AI's worst harms aren't designed—they're emergent. The temperance crowd can't tell the difference. That's why they're useless, and understanding matters more.