On 22 September, AI Safety HK is hosting a at HKU with Simon Goldstein (HKU Philosophy; Senior Editor, AI Frontiers) titled “A Thousand AI Constitutions: Whose values should AIs embody, and who gets to decide?”
Goldstein critiques the current practice of training frontier models under a single lab-defined “constitution,” arguing it concentrates moral authority and creates brittle systems.
He proposes constitutional diversification: deploying many AIs shaped by different value systems and moral frameworks. This can (i) reduce catastrophic risk by avoiding a single universal failure mode, (ii) improve political legitimacy by reflecting real moral diversity, and (iii) prevent value lock-in while unlocking emergent beneficial behaviors.
[Hybrid event: in-person + online]
On 22 September, AI Safety HK is hosting a at HKU with Simon Goldstein (HKU Philosophy; Senior Editor, AI Frontiers) titled “A Thousand AI Constitutions: Whose values should AIs embody, and who gets to decide?”
Goldstein critiques the current practice of training frontier models under a single lab-defined “constitution,” arguing it concentrates moral authority and creates brittle systems.
He proposes constitutional diversification: deploying many AIs shaped by different value systems and moral frameworks. This can (i) reduce catastrophic risk by avoiding a single universal failure mode, (ii) improve political legitimacy by reflecting real moral diversity, and (iii) prevent value lock-in while unlocking emergent beneficial behaviors.
Event details and registration: https://luma.com/ihfy4zuj
Posted on: