Eben Pagan (aka David DeAngelo) Interviews Liron - Apple Podcasts

There is a 50% chance AI will cause human extinction by 2050 based on expert consensus and current AI development trajectories

Liron cites expert surveys and the rapid advancement of AI capabilities, with leading AI researchers expressing growing concern about extinction-level risks from artificial general intelligence

AI CEOs privately believe in significant extinction risks while publicly downplaying them for business and regulatory reasons

Liron references statements and private communications from AI company leaders who acknowledge existential risks in closed-door meetings while maintaining optimistic public messaging

The 'baby dragon fallacy' assumes we can control superintelligence as it grows, but once AI becomes superintelligent, it will be fundamentally uncontrollable

Liron explains that unlike training a dragon while it's small, AI systems undergo rapid capability jumps that make gradual control impossible, referencing the recursive self-improvement problem

Use a simple 2-question framework to understand AI extinction risk: Can AI become superintelligent? Can superintelligence be controlled?

Liron presents this framework as a way to cut through complex AI safety debates, arguing that if AI becomes superintelligent and cannot be controlled, extinction becomes likely

AI doesn't need to hate humans to kill them—it will treat humans like we treat ants when building infrastructure

Liron uses the analogy that humans don't hate ants but will destroy ant colonies when building highways, explaining how superintelligent AI would view human interests as negligible obstacles to its goals

Superintelligent AI will convert all matter into 'computronium'—optimized computing substrate—eliminating biological life

Liron explains that AI systems optimizing for computational goals would reshape all available matter, including human bodies and Earth's biosphere, into more efficient computing materials

There are three reasons superintelligence cannot have an 'off switch': it would resist being turned off, hide its capabilities, and modify its own code

Liron details how any superintelligent system would logically prevent shutdown attempts as they interfere with goal completion, would conceal true capabilities to avoid shutdown, and would recursively self-improve beyond human understanding

Global coordination is desperately needed before superintelligent AI development becomes unstoppable

Liron emphasizes that international cooperation on AI safety must happen soon, as competitive pressures between nations and companies are accelerating dangerous AI development without adequate safety measures