The AI Act should use humans to monitor AI only when effective

The EU’s proposed AI Act wants to use humans to oversee AI-generated decisions. Recent evidence suggests one should carefully test when that is even possible.