Dr. Darrell Jaya-Ratnam
Dr. Darrell Jaya-Ratnam

The search for a job has turned into a lottery, and artificial intelligence is giving both sides the tickets.

After spending 40 years watching recruitment evolve, the latest transformation worries me more than anything that has come before. We are building a system in which candidates use AI to present themselves as the ideal employee, while employers use AI to identify the ideal candidate.

Each side is becoming better at manipulating the other's technology, and the person who actually deserves the job risks getting lost between the two.

The numbers suggest this is already happening. Greenhouse's 2026 Candidate AI Interview Report found that 63% of job seekers had experienced an AI interview, up 13 percentage points in six months. Yet 70% of those interviewed said they had not been clearly told beforehand that AI would evaluate them. We are automating one of the most consequential decisions in a person's life while leaving many of the people affected uncertain about how it works.

The public debate tends to focus on whether a machine should read someone's CV. I think we are asking the wrong question. Human recruiters have always worked under constraints. Give an employer 2,000 applications and limited time, and some CVs will receive little more than a cursory glance. I remember applying for jobs decades ago and seeing applications dismissed through arbitrary shortcuts. Human involvement has never guaranteed careful consideration, impartiality, or even basic consistency.

So why do we assume that putting a human back into the process automatically makes it fair?

Human judgment carries its own baggage. The halo effect, familiarity bias, and unconscious preferences influence how we assess people, often without our knowledge. I have spent much of my career confronting the halo effect myself. As a short, bald Asian man, I have competed against people whose appearance gave them an immediate advantage. Not being aware of a preference and not having a preference are two very different things, and this misunderstanding is influencing hiring decisions.

Stanford research examining four million applications processed through an AI screening tool found that Black and Asian candidates faced disadvantages in many assessments. AI does not eliminate prejudice. It can reproduce assumptions embedded in training data and evaluation criteria. Yet systems can be designed to exclude irrelevant personal characteristics and normalize assessment criteria far more easily than you can train humans to avoid their preferences.

To achieve this, employers must be prepared to design and test these systems properly.

The immediate danger is an arms race between two sophisticated systems. Candidates use AI to construct persuasive CVs, optimize language, and send applications at scale. Employers deploy AI to filter them. Candidates need relevant experience, the ability to communicate it to AI, effective prompts, and a system that represents them accurately. The employer then runs that output through another system, which may reward the presentation the first AI has perfected.

Who wins?

Potentially, the person with the better AI, the better prompts, or simply the luckier combination of algorithms. The employer may overlook an exceptional candidate, while an unsuitable applicant advances because their AI knows how to play the game. Multiply this across thousands of vacancies, and the recruitment process starts resembling an AI lottery.

There is a better way to use this technology. AI should help employers understand people, including what they struggle to communicate on paper. A CV records what someone chooses to write about their career. A meaningful interview can uncover why they made certain decisions, how they handled difficult situations, and what they learned from failure. Someone changing industries may have valuable transferable skills without knowing how to explain their relevance to a new employer.

This requires AI that can listen and ask intelligent follow-up questions with genuine curiosity, not simply AI that treats all candidates the same. I have seen how differently people respond when they feel free from the immediate judgment of another person. In our experience developing conversational and probing AI, individuals have openly discussed workplace failures and raised issues they might never have volunteered to a colleague. That information has enormous value. It gives employers access to a richer account of human capability, including the experiences that never make it onto a CV.

AI should also support selection without owning the final decision. Employers can use it to identify clear matches, clear mismatches, and candidates who deserve closer examination. Human oversight must remain meaningful, particularly for borderline cases.

Companies should also disclose AI involvement, explaining what their systems evaluate, and ensure the AI provides candidates with meaningful reasons for rejection. I would go further. Employers should build and audit their own assessment data against the actual performance of successful employees, examining whether their definition of excellence inadvertently rewards a particular background or appearance to the detriment of company performance.

A company cannot identify its biases if it has never examined the standards through which it measures talent.

The future presents two very different possibilities. One gives us workplaces where ability receives a fairer assessment, transferable skills become visible, and people find opportunities they might previously have missed. The other gives us automated application factories, opaque rejection decisions, constant employee turnover, and businesses struggling to retain people who were never properly matched to their roles.

The consequences will reach every household. Employment determines financial security and the ability to plan a future. The World Economic Forum estimates that structural labor-market changes could create 170 million jobs and displace 92 million by 2030. The quality of how we match people to those opportunities will matter enormously.

Business leaders must stop treating AI recruitment as an efficiency project and start treating it as a responsibility to the wider stakeholders such as customers, shareholders, and colleagues. CEOs and chief human resources officers should demand independently tested systems, meaningful human control, and evidence that hiring decisions predict actual performance. Technology should help candidates reveal their capabilities through genuine conversation, not simply compete through increasingly polished applications.

We are deciding who gets access to the next generation of work. That decision deserves more intelligence than an algorithm sorting CVs into yes and no. AI can make recruitment more human and equitable, but that future depends on the standards employers are willing to enforce today.

About the Author

Dr. Darrell Jaya-Ratnam is the managing director of DIEManalytics. A Cambridge engineering graduate with a PhD in ballistics, he has over 25 years of expertise in decision support, operational analysis, and artificial intelligence. He previously worked with the UK Ministry of Defence and sits on the MOD's AI Ethics Advisory Panel.