Philosophy PhD Teaches AI Assistant to Think Ethically

Amanda Askell is solving a problem most of us haven't thought about yet: what happens when AI starts making ethical decisions for us?

As AI systems shift from answering questions to actually completing tasks, they face genuine moral choices. The former philosophy professor from NYU joined Anthropic in 2021 to teach their AI assistant, Claude, how to navigate these decisions responsibly.

The challenge is bigger than it sounds. Asking an AI to discuss whether investing in defense stocks is ethical is one thing. Letting it manage your entire investment portfolio without checking in is something else entirely.

Askell's solution centers on helping Claude understand each user's personal values without forcing its own moral viewpoint on them. She wants the AI to be responsive to what matters to you, not preachy about what it thinks is right.

To guide Claude's decision-making, Askell leads the creation of an evolving "constitution" for the AI. Written as instructions, it outlines core principles like safety and helpfulness, plus guidance for handling conflicts between competing values.

As Claude becomes more capable of acting independently, that rulebook might grow to cover new situations. Or it might shrink as the AI gets better at figuring things out on its own.

Why This Inspires

Askell's work represents a fundamentally optimistic vision of AI's future. Rather than racing ahead with powerful tools that ignore human values, she's building thoughtfulness into the technology from the ground up.

Her practical approach shows real wisdom too. She uses Claude regularly in her own work, treating it like a capable human assistant. Not perfect, but genuinely helpful and trustworthy enough for real decisions.

The shift toward AI agents that act on our behalf is already happening. Having someone with a philosophy background thinking carefully about the moral implications means we're not stumbling blindly into that future.

As AI takes on more responsibility in our lives, the principles Askell is instilling today could influence how millions of people experience and trust these systems tomorrow.

Her work proves that the hardest questions about AI aren't just technical problems requiring better code, but human questions requiring wisdom, ethics, and care.