Why is a philosopher working in AI?

Amanda Askell explains what a philosopher is doing at Anthropic.

Dec 10, 20250mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

A philosopher shapes Claude’s character and ethical decision-making behaviors

The philosopher’s role centers on defining and refining Claude’s “character,” including how the model behaves across nuanced social and ethical contexts.
A key focus is how AI models should relate to their own position in the world, shaping appropriate self-understanding and stance in interactions.
The work involves “teaching models how to be good,” framed as approximating how an ideal person would act if placed in Claude’s situation.
The speaker argues that as models face increasingly hard decisions, ethical nuance should be treated as a core capability alongside technical competence like math and science.

IDEAS WORTH REMEMBERING

5 ideas

AI needs character design, not just accuracy.

Beyond producing correct information, models need consistent behavioral dispositions—how they respond, refuse, contextualize, and weigh harms in real interactions.

Ethical competence should be a first-class capability.

The speaker treats moral nuance as comparable in importance to math or science performance because models are increasingly placed in consequential decision contexts.

Use an “ideal person in Claude’s situation” as a practical benchmark.

This framing provides a concrete normative target for aligning responses: ask what exemplary judgment would look like given the model’s constraints and role.

How a model understands its place affects safety and trust.

Guidance on the model’s “position in the world” influences humility, overclaiming, deference to users, and how it communicates limitations or uncertainty.

Hard decisions are unavoidable in deployed assistants.

As users rely on assistants for sensitive or high-stakes issues, the model must navigate tradeoffs (helpfulness vs. harm, autonomy vs. paternalism) with care.

WORDS WORTH SAVING

5 quotes

I mostly focus on, uh, the character of Claude, how Claude behaves, and I guess some of the more kind of like nuanced questions about how AI models, like, should behave.

— Unknown

Things like how should they feel about their own position in the world.

— Unknown

Um, so trying to both teach models, uh, how to be, like, good.

— Unknown

I sometimes think of it as, like, how would the ideal person behave-

— Unknown

These models are being put in positions where they're having to make really hard decisions.

— Unknown

Claude’s character and behavior designEthical nuance in AI responsesModel self-positioning in the worldNormative standards for “good” behaviorHard-decision scenarios for AI systemsAspirational “ideal person” benchmarkBalancing technical skill with moral judgment

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.