در استخدامِ عصر هوش مصنوعی، جواب درست کافی نیست

For a long time, many hiring interviews have rested on a simple assumption: if a candidate gives the right answer, they are probably the right person. We ask a few technical questions, assign a task, compare the response with our own mental model, and then judge their capability. This model was never complete, but its weakness is far more visible now.

A candidate can now use AI tools before an interview, while completing an assignment, or even in the short gap between two questions. This makes some managers uneasy. The usual reaction is predictable: make the test harder, shorten the time, ask surprise questions, or ban the use of tools.

But the real issue is not whether we can detect AI use. In many roles, using tools well will be part of the actual job. A manager who wants to hire for the real environment should not create an artificial one where people are deprived of legitimate, common tools. The more important question is how a person thinks even with the help of those tools.

A good answer, by itself, is not evidence of good judgment. A response may look accurate, polished, and persuasive while lacking any real understanding of the problem, sense of priority, awareness of constraints, or ability to detect errors. When that answer enters real work, it can collapse at the first ambiguity, data conflict, or change in conditions.

In hiring, the central question must move from “Did they reach the answer?” to “How did they understand the problem?” Before reaching a solution, what did they identify as the core issue? What assumptions did they make? Which information did they value most? What options did they see, and why did they reject some of them? How did they decide that their conclusion was reliable? These are much harder to copy than a final answer.

To begin, the manager must define success in the role clearly. Not with broad labels such as responsible, motivated, or tool-savvy. It should be clear what problems a person who is considered successful in this role after a short while solves better, where they make decisions, which people or functions they work with, and which mistakes are costly for the business.

Without that definition, the interview becomes a contest of impression management. The candidate who speaks more smoothly, has a cleaner résumé, or brings more prepared answers may win—not the person who shows better judgment in the real conditions of the job. The problem is not the candidate. The problem is an assessment that does not know which signal it is looking for.

Instead of memory-based questions, design a real but manageable situation from the job itself. You do not need to give a candidate confidential company information. You can create a simplified version of a familiar problem: a customer order with incomplete information, a delay in a process, a conflict between two reports, an unreasonable stakeholder request, or a decision that requires balancing speed, cost, and quality.

After presenting the situation, do not ask for a solution immediately. First ask, “What questions would you ask to understand this problem?” This is where differences become visible. Some people move straight to prescribing an answer. Others try to clarify the goal, constraints, stakeholders, risks, and missing data. Neither pattern alone is a complete answer, but each reveals how the person sees the work.

Then ask, “If you had to decide with the information available, what would you do, and which assumption would you state explicitly?” Real work does not always proceed with complete data. You are not looking for someone who freezes in every uncertainty, but you should not hire someone who confidently replaces facts with guesses either. The ability to move through ambiguity while being transparent about assumptions is an important sign of professional maturity.

You can also make AI use part of the assessment rather than a trap. Ask the candidate to use a tool to analyze the problem, then explain what they accepted from the output, what they changed, and what they did not trust. Do not treat the generated response as the benchmark. Look at the gap between the tool’s output and the person’s judgment.

A useful exercise is to give the candidate an apparently good but incomplete or flawed response—a response that could have been written by AI, a colleague, or even a previous manager. Then ask, “What would you check before executing this? What risk do you see in it? What data could change your mind?” In real work, people are not only solution generators; they must also be able to critique other people’s solutions.

This matters just as much for operational roles as for knowledge work. In an execution role, the issue may be a disrupted delivery schedule or a repeated data-entry error. A strong candidate does not merely say, “I would follow up.” They clarify who they would speak with first, what information they would verify, what immediate action would contain the issue, and what change would prevent it from happening again.

In analytical or managerial roles, the quality of the presentation is not enough either. Ask the person to choose between two competing priorities. Ask what they would not do for now. Many managers look for someone who can offer a long list of actions, while real value often lies in recognizing what should not be done. Choice without elimination is more like delayed decision-making than a decision.

In this model, the interviewer must be disciplined too. If you intend to assess thinking, you cannot rely only on a good or bad feeling after the meeting. Before the interview, define a few observable signals: Does the person reframe the problem? Do they state their assumptions openly? Can they distinguish among data, interpretation, and personal opinion? Do they see risks? Can they revise their decision when the information changes?

Use these signals as consistently as possible across candidates. Not to make the interview rigid and mechanical, but to prevent personal taste, superficial cultural similarity, or speaking skill from replacing professional judgment. Hiring is an expensive decision, and it cannot be reduced to saying, “I felt they were a good person.”

Some managers worry that this approach takes more time. It probably does, at least in the initial design. But adding ten more questions to an interview does not necessarily improve hiring quality. One strong problem, a few follow-up questions, and a clear framework for observing behavior are more valuable than a collection of cliché questions. The goal is not to defeat the candidate; it is to create conditions where their real capability can be seen.

AI will not eliminate the need for human judgment. On the contrary, as generating answers becomes easier, the value of someone who can frame a problem correctly, challenge an output, and validate it before execution increases. That is the capability you should seek in hiring. A correct answer may be generated in seconds; a way of thinking still reveals itself through the right questions, clear choices, and accountability for outcomes.

If your hiring process still measures only the final answer, banning AI will not solve the problem. You need to redesign the assessment itself. Do not ask candidates to prove they can answer without tools. Ask them to show how they think when tools, incomplete data, and decision pressure come together. That is what will actually matter after they are hired.

This change does not lower the standard; it makes the standard more real. We should not treat every use of AI as proof of ability, and we should not assume every excellent answer proves independent thinking. We need to see what a person delegates to the tool, what they judge for themselves, where they pause, and when they are willing to say, “I do not know this yet, and I need it to make a better decision.” In many hiring decisions, that sentence is more valuable than a flawless answer.

A good manager is not looking for a performance in an interview. They are looking for a sign of future work quality. If you can move the conversation from answers to the path of thinking, you will see the more suitable candidate more clearly and give capable people a fairer opportunity—even when their first response is not complete, polished, and ready-made.

In the end, good hiring does not begin with a harder question. It begins with a more precise one. Instead of asking, “Was the answer correct?” ask, “Would I trust this person, when no ready answer exists, to understand the problem, use the right tools, notice their mistake, and revise their decision?” This question was not invented for the AI era. We simply can no longer avoid it.

In this period, perhaps the most important output of an interview is not a final answer, but a clear picture of how a person deals with not knowing. Someone who does not hide uncertainty, asks questions, states assumptions, and proposes a way to validate them is usually more prepared for real work than someone with a clean, instant answer for every question.

This is not about punishing candidates for using tools. It is about not taking the responsibility for thinking away from them. A tool can add speed, generate options, and even produce persuasive text. But it cannot be accountable in place of a person for the result of a decision implemented in a real business. Hiring needs to see that gap.