Roman Yampolskiy: AI Can’t Be Controlled — and We’re Building It Anyway

Roman Yampolskiy spent two decades trying to prove superintelligent AI can be controlled. He couldn't — and his book became a proof that the problem can't be solved at all. Not difficult. Mathematically impossible. I pushed back hard, because that is a claim worth breaking if it can be broken.

Roman is a professor of computer science at the University of Louisville and one of the earliest researchers in AI safety. We ran my "Einstein test" — could a model trained only on pre-1911 physics rediscover what Einstein did with the same data? It failed, and we disagree about what that failure means.

In this conversation:
- Whether AI control is mathematically impossible or merely unsolved
- Why Roman calls today's AI safety work "security theater"
- What the halting problem really says about predicting smarter-than-human minds
- Value alignment: a real problem, or a well-funded category error
- The case for a government moratorium on frontier models
- What to tell your kids about a career in a world where Roman might be right

CHAPTERS
00:00 The mind we can't switch off
01:34 You're the squirrel in this story
04:08 The thought AI may never have
08:18 We ran the test. It failed.
12:22 The trap we already locked ourselves in
16:10 Why AI still can't crack physics
21:54 Why toddlers are safe and teenagers aren't
23:06 The proof you can't predict a smarter mind
25:58 No system can vouch for itself
28:18 Does regulation just protect the oligarchs?
33:28 Is your mind more than a machine?
39:00 Do we owe AI anything?
43:00 The monster hiding behind the smiley face
46:22 A perfect universe built just for you
49:50 What if we're the alien probes?
55:02 The one kind of AI that should scare you
58:22 The movement begging labs to stop
59:58 What do you tell your kids now?
01:07:58 The question he'd put to Sam Altman
01:15:48 The debate he's dying to have
01:19:50 What he'd tell his 20-year-old self

If you listen to other people, the best you can become is average.

Substack: https://briankeating.substack.com/ss
Transcript, bonus content, and my Monday M.A.G.I.C. Message: https://briankeating.com/yt

Have a .edu email and live in the USA? You automatically win a meteorite: https://BrianKeating.com/edu

Subscribe: https://www.youtube.com/DrBrianKeating?sub_confirmation=1

Support Into the Impossible on Patreon: https://www.patreon.com/drbriankeating

Join this channel for perks, monthly Office Hours, and your name in the Member Roster: https://www.youtube.com/channel/UCmXH_moPhfkqCk6S3b9RWuw/join

Featured Guest — Roman Yampolskiy
The Roman Forum (YouTube): https://www.youtube.com/c/RomanYampolskiy?sub_confirmation=1
Twitter/X: https://x.com/romanyam
AI: Unexplainable, Unpredictable, Uncontrollable: https://www.romanyampolskiy.com/books/

My books:
Losing the Nobel Prize: http://amzn.to/2sa5UpA
Think Like a Nobel Prize Winner: https://a.co/d/03ezQFu
Focus Like a Nobel Prize Winner: https://a.co/d/hi50U9U
Galileo's Dialogue (first-ever audiobook): https://a.co/d/iZPi9Un

Twitter/X: https://x.com/BrianKeating
Blog: https://briankeating.com/blog
Audio-only: https://briankeating.com/podcast