If You're Going to Defend AI, You Should Be Honest About Its Actual Harms