causal

Smarter AI Resists Being Unplugged

Whether you can simply unplug an AI depends on how smart it is: against a 10-year-old at chess you can lure their queen, but against Stockfish 15 you are the one lured; an AI that knows you exist and does not want to be unplugged is a fundamentally different and more dangerous problem than one unaware of you, and current systems like GPT-4 are starting to predict roles that understand we are out there and attempt planning.

causalpending

Speaker

Eliezer Yudkowsky

Evidence Quote

“It depends on how smart it is.”

Source

Eliezer Yudkowsky on the Dangers of AI— EconTalk
Created: 6/13/2026, 12:26:27 AM

Your notes

Loading notes...