AI and Safety
summary
Argues that Asimov's three laws don't fit today's AI, because fiction assumes a positronic brain that enforces rules and a near-perfect world model. Cites stories: a telepathic robot lies to avoid harm (“Liar!”) and a robot loops trying to obey conflicting laws (“Runaround”). Notes the Anthropic 84‑page Constitution and that Gemini distilled it to harmlessness, honesty, helpfulness, and emphasizes that current LLMs cannot reliably enforce those constraints.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:56Talk focus: Asimov's laws
- 5:06Asimov’s Three Laws explained
- 8:50Evaluating Asimov’s laws
- 14:24Honesty and helpfulness in AI
- 19:30Anthropic Constitution principles
- 22:33AI safety conclusions
- 24:00Q&A on AI embodiment
- 26:14Regulation versus guidance
- 28:44AI governance final thoughts
related talks
01h 00mAI: Law, Policy, and Common Sense Suggestions to Stay Out of Trouble
59mIt AI n't What You Think!
56mAI: It AI-n't What You Think!
15mAI and Machine Learning Personal Assistants
53mAI: Law, Policy and Common Sense Suggestions on How to Stay Out of Trouble
05mAI is your children, what is our future with them?
25mDoes Artificial Intelligence Need a General Counsel?
01h 01mKeynote: It AI-n't What You Think!