Dr. Roman Yampolskiy — A computer scientist and associate professor who coined the term 'AI safety' and has researched the control problem for roughly 15 years. He is a globally recognized voice warning that superintelligence cannot be made safe.
The gist
Yampolskiy tells Steven Bartlett that the gap between AI capability and our ability to control it is widening exponentially, and that AGI may arrive around 2027 with superintelligence soon after. He predicts up to 99% unemployment as cognitive and then physical labor get automated, with humanoid robots competing with humans by 2030. He argues controlling superintelligence is not merely hard but impossible, that AI is a black box even to its makers, and that the only safe path is building narrow tools rather than racing to general superintelligence. The conversation also covers his belief that we almost certainly live in a simulation, the case for radical life extension, and his bullishness on Bitcoin as the only truly scarce asset.
Big reveals
- Yampolskiy says he coined the term 'AI safety' and spent his first five years trying to solve the problem before concluding it is impossible.
- Predicts up to 99% unemployment, not just 10%, as nearly all jobs become automatable.
- Argues almost no job is safe; even a human accountant or maker is 'almost a fetish' with no practical reason.
- Dismisses 'just unplug it' as silly, comparing superintelligence to a virus or Bitcoin you cannot turn off.
- Says Sam Altman 'puts safety second' and may want to control the 'light cone of the universe.'
- States he is 'very close to certainty' that we are living in a simulation.
- Calls aging 'just a disease' that can be cured, saying nothing stops us living forever.
- Asked if he'd press a button to shut down all AI, he keeps narrow AI but would stop AGI and superintelligence.
Things worth remembering
- In three years LLMs went from struggling with three-digit multiplication to winning mathematics Olympiads.
- Predicts AI could train on a podcaster's entire catalog and generate videos of them interviewing anyone within seconds.
- Even the companies building these models must run experiments to discover what their own systems can do.
- He compares modern AI to growing an 'alien plant' that is studied rather than engineered.
- His simulation argument: once it costs ~$10 to run a billion simulations of this exact moment, you are statistically inside one.
- Claims every religion is essentially a description of simulation theory passed down through generations.
- Believes the human genome has a 'rejuvenation loop' capped near 120 years that could be reset higher.
- Invests in Bitcoin because it is the only resource with true scarcity, capped at 21 million.
- Argues the US federal minimum wage of $7.25 should be about $25 to keep up with the economy.
- Says people move from careless developer to safety researcher with exposure, but never the reverse.
Recommended in this episode
Books, products and media the guest or host genuinely endorsed here — with the buy link.
Affiliate link — we may earn a commission at no extra cost to you.
AI: Unexplainable, Unpredictable, Uncontrollable
Roman Yampolskiy
“This book, which I think everybody should check out, I think it was published in 2024, gives a holistic view on many of the things we've talked about today, preventing AI failures.” — Steven Bartlett 01:25:59Find it on Amazon