> By sending an email that repels all but the most gullible the scammer gets the most promising marks to self-select, and tilts the true to false positive ratio in his favor.

There's less need to identify the most gullible when AI drops the incremental cost of responding closer to zero.