Post APZHjbSVdHwksvLOzo by maria_antoniak@sigmoid.social
 (DIR) More posts by maria_antoniak@sigmoid.social
 (DIR) Post #APZHjbSVdHwksvLOzo by maria_antoniak@sigmoid.social
       2022-11-13T16:34:36Z
       
       0 likes, 0 repeats
       
       I've been reviewing, and one thing on my mind again is that as a reviewer, I get more traction with the other reviewers by framing ethical issues "technically" rather than as "harms." And it's true that ethical problems very often go hand in hand with technical problems. If you train a "gender" classifier on names, there will be serious accuracy issues that are rarely probed or described. And if you make that mistake, you likely make other technical mistakes too. But...
       
 (DIR) Post #APZHjc1xVUIqerviFc by maria_antoniak@sigmoid.social
       2022-11-13T16:39:52Z
       
       0 likes, 1 repeats
       
       ...this also makes me think again about this paper from last year's FAccT, and how the values described connect to reviewing. I really think it's worth reading this paper if you're reviewing ML papers."The Values Encoded in Machine Learning Research" (Abeba Birhane et al.)https://facctconference.org/static/pdfs_2022/facct22-14.pdf(s/o to @emilymbender who I think has made a similar point about ethical/technical issues going hand in hand)