Roman Yampolskiy vs Joscha Bach: Can Anyone Control Superintelligence?

Roman Yampolskiy says keeping a superintelligence safe forever is as impossible as building a perpetual motion machine. Joscha Bach says today's AI is more controllable than anyone expected, and that shutting it down would doom us anyway.

📬 Get the transcript, fascinating bonus content, and my Monday M.A.G.I.C. Message: https://briankeating.com/ai

Roman Yampolskiy is a computer scientist at the University of Louisville who has spent more than a decade on the AI control problem. Joscha Bach is a cognitive scientist who studies how minds, biological and artificial, build a model of reality.

We cover: why a perpetual safety device may be as impossible as perpetual motion, what happens when AI research automates itself (Roman puts it at 28 percent today), whether nuclear deterrence is the right analogy, who really benefits when AI labs ask to be regulated, and why consciousness may have nothing to do with whether AI is safe.

Being conscious does not make you kind.

00:00 Simulation theory meets certain doom

03:25 Is P(doom) even a real number?

06:05 Why a simulated world still matters

09:26 Should we have skipped the Trinity test?

13:04 Did nukes actually keep the peace?

17:32 Why haven't we wiped out the spiders?

22:35 Is AI doom falsifiable?