For generations, getting a job required convincing another person that you could do it. A résumé might get you through the door, but eventually a manager had to make a judgment about your experience, ability, and character.
Read more George Clooney, Mark Ruffalo, and the Free Speech Straw Man: How Cowardice Sanitizes Hate
That process was never perfect. Human beings bring their own prejudices and preferences to hiring. But the growing use of artificial intelligence in recruitment introduces a different problem: the person making the first judgment may no longer be a person at all.
Employers increasingly use automated systems to source candidates, screen résumés, administer assessments, and rank applicants. The appeal is obvious. A company receiving thousands of applications cannot have a manager read every résumé carefully. Software can process information faster and impose consistent rules.
But efficiency is not the same thing as judgment. And if employers are not careful, the pursuit of efficiency can quietly change what “merit” means.
Graphic: X Post
Automated screening is most useful when an employer knows exactly what it needs and can reliably identify those qualifications. Problems arise when the system begins treating easily measurable characteristics as substitutes for actual ability.
A résumé may contain a job title, degree, certification, employment dates, and a collection of keywords. Those details are convenient for software to process. They are not necessarily the best evidence of whether someone will succeed in a particular job.
Consider two candidates for the same position. One has followed a conventional career path, accumulating recognizable job titles at well-known companies. The other changed industries, learned new skills independently, spent several years outside the workforce, or gained experience through smaller organizations.
A rigid screening system may find the first résumé easier to classify. A thoughtful hiring manager might find the second candidate more interesting.
That distinction matters because unconventional career paths are not necessarily evidence of inferior ability. They may reflect entrepreneurship, military service, caregiving, economic disruption, immigration, education, illness, or simply a willingness to change direction.
When an algorithm favors the candidate whose résumé most closely resembles the profiles it has been trained to recognize, it can reward familiarity rather than potential.
One of the most attractive arguments for automated hiring is that machines can remove human bias. There is some logic to this. A computer does not have a personal grudge against a candidate, nor does it instinctively favor someone because they attended the same university.
But an algorithm does not need personal prejudice to produce biased results.
The Equal Employment Opportunity Commission has warned that AI and other automated systems used in employment decisions can create discriminatory barriers. The agency has specifically raised concerns about automated résumé screening, assessments, and other technologies that may disadvantage applicants with disabilities or create disparate effects.
The underlying problem is straightforward: an algorithm learns from data and instructions created by people. If the assumptions built into those systems are poor, automation can reproduce those assumptions at enormous scale.
An employer may therefore replace one biased gatekeeper with another except the second gatekeeper can evaluate thousands of people before anyone realizes there is a problem.
There is another problem that deserves more attention.
Suppose an employer discovers that many of its successful employees attended certain universities, worked for particular companies, or used particular terminology on their résumés. It may be tempting to treat those characteristics as indicators of future success.
Read more James Talarico’s faulty government charity
But correlation does not establish competence.
A person may have attended an elite university without being particularly capable at the job in question. Another person may have acquired exactly the right skills without having the educational or professional credentials that an algorithm recognizes.
This is especially important as employers increasingly talk about skills-based hiring. If the goal is genuinely to hire for skills, organizations should evaluate whether their technology measures those skills or merely identifies proxies that are easier to process.
An algorithm that rewards the appearance of qualification can undermine the very meritocracy it was supposed to improve.
None of this means employers should abandon AI.
Used properly, automated systems can perform valuable administrative work. They can organize large applicant pools, identify potentially relevant experience, reduce repetitive tasks, and help recruiters spend more time evaluating candidates.
The mistake is allowing convenience to become authority.
A screening system should assist a hiring decision, not quietly become the hiring decision.
That requires employers to understand what their tools are actually measuring. They should test whether automated criteria are genuinely related to job requirements, monitor outcomes, investigate unexpected disparities, and maintain a meaningful opportunity for human review.
has already recognized technology-related employment discrimination as an enforcement concern, including the use of algorithmic decision-making and automated recruitment and selection tools.
Employers should also remember that legal compliance is not the only reason to scrutinize these systems. A company can create a technically efficient hiring process that is still strategically foolish.
If a system consistently rejects people who could have become excellent employees, the organization is not saving time. It is losing talent.
The central question is not whether AI belongs in hiring. It does.
The question is what we want AI to optimize.
If the answer is simply speed, employers may build remarkably efficient systems for rejecting people. If the answer is finding capable workers, the technology must be judged by whether it actually helps identify capability.
That means preserving room for evidence that does not fit neatly into a predefined pattern: transferable skills, unusual experience, demonstrated ability, adaptability, and evidence of learning.
The irony is that technology could ultimately make merit-based hiring better—but only if employers resist the temptation to define merit according to whatever information happens to be easiest for a machine to process.
A résumé is a document. A candidate is a human being.
Confusing the two may be the most consequential mistake in the new age of automated hiring.
Michelle Brenier is a SaaS and technology writer specializing in AI, recruitment technology, and the changing American job market. He writes about resumes, career development, job applications, and the evolving role of technology in employment for Jump Resume Builder: