AI Constitutionalism: Technical Foundations, Legal Questions, and the Future of the Field

• Semester: Spring 2027
• Course ID: 296W
• Credit Hours: 2
• Unique: 29889

Registration Status: Open

Sign in to see meeting information.

Description

This course examines AI constitutions: the public documents that frontier AI developers use to set out the values, priorities, and rules their models are meant to follow. Students learn how a written constitution is turned into model behavior through training, what it can and cannot control, and why developers have moved from rigid rules toward written principles that ask models to exercise judgment. No technical background is assumed. The goal is enough fluency to evaluate these documents as legal and policy instruments.   The course then asks the questions a lawyer or policymaker would ask of any governing text. Who writes it, and by what process? How does it resolve conflicts between competing values? How is it interpreted, amended, and enforced, and by whom? Throughout, students consider a central tension. These documents shape how billions of people get information and advice, yet they are drafted privately and enforced without outside review. That tension leads to questions of process and accountability. Students examine whether deliberative democratic methods could make the drafting of AI constitutions more legitimate. These methods include citizen assemblies, deliberative polls, representative panels, and structured public comment. The course also considers the limits of these methods: how members would be selected, how their conclusions would become technical specifications, and whether a single public process could speak for a global user base. We then consider AI constitutions as public commitments. If a model behaves contrary to its published values, is that a deceptive practice, a breach of contract, or evidence of negligence? And should regulators attach consequences? Finally, we turn to the First Amendment. Government is the most obvious vehicle for public input into AI values, but laws or informal pressure dictating those values raise serious free speech concerns, and they risk turning democratic accountability into political control. The final unit looks ahead to swarm governance. AI is increasingly deployed as many agents working together across different developers and on behalf of different people, often with little human supervision. Individual agents may each follow their own constitutions while the group behaves in ways no one intended. And as humans rely more on AI to oversee other AI, preserving meaningful human authority becomes harder. Students consider whether this future calls for constitutions written for groups, new rules for how agents coordinate, independent oversight institutions, or legal and policy frameworks borrowed from corporate, antitrust, and administrative governance. Assessments include short response papers and a final take-home exam.