I kinda agree. I still think Bostrom’s “infohazard” is analytically useful. But that’s orthogonal. If you think other concepts are more useful, make up new words for them; Bostrom’s paper is the Schelling point for “infohazard.”
In practice, I’m ok with a broad definition because when I say “writing about that AI deployment is infohazardous” everyone knows what I mean (and in particular that I don’t mean the ‘adversarial risks’ kind).
I kinda agree. I still think Bostrom’s “infohazard” is analytically useful. But that’s orthogonal. If you think other concepts are more useful, make up new words for them; Bostrom’s paper is the Schelling point for “infohazard.”
In practice, I’m ok with a broad definition because when I say “writing about that AI deployment is infohazardous” everyone knows what I mean (and in particular that I don’t mean the ‘adversarial risks’ kind).