Yay new case study? :-\
(1) Consider the use cases for this. Nothing in readme about how this shouldn't be used, privacy, etc.
(2) Examples in the comments of running this on images of Black faces, and the result coming out obviously white. I question the training data. :(
(1) Consider the use cases for this. Nothing in readme about how this shouldn't be used, privacy, etc.
(2) Examples in the comments of running this on images of Black faces, and the result coming out obviously white. I question the training data. :(
Just had a quick look at the research paper, and did a control-F for "ethics", "privacy", and "bias". 👎
I'm not saying don't build things! But all I want for christmas is some ethical reflection.
I'm not saying don't build things! But all I want for christmas is some ethical reflection.
Since I noticed that paper was published at CVPR, here's an article from a CVPR workshop a few years ago from a couple of my collaborators: "Designing a Moral Compass for the Future of Computer Vision using Speculative Analysis" openaccess.thecvf.com/content_cvpr_2…
Excellent point... a side effect of the likely horrific training data is that at least if this technology is being used for evil, it will be very bad at identifying people of color?
More examples of the racial bias in this example from yesterday. @osazuwa points out how demonstrating ML's limitations helps us understand how it can be made useful. Remember that it's also so we can understand and mitigate harms.
