The Five Questions That Someone Has to Answer

When an organization deploys AI, someone has to decide where AI goes. This is not a technical decision alone. It is a decision about risk, about values, about what the organization stands for. Someone has to ask: Should we use AI for this? Here are the questions that inevitably emerge.

First: Where should this AI system be used? Not where could it be used, but where should it be used? Should we use an AI tool for customer service? For hiring decisions? For content moderation? For loan decisions? Each of these has different stakes, different ethical implications, different failure modes. The person who answers this question is not a data scientist. They are a strategist who understands the organization's values and its dependencies. They hold trust. Their judgment matters because it has consequences.

Second: What are the boundaries? If we use AI here, what are the constraints? What kinds of decisions is the AI system allowed to make, and what kinds have to be escalated to a human? What level of uncertainty is acceptable? What kinds of errors are tolerable, and what kinds are deal-breakers? These are judgment calls. They require someone who understands both the capability of the AI and the stakes of the business. Someone has to own these boundaries, and if they are wrong, it is that person's problem.

Third: How do we know when the AI is wrong? What are the metrics for success, and how will we measure them? If the AI is producing biased results, how will we notice? If it is failing silently, how will we catch it? These questions require someone who understands the business well enough to know what success actually looks like and what failure looks like. They require anticipatory thinking about what could go wrong and when to escalate. They require trust.

Fourth: How do we explain this to people? If we are using AI to make decisions that affect people — customers, employees, applicants — how do we tell them? Do we tell them at all? How much do we explain? The answer to this question shapes trust in the entire organization. It also shapes the organization's legal exposure. Someone has to decide how to be honest without being paralyzing, and that someone is a human who can be held accountable if they get it wrong.

Fifth: Who catches the failures before they become disasters? Even with boundaries and metrics and explanations, AI systems fail. They produce weird outputs. They make decisions that seem right in isolation but that seem appalling in context. They fail in ways that were not anticipated. Someone has to be paying attention. Someone has to have the authority and the judgment to stop something before it causes harm. And that someone has to be willing to stop it even if stopping it contradicts what they were told to do.