2020-04-14 HR Examiner article Mochael Kannisto PhD Maybe We Need An AI Safe Word stock photo img cc0 by redrecords red led traffic cone 2743739 544x363px.jpg

Michael Kannisto, Ph.D. discusses how recent academic research is calling into question the validity of using AI to select employee candidates

Have you noticed the recent appearance of serious academic research that is calling into question the validity of using AI to make candidate selection decisions? Don’t worry — there’s probably no need to cancel your current contracts, but things are about to get interesting.

The HR tech industry has been notoriously opaque about how their products work. One well-known firm calls me every month to ask about scheduling a demo. I ask them to send me their validation study, and any other peer reviewed research that supports the use of their technology to make selection decisions. During my most recent monthly conversation, the salesperson surprised me by agreeing to send over a proper study. A few minutes later, an e-mail arrived with an attachment. It was an eight-page document containing not a single number. (On the other hand, it did contain several cartoon drawings. So there’s that.)

It’s no secret that these HR tech companies are swimming in investment cash. I sometimes wonder if it might be someone’s job to actually figure out how to spend all the money they receive. And the maniacal pressure to deliver returns on these investments may be contributing to the lack of dialogue about the utility of these tools. Their argument is less about whether the algorithm is a good predictor of a particular outcome, and more about how other companies are using it so “you should too.” I’ll leave those discussions to others, but the long-overdue objective research that is starting to show up in academic journals will add a needed point of view.

Once people start getting their hands on these algorithms and performing legitimate studies on the outcomes it will become easier for customers to make better purchasing decisions. We may find that the use of these tools is completely unfounded, and creates at best a random outcome (and at worst a tool for automating systemic discrimination). Or we may find that they work perfectly well, and with a few adjustments will help companies effectively sort and identify talent. At the moment, no one knows. The risks and vulnerabilities are probably unprecedented.


Michael R. Kannisto, P.h.D, HRExaminer Editorial Advisory Board Contributor.

In the meantime, the vendors keep selling and selling. One observation is the way this generation of salespeople is bypassing the traditional system of targeting the decision-maker. When Applicant Tracking Systems appeared on the scene, companies pitched to the head of recruiting. The current pool of HR tech providers uses a multi-pronged strategy – they approach TA, HRIS, Talent Management, internal innovation labs/incubators, and executives looking to be innovative. They propose pilots, partnerships, studies, and other projects designed to just get the technology in the door.

Issues such as transparency, ethics, bias, and privacy are intersecting in ways never seen before. These tools are not like typing tests; they make selection decisions based on decision-making criteria that no one will explain (or perhaps can’t explain).

You will most assuredly be having conversations about a piece of AI HR technology in 2020. How transparent will your organization be? Will candidates get copies of their results? How will you handle data privacy? Will you allow technology to remove candidates from consideration without the involvement of a human being?

I recommend the following approach to evaluating a steadily-growing assortment of vendors and products.

  • Determine and document your company’s position on AI

A few hours spent with your employment counsel, your privacy officer, your information security officer, your CIO, and your employment branding/candidate experience leader will help you articulate and document your stand on the evaluation of AI technology.

  • Meet regularly

Once you have determined your position on AI, have representatives from each of these groups meet regularly to discuss proposed technology purchases, coordinate calls with vendors, and review current external trends (lawsuits, research, media).

  • Have a “Safe Word” (to activate a pause)

Perhaps because of the approach used to sell this technology (multiple simultaneous touch points) it seems that HR technology evaluation moves really quickly. Giving every member of the governance committee the ability to “stop the line” and pause the process is a safeguard that seems appropriate with such new and novel applications of technology.

  • Measure performance

Thoughtful, well-planned, intentional implementations should be measured for performance. Did your new system impart bias? Did your applicants hate it? Is it solving the problem that prompted you to implement it?

  • Revisit your technology to see how it’s evolved (or your organization has evolved)

These products change. Upgrades and product enhancements can significantly alter the way tools appear to users, and “machine learning” implies that the algorithms will evolve. Evaluate technology regularly on an ongoing basis.

As the academic world begins to explore the wisdom of automating human capital processes with AI, it is more important than ever that the evaluation of these tech tools remain firmly within the scope of people. At least for now.

Read previous post:
Talentism: Principles for Purpose in a Pandemic

“Talentism is laser-focused on the systematic realization of human potential. What gets my attention is the compelling way they deliver...