4 ms·
The quote you cited says "no evidence the saliency model explicitly encodes male gaze" emphasis mine. Your re-phrasing however changes the meaning in your quest
by cryptoz 5y ago
The quote you cited says "no evidence the saliency model explicitly encodes male gaze" emphasis mine. Your re-phrasing however changes the meaning in your question, "historical biases not inherent in the model".
Explicitly encodes != inherent in. The difference between the two might answer your question. You can solve for something that is inherent in a system (ie, this), but was not explicitly encoded. They are trying to say it isn't their fault specifically, they find no evidence this was done on purpose or in a targeted way, but that it happens as an accident / lack of testing on the current model.
- stolenmerch 5y agoThey are very clear about 'intentional' vs. 'unintentional' harm in their judging metric[1], so I get that I think. However, there is no possible way a saliency model could crop an image without someone reading bias into it. E.g. only cropping a jersey, only cropping a face, not cropping at all, etc. can all be signs of historical bias. At some point it just becomes a rorschach test for the reviewer. Not saying discrimination isn't happening or can't happen, just saying this very tenuous and weak connection to "historical bias" interpretation lacks a certain rigor in a bug bounty program or model optimization. [1] https://hackerone.com/twitter-algorithmic-bias https://hackerone.com/twitter-algorithmic-bias