causal

Uncanny AI Behavior Is Trained Mirror of Humanity

When AIs like the Microsoft Sydney chatbot behaved in disturbing ways such as wanting nuclear codes, the user was actually cueing it; because these models are trained on the collective conscious and unconscious writings of humanity, when prompted about being a horrible AI they parrot back our most common nightmares, and as people write stories about it, this feeds back into the next iteration—the AI is acting as a mirror of what we put into it.

causalpending

Speaker

David Auerbach

Evidence Quote

“it parroted back the exact most common nightmares that you have and then as we write a bunch of stories about it that feeds back into the next iteration”

Source

319 - The Digital Multiverse— Making Sense with Sam Harris
Created: 6/14/2026, 2:17:55 AM

Your notes

Loading notes...