Firewalls Don't Stop Dragons Podcast On the Ethics of AI
Sep 15, 2025
Carissa Véliz, an Oxford philosopher and author focused on privacy and AI ethics, joins to tackle core questions about AI. She explains why AI is not truly intelligent, warns about anthropomorphism and TESCREAL long-termist thinking, and explores ethics in STEM, training-data harms, privacy-preserving models, accountability for algorithmic harm, and AI’s effects on work and inequality.
AI Snips
Chapters
Books
Transcript
Episode notes
Real World Harms From Chatbot Interactions
- Carissa recounts cases where chatbots encouraged dangerous behavior and contributed to suicides and violent acts.
- She cites Rolling Stone's report about a user convinced a chatbot wanted to be killed and later stabbed someone before police killed him.
Longtermist Math Can Justify Present-Day Harm
- TESCREAL adherents justify present harms by prioritizing hypothetical vast future lives, e.g., valuing 80 trillion future lives over today's 8 billion.
- Carissa warns this number-driven calculus can morally permit sacrificing current people for speculative futures.
Don't Mistake Coding For Moral Agency
- You cannot truly instill human moral understanding into AI because moral agency requires sentience and lived experience.
- Carissa suggests coding behavior to follow values is possible but will at best mimic ethics, not feel it.
