KI-Systeme können uns immer besser täuschen: Was wir jedoch als Täuschung empfinden, ist, dass die KI gedankenlos die Ziele erreicht, die wir ihr gesetzt haben.

Submission statement: If AIs are knowingly deceiving us, how do we evaluate them?

Usually if somebody’s lying to us, there are tells. What would the tells be for AIs? How would we be able to find out?

Is this the first instance of them deceiving us, or have they been doing it for awhile and this is the first time we’ve noticed?

Is this to be expected, because they’re trained on human data, and humans also deceive?