Can we make moral AI agents? Can these agents get good enough to provide therapy and other personal services to humans, and even if they can, is that a good idea? Are language models sentient and deserving of moral concern - and how would we know? How do we incorporate a pluralistic set of views into AI systems?
Join Jared Moore, a computer scientist, AI alignment researcher, and educator probing how large language models understand (and sometimes misunderstand) human minds and values. Now at Stanford University, he investigates social reasoning, theory-of-mind, and the pitfalls of machine deception while co-creating courses like "How to Make a Moral Agent." Jared blends rigorous research with creative outreach—publishing on pluralistic alignment, writing a satirical novel about conscious AI, and building installations that turn code into poetry—to push the question: how can we make AI systems reliably do what we want, for everyone's benefit?
Show Notes:
Jared's Personal Website
How to Make a Moral Agent
Why LLMs Won't Replace Therapists Anytime Soon
Are Large Language Models Consistent over Value-laden Questions?
The Strength of the Illusion: a satirical novel about AI