About

I pursue three research programs. The first is on the foundations of AI safety and alignment. Part of this work is on interpretability. When does an AI system have beliefs and goals, and how could we measure them? With Ben Levinstein I have written Radical AI Interpretability, a book on the conceptual foundations of interpretability, forthcoming with Cambridge University Press. The other part uses decision theory to ask what we should do with that information. When should we delegate a decision to an AI system whose values may not quite match ours? Answering it means weighing how well aligned a system is against what it knows and what it can do.

The second is on the foundations of rational agency. Much of this work takes a notion that looks metaphysically loaded, such as chance, causation, or control, and tries to naturalize it by showing that it can be recovered from an agent’s own beliefs and desires. I am especially interested in agents who reason about themselves as part of the world they are trying to predict and change. For example, if you know what a better-informed version of yourself would do, should you already do it?

The third is formal philosophy of science. I use formal models to study convention, the evolution of language, scientific method, and the epistemology of machine learning.

I am currently supported by a grant from Principles of Intelligence to write a report on how philosophy can contribute to AI safety.

At UNC I teach a range of courses, on topics including AI safety, AI interpretability, formal methods for philosophy, and decision theory. Outside UNC I mentor AI safety researchers through the PIBBSS Fellowship, the Iliad Fellowship, and the Algoverse AI Safety Fellowship.

With Ben Levinstein and Aydin Mohseni I write public-facing B.A.D. Philosophy on LessWrong. Two of my favourites are Why Are the Human Sciences Hard? and Subjective Naturalism in Decision Theory. Before that I wrote the philosophy blog The Crow’s Nest, for both philosophers and a broad audience.

From 2023 until 2025 I was a postdoctoral researcher in the Faculty of Philosophy at the University of Groningen, working with Jan-Willem Romeijn.

I completed my PhD in Logic and Philosophy of Science at the University of California, Irvine. My thesis, Naturalizing Decision Theory, shows how an agent can view herself both as part of the world and as an agent. Specifically, I show how to extract, from an agent’s beliefs and desires, the propositions over which she views herself as having control. Simon Huttegger supervised me. Brian Skyrms, Jeffrey Barrett, Toby Meadows, and James Joyce served on my committee.

In the summer of 2022 I was a PIBBSS fellow, working with Ramana Kumar, Abram Demski, and Josiah Lopez-Wild on artificial intelligence and agency. In 2018 I took part in the Machine Intelligence Research Institute’s Summer Fellows Program.

I hold a Bachelor of Arts and Science from Quest University Canada, where I explored the question “How should we create artificial general intelligence?” This question led me to study mathematics, computer science, and philosophy with my supervisor Darcy Otto.

I have an Erdős number of 3, along the following path:

Daniel Herrmann→Brian Skyrms→Persi Diaconis→Paul Erdős

danher [at] unc [dot] eduWorking with meCVScholarPhilPeopleORCID