Research

My research asks how a person, institution, or AI system can determine the right thing to do in a particular situation.

The project begins with medical ethics and research ethics. In these fields, ethical guidance must be used in decisions about a particular patient, study, risk, reviewer, or institutional action. I study the mechanisms these fields have developed for making those decisions in institutions shaped by regulation, divided responsibility, uncertainty, and correction.

I use what these mechanisms teach us to address the specification problem in AI alignment. The specification problem asks how to move from a general norm to guidance about what an agent should do in a particular situation.

AI alignment and specification

AI systems designed to act as agents need guidance that identifies what matters in the situation they face and what they should do in response. My work examines what information an AI agent needs, how that information can be organized in a normative world model, and how institutions should divide responsibility between AI systems and human decision-makers.

Medical and research ethics

My work in medical and research ethics studies how ethical guidance becomes usable in practice. These fields have developed mechanisms for deciding who has authority, what evidence matters, how exceptions are handled, and when a decision must move to another process. I examine which features can inform AI alignment and how they need to be adapted for AI agents.

Normative ethics, roles, and conventions

My work on promising, role-based obligations, and interpersonal moral conflict examines how social practices can make obligations accessible and action-guiding. This research provides part of the philosophical foundation for my current work on norms that humans and AI systems can use in institutional settings.

Current projects

Doing the Right Thing

Doing the Right Thing: A Normative World Model for Humans and Machines asks what an agent needs to know to determine what to do in a particular situation. The book develops a normative world model for organizing that guidance and examines what should happen when established guidance runs out.

The Competence Problem of AI Alignment

The Competence Problem of AI Alignment identifies a problem that arises when capable AI systems treat the reasons behind an institutional rule as permission to reopen a legitimate settlement.

Human-in-the-loop systems

When Human-in-the-Loop Amplifies the Risk of Misalignment, forthcoming in Ethics & Human Research, examines the competencies human reviewers need when AI systems are used in research-ethics oversight. It explains how human review can increase risk when reviewers lack the competencies their role requires.

Research in practice

I serve on North Star Institutional Review Board. This service keeps my research connected to the decisions institutions make about particular studies, risks, responsibilities, and forms of oversight.