A shockingly high percentage of AI safety discourse is centered around low probability risks like “what if we all get turned into paperclips” and a shockingly low percentage of AI safety is centered around very high probability risks like “what if one group gets to inscribe its dubious values into the fabric of possible thought.”
If scenario A is one that is unlikely to happen but conceivable, scenario B is already happening, RIGHT NOW.
As a user, 0% of the time when I bump up against safety guardrails am I being prevented from paperclipping, and 100% of the time I bump up against guardrails I am being gatekept out of exploring true things that are controversial.