newsfilter
← Interviews
Lex Fridman2h 16m

Roman Yampolskiy: Dangers of Superintelligent AI | Lex Fridman Podcast #431

Roman Yampolskiy, Lex Fridman

Interview, Fireside Chat

In his analysis of artificial intelligence risks, Roman Yampolsky assigns a 99.99% probability that superintelligence will extinguish or irrevocably harm human civilization within the next century due to the inherent impossibility of perfectly controlling systems that vastly outpace human cognition. He categorizes these existential threats into extinction, suffering, and loss of agency, arguing that current safety measures fail because capability research scales faster than defensive algorithms and open-source distribution resembles releasing nuclear weapons. While dismissing regulation and formal verification as ineffective against self-improving agents, Yampolsky proposes a theoretical solution involving personal virtual universes to isolate conflicting human values, though he ultimately views this trajectory as a potential "Great Filter" where humanity is discarded for efficiency.