When Human Judgment Isn't Enough
Why Recruiters Must Earn Their Authority in AI-Enabled Recruiting
Perhaps due to our insecurity or fear of losing our jobs, we have made human-in-the-loop the centerpiece of both law and AI governance, often without asking whether the human in that loop is competent or capable of sound reasoning.
The National Institute of Standards and Technology, an agency of the U.S. Department of Commerce, and its Artificial Intelligence Risk Management Framework emphasize defining human roles and responsibilities and understanding the limitations of human-AI interactions. They do not recommend just accepting human involvement as sufficient.
An AI algorithm may be as competent as or more competent than a human recruiter, and there is the rub. How to tell?
If an algorithm says that Candidate A fits the job requirements at 85% and Candidate B at only 63%, and a recruiter recommends the first candidate, can they explain why, or did they just rubber-stamp the recommendation?
To have faith that the recruiter is competent and at least as good as the algorithm, I believe they should be qualified at three levels.
The first is recruiting competence. The recruiter needs to understand both what the hiring manager is seeking and what the organization needs. They need skills in assessment techniques and their limitations, and in interpreting a candidate’s answers in relation to the job and the hiring manager, not just the specific requirements. They need to “read between the lines,” so to speak, and have some level of intuition.
Second is AI competence. They must understand the algorithm performing the evaluation and know what it was designed to interpret. They must know what it uses for inputs, its limitations, and possible failure modes. This does not require being a data scientist but does require a basic understanding of how these systems work,
An example is an airline pilot trained to this level of competence. They do not need to understand every line of code, but they must understand what the algorithm or software is supposed to do and how to recognize when it becomes unreliable.
