Howard Levitt: Why AI may be (re)humanizing the hiring process

Article content

Human rights legislation prohibits discrimination in employment on grounds including age and disability. It also addresses apparently neutral requirements that have discriminatory effects. Illegal discrimination does not require discriminatory intent. An employer can sincerely believe its hiring process is neutral and still have a human rights problem. Technology may simply make the problem harder to detect.

Article content

Suppose an employer uses an automated video interview that evaluates speech, facial expression, response time or other behavioural characteristics. An applicant with a disability may communicate differently, take longer to respond or have facial movements that the system regards as unusual. The candidate may therefore score poorly despite being perfectly capable of doing the job.

Article content

Once the employer knows that a disability affects the assessment, the duty of accommodation arises.

Article content

And HR departments should remember an important principle: Treating everyone identically is not treating everyone equally.

Article content

If every applicant must complete the same assessment, that does not end the human rights analysis. More likely, it compounds the difficulty.

Article content

Article content

The employer may have to find another way of assessing a qualified applicant. “Our recruiting company doesn’t offer another option” is not an answer. The employer chose the system. It remains responsible for the hiring process.

Article content

Employers are accustomed to buying software without knowing exactly how it works. That is manageable when the software schedules interviews or organizes resumés. It is different when the software decides who gets an interview.

Article content

Before adopting an AI recruitment system, an employer should be able to answer some basic questions.

Article content

Article content

An employer does not need to become an AI engineer. But it should understand enough about the product to defend the decision it makes with it.

Article content

There is another development employers should know about.

Article content

Applicants are not simply using AI to polish their resumés. They are also gaming the system, attempting to manipulate the software screening them.

Article content

Article content

Researchers have documented prompt-injection techniques in which hidden instructions are placed in resumés so that an AI system reading the document receives instructions that are invisible or meaningless to the human recruiter. They are designed to put the applicant at the front of the line. These methods will evolve further.

Article content

The point is that employers may no longer be evaluating a document created simply to communicate information. They may be evaluating a document designed to influence the software evaluating it. That is a very different resumé.

Article content

The solution is surprisingly old fashioned.

Article content

Verify. If the degree matters, check it. If the professional designation matters, confirm it. If the candidate says they managed 200 employees, ask what that actually involved. If they claim to have increased sales by 40 per cent, ask how they calculated it. If they say they solved a major operational problem, make them explain what they did, what went wrong and what they would do differently now.

Article content

And ask their references questions that cannot be answered from the resumé.