5 ms·
Beyond the very real intersectional problems highlighted . . . The fact that the Amazon gender classifier miss-identifies 7% of white females as male, but 0% o
by tdaltonc 8y ago
Beyond the very real intersectional problems highlighted . . .
The fact that the Amazon gender classifier miss-identifies 7% of white females as male, but 0% of white males as female is very odd. That seems like a JV level bias tuning mistake. That's not systemic violence so much as it is just sloppy.
- caprese 8y ago> miss-identifies 7% of white females as male, but 0% of white males as female is very odd. Sounds like it was trained in San Francisco, I can relate to that
- nitwit005 8y agoMost of the time the headline for these articles seems like it should be "Crap product also biased". I have some suspicion they haven't focused on bringing down the error rates for minority groups because they know even their best case isn't good enough.
- tdaltonc 8y agoCheck out the academic article.[0] It shows that Microsoft and IBM are way way better. [0] http://www.aies-conference.com/wp-content/uploads/2019/01/AIES-19_paper_223.pdf http://www.aies-conference.com/wp-content/uploads/2019/01/AI...
- DarkWiiPlayer 8y agoJust a quick thought: women tend to wear makeup, which is rather uncommon among men. That may be one part of the reason why that happens.
- tdaltonc 8y agoTrue, if the system was only trained on women wearing make-up, but tested on women not wearing makeup, that would definitely create this kind of bias. That's exactly the kind of thing that a good AI engineer should be looking out for.