The issue is not whether artificial intelligence can be trusted but how systems are designed to manage its imperfections.