A big problem here is that AI can be responsible for "anchoring", so even if its characterization is known to be false and corrected, the residue of that false

A big problem here is that AI can be responsible for "anchoring", so even if its characterization is known to be false and corrected, the residue of that false impression colors how everything is interpreted thereafter. The "yeah, but still" effect, you might say.