Doom Debates Former MIRI Researcher Solving AI Alignment by Engineering Smarter Human Babies
25 snips
Oct 25, 2025 Tsvi Benson-Tilsen, a former MIRI researcher and co-founder of the Berkeley Genomics Project, advocates for engineering smarter humans as a solution to AI alignment challenges. He discusses the alarming P(doom) estimates and the urgent need to slow AGI development. Delving into human germline engineering, Tsvi shares insights on chromosome selection and its potential to enhance intelligence significantly. He also debates societal stigma around AGI research and outlines an ambitious timeline for creating genetically enhanced humans to tackle the impending AI risks.
AI Snips
Chapters
Transcript
Episode notes
Alignment Progress Is Essentially Zero
- Tsvi judges current alignment progress at essentially 0% and thinks creating AGI without solving alignment has >90% chance to kill everyone.
- He estimates ~20% P(doom) by 2050 and believes timelines are measured in decades, not years.
Pause AGI And Build Social Taboos
- Pause AGI development and stigmatize gain-of-function AGI research like dangerous virology work.
- Use protests, public awareness, and governmental measures to slow capability growth now.
Treat AGI Research Like Gain-Of-Function Biology
- Tsvi proposes an international treaty plus a social regime to ban conceptual AGI research, comparing it to gain-of-function virus work.
- He emphasizes academics should avoid researching core algorithms of intelligence.