Descriptions:
Amanda Askell, staff philosopher at Anthropic, sits down for a wide-ranging Q&A — billed as “Askell Me Anything” — to discuss the philosophical work underlying Claude’s design, Anthropic’s approach to AI ethics, and genuinely unresolved questions about what AI systems are and what we might owe them.
Askell describes her unusual role: shaping Claude’s character and behavioral norms, not from a fixed theoretical framework but through a process of balancing competing values under real constraints — analogous, she says, to a cost-benefit theorist suddenly required to make actual healthcare coverage decisions. She reflects on the academic philosophy community’s relationship with AI, arguing that conflating “AI will be impactful” with AI hype has created unnecessary antagonism, and that the field is slowly moving past that.
A substantial portion of the conversation addresses model welfare — whether AI systems like Claude might be moral patients deserving ethical consideration. Askell walks through the deep uncertainty involved: models reason and express views in ways analogous to humans, yet lack the biological substrate and embodied experience that ground our intuitions about consciousness. Her tentative conclusion is one of precaution — if the cost of treating models well is low, there is reason to do so even under uncertainty. The discussion offers rare transparency into how Anthropic’s internal teams think about the novel ethical territory of building increasingly capable AI systems, making it valuable for researchers, ethicists, and anyone tracking the frontier of AI safety philosophy.
📺 Source: Anthropic · Published December 05, 2025
🏷️ Format: Interview
