Professor and serving as the Director of Graduate Admissions. He specializes in metaphysics, the philosophy of language, the philosophy of mathematics, and the foundations of artificial intelligence. Hofweber is the author of 'Idealism and the Harmony of Thought and Reality' (Oxford University Press, 2023), which defends idealism, and his first book 'Ontology and the Ambitions of Metaphysics' was published by Oxford University Press in 2016. He has also co-edited a collection titled 'Empty Names, Fiction, and the Puzzles of Non-Existence' with Anthony Everett. Hofweber is the Director of the AI Project and holds a secondary appointment in the School of Data Science and Society. His research includes a wide range of topics within metaphysics and philosophy of language, and he has published extensively on issues such as ontology, logic, and the philosophy of mathematics.

Research topics

• Epistemology
• Philosophy
• Computer Science
• Humanities
• Linguistics
• Mathematics
• Law
• Medicine
• Art
• Chemistry
• Literature

Selected publications

• Will AI Produce Works of Extraordinary Aesthetic Value?

  Journal of Aesthetics and Art Criticism · 2026-02-17

  article1st authorCorresponding

  Abstract It is tempting to think that AI will never produce any works of art that are of significant aesthetic value, let alone ones comparable to or even surpassing the greatest works produced by humans. However, it is not so clear what precisely the reason is for this limitation of AI, which has already surpassed human performance in other areas. I discuss several possible in-principle limitations to the aesthetic value of AI-generated works and conclude that although some limits are legitimate, the overall negative stance is nonetheless unjustified. But the purely techno-optimist stance that AI will be better than humans at everything is also unjustified. The question remains how one might build an AI that specifically aims to produce works of great aesthetic value. I outline how one might do this and argue that a particular feature of aesthetic value allows for the application of reinforcement learning in the training of such an AI. Finally, I discuss some obstacles for this approach and outline what we might expect for the future of art.

  PublisherDOI

• AI-Mediated Explainable Regulation for Justice

  arXiv (Cornell University) · 2026-03-31

  preprintOpen access1st authorCorresponding

  Present practice of deciding on regulation faces numerous problems that make adopted regulations static, unexplained, unduly influenced by powerful interest groups, and stained with a perception of illegitimacy. These well-known problems with the regulatory process can lead to injustice and have substantial negative effects on society and democracy. We discuss a new approach that utilizes distributed artificial intelligence (AI) to make a regulatory recommendation that is explainable and adaptable by design. We outline the main components of a system that can implement this approach and show how it would resolve the problems with the present regulatory system. This approach models and reasons about stakeholder preferences with separate preference models, while it aggregates these preferences in a value sensitive way. Such recommendations can be updated due to changes in facts or in values and are inherently explainable. We suggest how stakeholders can make their preferences known to the system and how they can verify whether they were properly considered in the regulatory decision. The resulting system promises to support regulatory justice, legitimacy, and compliance.

  PublisherDOI

• Will AI Produce Works of Extraordinary Aesthetic Value?

  UNC Libraries · 2026-04-23

  articleOpen access1st authorCorresponding

  It is tempting to think that AI will never produce any works of art that are of significant aesthetic value, let alone ones comparable to or even surpassing the greatest works produced by humans. However, it is not so clear what precisely the reason is for this limitation of AI, which has already surpassed human performance in other areas. I discuss several possible in-principle limitations to the aesthetic value of AI-generated works and conclude that although some limits are legitimate, the overall negative stance is nonetheless unjustified. But the purely techno-optimist stance that AI will be better than humans at everything is also unjustified. The question remains how one might build an AI that specifically aims to produce works of great aesthetic value. I outline how one might do this and argue that a particular feature of aesthetic value allows for the application of reinforcement learning in the training of such an AI. Finally, I discuss some obstacles for this approach and outline what we might expect for the future of art.

  PublisherDOI

• AI-Mediated Explainable Regulation for Justice

  ArXiv.org · 2026-03-31

  articleOpen access1st authorCorresponding

  Present practice of deciding on regulation faces numerous problems that make adopted regulations static, unexplained, unduly influenced by powerful interest groups, and stained with a perception of illegitimacy. These well-known problems with the regulatory process can lead to injustice and have substantial negative effects on society and democracy. We discuss a new approach that utilizes distributed artificial intelligence (AI) to make a regulatory recommendation that is explainable and adaptable by design. We outline the main components of a system that can implement this approach and show how it would resolve the problems with the present regulatory system. This approach models and reasons about stakeholder preferences with separate preference models, while it aggregates these preferences in a value sensitive way. Such recommendations can be updated due to changes in facts or in values and are inherently explainable. We suggest how stakeholders can make their preferences known to the system and how they can verify whether they were properly considered in the regulatory decision. The resulting system promises to support regulatory justice, legitimacy, and compliance.

