We are often more certain than the evidence really allows.
More quotesRecruiter 83 receives an AI recommendation, reads the résumé, agrees and saves four minutes.
Recruiter 84 receives another recommendation after twenty sensible ones. It is specific, calm and wrong. She accepts it in twelve seconds.
These two recruiters are illustrative, not individuals reported in the source. They show how competence can remove the human behaviour a safety process supposedly relies upon. Correct recommendation after correct recommendation teaches that scrutiny wastes time. Speed is praised. Agreement clears the queue. Doubt creates work. By the first serious error, the operating habit is already submission.
In the source, 181 professional recruiters assessed 44 applications with AI of different quality. Those using the better system performed worse than those using the weaker one. They spent less effort, followed recommendations more readily and failed to improve. The weaker system demanded attention because disagreement remained an ordinary part of using it.
“A human remains accountable” is policy wallpaper. Accountability does not live in a sentence at the bottom of the procedure. It lives in time allowed per case, access to the underlying material, authority to refuse and consequences that do not punish somebody for slowing the machine down. Put the recommendation first, measure throughput and celebrate a high acceptance rate; the click has been designed.
The efficiency calculation counts minutes saved on routine cases. It needs to include missed exceptions, declining human skill and the cost of recovering when the system moves outside familiar conditions. Otherwise the organisation removes checking from the job, keeps accountability in the policy and discovers the difference in rejected applicants.