Re:Facial Recognition Is Accurate, If You’re a White Guy
by mabbo on Saturday February 10, 2018 @10:07AM
(#16347580)
From a competitive ML point of view, this is actually great news (hear me out here!). What this means, to me, is that there's a large space of room where the algorithms can improve, and likely all they need is better training data to do so.
If we can get 99% accuracy on a subset of the data (white males) while there isn't any real difference in the difficulty between that subset and another then with better data the other subsets should perform equally well.
Re:Facial Recognition Is Accurate, If You’re a White Guy
by vgf on Saturday February 10, 2018 @09:00AM
(#16347288)
From a privacy perspective I would personally see it as a positive if my skin color made it harder for me to be accurately detected by facial recognition algorithms. This seems like severely misplaced outrage. Be careful for what you wish for..
Re:Facial Recognition Is Accurate, If You’re a White Guy
by vgf on Saturday February 10, 2018 @10:08AM
(#16347583)
That seems like the... wrong issue to highlight. In this case it's all about the skin color of the people pictured in the data sets. E.g. the potential "victims" when this stuff gets inevitably gets misused. Not about the skin color of the developers.
I don't think anyone in the "AI community" is against having better data sets.
If we can get 99% accuracy on a subset of the data (white males) while there isn't any real difference in the difficulty between that subset and another then with better data the other subsets should perform equally well.
I don't think anyone in the "AI community" is against having better data sets.