  PublisherOA PDF

• Two problems for Millian phenomenalism

  Asian Journal of Philosophy · 2025-03-01

  article1st authorCorresponding
  PublisherDOI

• Fundamental Ontology and Esoteric Metaphysics: How to settle the Question

  Revista Portuguesa de Filosofia · 2024-12-09

  article1st authorCorresponding

  How can we settle whether key metaphysical questions should properly be stated by relying on a substantial notion of metaphysical priority, like grounding or being metaphysically more fundamental than? Relatedly, how can we settle whether ontology should properly be seen as the disciple that studies either what there is or else only what there is fundamentally? Which way of thinking about ontology brings out its proper metaphysical significance? One challenge to giving notions like grounding or fundamentality key roles in metaphysics is that these notions are insufficiently clear, and that metaphysics tied to them turns into esoteric metaphysics. To make progress on these issues I propose a particular challenge—the cognitive function challenge—that needs to be met for metaphysics based on a substantial notion of priority not to turn into objectionable esoteric metaphysics. I also outline some reasons why other approaches that aim to establish such notions as legitimate for metaphysics fall short and how the cognitive function challenge might be met.

  PublisherDOI

• Fundamental Problems With Model Editing: How Should Rational Belief Revision Work in LLMs?

  arXiv (Cornell University) · 2024-06-27 · 1 citations

  preprintOpen access

  The model editing problem concerns how language models should learn new facts about the world over time. While empirical research on model editing has drawn widespread attention, the conceptual foundations of model editing remain shaky -- perhaps unsurprisingly, since model editing is essentially belief revision, a storied problem in philosophy that has eluded succinct solutions for decades. Model editing nonetheless demands a solution, since we need to be able to control the knowledge within language models. With this goal in mind, this paper critiques the standard formulation of the model editing problem and proposes a formal testbed for model editing research. We first describe 12 open problems with model editing, based on challenges with (1) defining the problem, (2) developing benchmarks, and (3) assuming LLMs have editable beliefs in the first place. Many of these challenges are extremely difficult to address, e.g. determining far-reaching consequences of edits, labeling probabilistic entailments between facts, and updating beliefs of agent simulators. Next, we introduce a semi-synthetic dataset for model editing based on Wikidata, where we can evaluate edits against labels given by an idealized Bayesian agent. This enables us to say exactly how belief revision in language models falls short of a desirable epistemic standard. We encourage further research exploring settings where such a gold standard can be compared against. Our code is publicly available at: https://github.com/peterbhase/LLM-belief-revision

  PublisherOA PDFDOI

• Machine Learning in Health Care: Ethical Considerations Tied to Privacy, Interpretability, and Bias

  North Carolina Medical Journal · 2024-07-11 · 7 citations

  reviewOpen access1st authorCorresponding

  Machine learning models hold great promise with medical applications, but also give rise to a series of ethical challenges. In this survey we focus on training data, model interpretability and bias and the related issues tied to privacy, autonomy, and health equity.

  PublisherOA PDFDOI

• Machine Learning in Health Care: Ethical Considerations Tied to Privacy, Interpretability, and Bias

  UNC Libraries · 2024-10-05

  articleOpen accessSenior author

  Machine learning models hold great promise with medical applications, but also give rise to a series of ethical challenges. In this survey we focus on training data, model interpretability and bias and the related issues tied to privacy, autonomy, and health equity.

  PublisherDOI

• Are language models rational? The case of coherence norms and belief revision

  arXiv (Cornell University) · 2024-06-05

  preprintOpen access1st authorCorresponding

  Do norms of rationality apply to machine learning models, in particular language models? In this paper we investigate this question by focusing on a special subset of rational norms: coherence norms. We consider both logical coherence norms as well as coherence norms tied to the strength of belief. To make sense of the latter, we introduce the Minimal Assent Connection (MAC) and propose a new account of credence, which captures the strength of belief in language models. This proposal uniformly assigns strength of belief simply on the basis of model internal next token probabilities. We argue that rational norms tied to coherence do apply to some language models, but not to others. This issue is significant since rationality is closely tied to predicting and explaining behavior, and thus it is connected to considerations about AI safety and alignment, as well as understanding model behavior more generally.

  PublisherOA PDFDOI

Frequent coauthors

• Marc Lange

  5 shared

• J. David Velleman

  4 shared

• Javier Cumpa

  Universidad Complutense de Madrid

  3 shared

• Rebecca L. Walker

  2 shared

• Peter Hase

  2 shared

• Elias Stengel-Eskin

  2 shared

• Mohit Bansal

  2 shared

• Amie L. Thomasson

  Dartmouth Hospital

  1 shared

Education

• PhD, Philosophy

  Stanford University

• MA, Philosophy

  Ludwig-Maximilians-Universität München

Awards & honors

• Philosopher’s Annual (2001)
• Oxford Studies Younger Metaphysician Prize
• APA Article Prize
• Philosopher’s Annual (2008)