I build Hiring AI, and I am arguing against one specific thing: AI that autonomously rejects candidates.
The reason is not that automate=bad. Hiring has no real clean ground truth to train on. "Good Hire" is measured late, confounded and survivorship-biased: you never see how the people you rejected would have done in the job. So a model that auto-rejects is confident it made the right call, and the candidate is left in the mud.
I'm not claiming humans are unbiased, as its quite the opposite some times, but the argument is "Give the human structure (fixed criteria, evidence over impression) to cut bias, and it will most likely make better decisions on itself". In this scenario, the machine's job would be that, to extract evidence, not to decide.
We optimize the process so we can get the best of both worlds, humans and machine.
I build Hiring AI, and I am arguing against one specific thing: AI that autonomously rejects candidates.
The reason is not that automate=bad. Hiring has no real clean ground truth to train on. "Good Hire" is measured late, confounded and survivorship-biased: you never see how the people you rejected would have done in the job. So a model that auto-rejects is confident it made the right call, and the candidate is left in the mud.
I'm not claiming humans are unbiased, as its quite the opposite some times, but the argument is "Give the human structure (fixed criteria, evidence over impression) to cut bias, and it will most likely make better decisions on itself". In this scenario, the machine's job would be that, to extract evidence, not to decide.
We optimize the process so we can get the best of both worlds, humans and machine.