The AI Safety Debate Is All Wrong