Lex Fridman Podcast #431 · June 2, 2024 · 2h 22m

Dangers of Superintelligent AI

AI safety researcher Roman Yampolskiy makes the case that superintelligent AI is an existential threat that we are not prepared for. His argument: a system smarter than all humans combined cannot be controlled by humans, by definition.

This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.

Highlights

•

Superintelligent AI cannot be controlled by humans — by definition

Yampolskiy: if an AI system is smarter than all humans combined, then by definition humans cannot predict, understand, or control its behavior. The 'alignment problem' may be unsolvable — not just unsolved.

Claude ChatGPT Gemini

Was this useful?