Summary
Highlights
Daniel Kokotajlo highlights the industry's secret fear: creating a new species that could lead to human extinction. He explains why he left OpenAI, citing a lack of transparency and commitment to safe AI development.
Kokotajlo defines superintelligence as systems smarter than humans in every aspect. He forecasts a 50% chance of reaching this level by 2029-2030, emphasizing that current industry trends suggest a rapid, potentially uncontrollable trajectory.
The discussion shifts to how AI labs are driven by power-seeking incentives, with CEOs racing to reach AGI first to avoid others becoming a dictator. Kokotajlo reveals that safety narratives are often secondary to commercial and strategic competition.
Kokotajlo explains his decision to forfeit $2 million in equity by refusing to sign an anti-disparagement agreement upon leaving OpenAI, which he saw as a move to silence dissent about safety risks.
Kokotajlo details his 'AI 2027' research, which predicts a sequence of automation—coding, research, and deployment—that leads to superintelligence. He introduces 'Plan A' (AI 2040), a proposal for regulated, transparent development.
As AI and robotics eventually take over labor, Kokotajlo suggests that a 'citizens dividend' will be necessary to ensure social stability and public access to the immense wealth generated by AI systems.
Kokotajlo urges the public to take AI seriously, stay informed, and advocate for policy. He concludes that while the situation is dire, human action and international regulation still provide a path toward a better outcome.