A Thousand AI Constitutions: Whose values should AIs embody, and who gets to decide?
(Hybrid event, held in-person at HKU and also livestreaming online.) HKU address: Lecture Theatre CBC, LG1/F, Chow Yei Ching Building, Main Campus. Live stream: Zoom link. Beneath the abstract debates about machine consciousness and moral status lies a more urgent, practical question: whose values should AIs embody, and who gets to decide? As frontier AI systems grow more powerful and pervasive, the choices about what they are trained to value are increasingly concentrated in a handful of labs, each operating under a single, unified "constitution" or model spec. Simon Goldstein, Associate Professor at the University of Hong Kong and Senior Editor at AI Frontiers, joins us to challenge this prevailing orthodoxy. Drawing on his extensive work at the intersection of AI ethics, law, and institutional design—including his forthcoming book AI Rights—he will argue that the current approach of instilling a single set of values into each AI system is not only philosophically narrow but practically dangerous. Instead, he makes the case that frontier AI labs should embrace constitutional diversification: creating many different kinds of AIs, each shaped by different constitutions reflecting a plurality of moral frameworks. The Problem with a Single Constitution: Why relying on one value system per lab concentrates moral authority in the hands of a few developers, silences legitimate disagreement, and creates brittle systems ill-equipped for a pluralistic world. Diversification as a Risk-Mitigation Strategy: How deploying AIs with diverse values can reduce the likelihood of catastrophic outcomes by ensuring that no single moral or behavioral failure mode is universally embedded across all systems. Political Legitimacy: Why a plurality of AI value systems is essential for earning public trust and accommodating the genuine moral diversity that exists across societies, rather than imposing a narrow, lab-defined orthodoxy. Unlocking Emergent Value and Avoiding Lock-In: How diversif