When I trusted artificial intelligence more than I should

A real experiment almost harmed one of my projects.

I am among the most convinced of AI’s value in human resources, and I use it to analyze resumes, write job descriptions, prepare interview questions, and compare candidates. But I have learned from a real situation that using a powerful tool does not mean handing over our minds to it or granting it absolute trust.

In one recruitment project, I used an artificial-intelligence tool to sort résumés and assess how well candidates matched the job requirements. I did not use it randomly; I had trained it well on the project’s requirements, defined the required experience, and explicitly asked it to look for anyRed FlagsInside every résumé.

I received an excellent analysis of one of the candidates. The experience is suitable, the skills are aligned, the assessment is reassuring, and the tool did not detect any fundamental problem. Based on this analysis, I sent the CV to the project.

But during the human review, the surprise emerged.

The candidate had listed in their experience a work period that begins in a month and year that have not yet arrived, i.e., they recorded future employment experience, a clear inconsistency that should have been identified immediately.

The information was present in front of the tool, and the instructions were clear, but it did not capture it. In turn, I relied on the result without conducting the final review myself.

In that moment I realized that the real danger is not that artificial intelligence makes a mistake; error is possible in any system. The danger is that we trust its answer to the point where we stop thinking and verifying.

Why didn't artificial intelligence detect the error?

Artificial intelligence may analyze a large amount of information quickly, but it does not always understand it in the same way a human does. It may focus on keyword matches, the number of years of experience, or skill alignment, and overlook a clear temporal inconsistency.

A result that appears organized and confident is not necessarily correct. Artificial intelligence can present an error in a very convincing manner, and that is where its danger lies.

Even when we train the tool well and give it precise instructions, we cannot assume that it will apply every standard correctly each time. Good training reduces the likelihood of error, but it does not eliminate it.

What could have happened?

It may seem like a simple mistake, but in recruitment it may lead to:

• Nominating a unsuitable person for the client or the project.
• Damaging the credibility of the hiring manager or consultant.
• Wasting the interview team's time.
• Making a decision based on inaccurate information.
• Excluding a qualified candidate or preferring a less competent candidate.
• Exposing the institution to professional, legal, or ethical risks.

The tool does not bear responsibility for the decision before the client, management or the candidate. The ultimate responsibility remains with the person who used it and relied on its outputs.

Artificial intelligence is an assistant, not a replacement for the mind.

This experiment does not mean that we should stop using artificial intelligence. On the contrary, it can save a great deal of time, help us organize data, make comparisons, and discover patterns that may be difficult to detect manually.

But we must put it in its proper place:A smart assistant provides recommendations, not the final decision-maker.

When screening résumés, for example, AI can conduct the initial analysis, but the specialist must review it themselves:

• The chronological sequence of experiences.
• Overlap or gaps between functions.
• Validity of years of experience.
• Alignment of titles with responsibilities.
• Qualifications, certificates, and dates.
• any possible exaggerations or contradictions
• The degree of the candidate’s suitability to the actual context of the job.

Human expertise does not consist merely of reading information; it also involves understanding context, asking questions, noticing what does not seem logical, and taking responsibility for the decision.

The lesson I learned

The biggest mistake we can make is not using artificial intelligence, but using it without review.

We may write the best instructions, train the tool on every detail, and ask it to look for errors and risks, yet it may fail to notice an obvious piece of information. Therefore there should always be a final stage called:Human Verification.

They should use AI to accelerate work, not to eliminate their role.

Let them propose, analyze, organize and compare, but do not let them make the decision for you. Because future competence will not be in the person who only uses artificial intelligence, but in the person who knows when to trust it, when to doubt its answer and when to intervene with their mind and experience.

Artificial intelligence may be an exceptional assistant, but it still needs an attentive human behind the screen.

By: Hadeer Nabil

HR Consultant | CEO

About the author Hadeer Al-Maghribi is a human-resources and business consultant with more than 11 years of experience since 2014, specializing in institutional development, performance management, transformation and restructuring. She designed performance-indicator systems that helped raise team productivity by up to 20% and built structures and functional roles that affected over 100 employees. She works with the Ro'aa Training Center and the Arab Academy for Science, Technology and Maritime Transport.