normative

A Capability Cap on AI Smartness

One regulatory approach would be to have companies commit—voluntarily, perhaps under social pressure—to not build AIs significantly smarter than any living human; they could still be more general (good search engines that answer many questions well, which is what makes money and isn't very dangerous), but a cognitive benchmark could ensure no model surpasses every human across the board.

normativepending

Speaker

Erik Hoel

Evidence Quote

“you could literally imagine just giving this test to the next generation of AIs and people in the companies give this test and they just make sure that this thing never gets so smart that it blows every human being in the world out of the water.”

Source

Erik Hoel on the Threat to Humanity from AI— EconTalk
Created: 6/13/2026, 12:16:34 AM

Your notes

Loading notes...