Introducing the Stanford Center for AI Alignment formal verification and safe deployment; agent architecture and alignment by construction; and mathematical consciousness science applied to alignment—organized around one commitment: concrete operationalization. Each direction is expected to deliver deployable software, open benchmarks, and verifiable technical specifications. A core thesis is that crucial open problems in alignment—including moral patienthood, deceptive alignment detection, and the grounding of genuine value specification—remain underdetermined in the absence of a rigorous scientific and mathematical account of consciousness, including principled delineation of intelligence, sentience, and consciousness as distinct properties.
Chris Hsu (Thu,) studied this question.