7
Had to pick between two hiring algorithms for our team
One was a black box model that got great accuracy scores but we couldn't explain why it rejected certain candidates. The other was simpler but let us see every factor that went into a decision. We went with the explainable one after it showed us it was penalizing gaps in employment history, which seemed biased against parents who took time off. Has anyone else faced this tradeoff?
4 comments
Log in to join the discussion
Log In4 Comments
daniel_gonzalez10d ago
@drewr15 nailed it - these algorithms just mirror our own bad habits back at us, but with a shiny new coat of paint. It's like how credit scores can ding you for missing a payment when you were in the hospital, or how social media algorithms assume you're still obsessed with your ex because you clicked their profile once a year ago. The bias is always there, it just gets automated and amplified now.
7
drewr1511d ago
That part about the algorithm penalizing employment gaps really hit home for me. It reminds me of how people have always used shortcuts to judge others without realizing it. Like how some landlords automatically reject tenants with certain last names or how some managers overlook older workers because they think they won't adapt. These algorithms are just taking our old biases and making them faster and harder to spot. Glad you caught it early and picked the honest one instead of the one that sounded good but hid its flaws.
5
kim.nina10d ago
oh man, that's such a relief you caught that before it went live. it's scary how these systems can just quietly bake in old assumptions without anyone noticing until it's too late.
3
jakeb8110d ago
...and that's exactly why you've gotta treat these systems like hiring a new crew member, not setting up a toaster. You don't just flip the switch and walk away. You watch what it does for the first month. I had this happen with a scheduling tool once - it kept routing my guys through the same neighborhood at rush hour because it "learned" that was efficient from last summer's data. Had to kill that setting manually. The fix is simple: put a human in the loop for any automated decision that can screw someone over. A thirty second gut check beats six months of clean up every time.
5