Roman Yampolskiy: Why Controlling Superintelligence May Be Impossible
Notas do episódio
Most AI researchers think alignment is a hard problem. Roman Yampolskiy argues it's an impossible one.
Yampolskiy is a computer scientist at the University of Louisville and one of the longest-standing voices in AI safety research. For over a decade he has argued that controlling a system more intelligent than ourselves isn't a challenge the field will eventually solve — it's a structural limit we can't get past. His position sits at the far end of AI safety discourse, and most of the field disagrees with him. In this conversation we get into why he holds it, what he thinks the rest of the community is missing, and what the next decade looks like if he's right.
What we cover:
- The impossibility argument: why he sees alignment not as a problem to solve, but as a wall.
- His timeline for ASI, and what — if anything — could meaningfully slow it down.
- Why he's skeptical of the alignment work being done at the frontier labs.
- What happens to economies, employment, and human agency in the years before ASI arrives.
- Whether any policy or technical move in 2026 actually buys time.
If Yampolskiy is wrong, this is one researcher's pessimism. If he's right, almost nothing else we're doing matters.
Timestamps:
- 00:00 Introduction
- 01:10 When Will AI Become Dangerous to Humanity?
- 02:37 AI Tried to Avoid Being Shut Down
- 03:13 When Did AI Try to Lie?
- 03:44 Have We Entered the Technological Singularity?
- 04:58 When Should We Expect the Emergence of AGI?
- 06:17 Is Polymarket an Indicator That an Event Will Happen?
- 07:20 What Will People Feel When ASI Appears?
- 09:03 Which Professions Will Disappear — and Which Won't?
- 09:44 What Is the Motivation of AI Creators?
- 11:32 Should We Expect No Limits on AI?
- 12:58 Is Nuclear Weaponry Comparable to AI?
- 13:42 Skepticism Around AI
- 15:28 Would Other Civilizations Also Create AI?
- 17:06 About the Simulation Hypothesis
- 18:17 Are We Living in a Simulation?
- 20:01 When Could the Simulation Have Been Started?
- 20:38 How to React Calmly to the Future?
- 21:30 How Can Humanity's Extinction Date Be Delayed?
- 22:25 76% of People Have Never Used AI
- 23:29 Investing Money in AI
- 23:59 AI Safety Departments in Companies
- 24:50 Elon Musk's Vision of the Future
- 26:36 Future Competitions Without AI
- 27:02 What Is the Probability That AI Will Destroy Us?
- 29:06 Does Humanity Have Only a Few Years Left?
- 29:51 Competition Between AIs
- 31:25 OpenAI's Contract With the Pentagon
- 32:23 Does Claude Have Consciousness?
- 35:05 The Future of AI
- 36:25 Scientists From the 2000s Reacting to AI
- 38:06 Is a Quantum Computer Possible?
- 39:26 The Microtubule Hypothesis and Neuromorphic Computers
- 40:01 Do We Have Only 2 Years Left Until the Point of No Return?
- 41:32 Is There a Ray of Hope?