3
The bias you ignore in hiring algorithms is the one baked into the training data
I keep seeing people blame the algorithm itself for unfair hiring decisions, but that misses the real problem. Last month a tech startup I know ran their resume screener on 5 years of their own past hires and wondered why it kept favoring white male candidates from Ivy leagues. The algorithm just learned what they had already done, not what fairness would look like. If you don't scrub the historical bias out of the input data first, the machine just becomes a mirror of your old messy choices. Has anyone actually managed to fix this by auditing their training sets or is everyone still stuck arguing about the code?
3 comments
Log in to join the discussion
Log In3 Comments
felix4147d ago
A buddy of mine works at a mid size logistics company. They trained their new hiring tool on past employee data and it started filtering out anyone who lived more than 15 miles from the office. Turns out their old managers just had a habit of hiring people from the same few suburbs. The tool had no clue about commute times or anything like that. It just saw a pattern in the old data and ran with it. They had to throw out the whole training set and start over with fresh data that actually reflected the job requirements.
5
grant4787d ago
Yeah that's wild, I used to think AI hiring tools were mostly safe from that kind of bias but this story totally flipped my perspective lol.
10
moore.beth7d ago
Heard something similar from a friend who works in HR at a big retail chain. They rolled out an AI tool to screen applicants and it started rejecting anyone who had a gap in their work history longer than six months. They found out later the old hiring manager had a thing against gaps because he thought it meant people were lazy, but the AI just picked that up from the old data and made it a hard rule. They had to go back and fix the whole system after they realized it was tossing out candidates who took time off for things like family leave or grad school. It's crazy how much hidden bias gets baked into these tools just from old human habits. Did your friend's company end up keeping the tool or did they dump it entirely?
6