Heard a data scientist say bias is just 'bad data' and nearly lost it

At a meetup in Austin last week some guy claimed algorithmic bias disappears if you just clean the data better. Has anyone else dealt with someone who thinks fairness is purely a math problem?

3 comments

3 Comments

owens.anthony1mo ago

Read a piece from some MIT researchers about how image recognition AI was labeled with "kitchen" as a woman's domain, and that bias stayed even after they cleaned up all the obvious bad data. The issue was the labels themselves came from a skewed way of looking at the world. Clean data won't fix a broken starting point. Heard another guy describe it like trying to fix a map of a city that was drawn by someone who only drove through rich neighborhoods. You can polish that data all you want, but the map is still wrong because the original viewpoint was narrow. Thinking bias is just about cleaning data misses how the whole system is built on choices people made.

blairc901mo agoMost Upvoted

Clean data won't fix a broken starting point" says it all, honestly.

max2231mo ago

I'm right there with you on this one. That "bad data" line would have made me lose it too, because it completely ignores how bias can be baked into the very questions we ask or the labels we use (even with perfect data). It's like they think fairness is just a simple math fix when it's really about messy human stuff like history and power. I've run into people who treat algorithmic fairness like a homework problem you solve by tweaking some numbers. Drives me nuts when they won't see the bigger picture.