3 ms·

Reading the blog post and the Twitter thread. Wow. 10 years ago it would have been seen as a comical blunder of a stupid AI. Something to be fixed for sure, bu

Reading the blog post and the Twitter thread. Wow. 10 years ago it would have been seen as a comical blunder of a stupid AI. Something to be fixed for sure, but not a Serious Social Issue by any means. Nowadays, it apparently warrants several follow-up articles in the mainstream media, "social" commentary and 3,309 retweets. From the blog post: “The bias of the Internet reflects the bias of society,” she said. In some cases - yes, but this one seems more like society deliberately projecting human motivations onto a primitive algorithm and jumping to conclusions about what its errors "really" mean. - For people who disagree, here is a scenario your might want to consider. Imagine that you built an image tagging service. Imagine that someone found a glitch in your service that they consider offensive. Imagine them tweeting about it (before or instead of contacting you directly) and getting a similar kind of reaction, complete with extensive social commentary and media coverage. Nice, big crowd of people using your company and your service as a convenient example of things-that-are-wrong-with-our-society. How would you feel in that case?

jnky 11y ago
I couldn't agree more. I'm all for not discriminating people, but on the other hand I firmly believe that technical progress should not be second to people's feelings.
taneq 11y ago
The issue here is that race relations in the U.S. are so convoluted and have so much history that it's literally impossible to tell what would be considered "racist" without a broad understanding of the culture and a comprehensive list of past racial slurs and grievances. Witness the KFC ad (https://www.youtube.com/watch?v=ZaIhf41ctkM https://www.youtube.com/watch?v=ZaIhf41ctkM) which was broadly labeled as "racist", despite the fact that the 'black people love fried chicken' stereotype is (as far as I know) only a U.S. construction.
- jnky 11y ago
  > The issue here is that race relations in the U.S. are so convoluted and have so much history that it's literally impossible to tell what would be considered "racist" without a broad understanding of the culture and a comprehensive list of past racial slurs and grievances. I understand, but isn't the logical conclusion that one cannot make a piece of technology (like an image caption bot) that is unaware of, for instance, such complex racial relations? And if so, should technological progress really be hampered by people's sense of outrage, even in the absence of malicious intent?
  - taneq 11y ago
    The logical (to me) conclusion is that people need to unbunch their panties and stop looking for things to be offended about. I don't think it's reasonable for anyone to expect an AI system, at our current level of technology, to have a complete understanding of every nuance of human pique to the degree where it will never do anything which could be interpreted as offensive by anyone. Hell, that's a far higher bar than we humans can hope to meet in today's 'outrage culture'.
gengkev 11y ago
Yes, I think people overreacted a bit. But part of the backlash was because if these algorithms were better trained on a diverse set of faces, they might not have made that mistake. I think that's a fair criticism.
- colllectorof 11y ago
  > if these algorithms were better trained on a diverse set of faces, they might not have made that mistake. Is this assumption based on something specific? I.e. are there good reasons to believe that they weren't trained on a diverse set of people and that such training would prevent this error from happening?

jnky 11y ago

I couldn't agree more. I'm all for not discriminating people, but on the other hand I firmly believe that technical progress should not be second to people's feelings.

taneq 11y ago

The issue here is that race relations in the U.S. are so convoluted and have so much history that it's literally impossible to tell what would be considered "racist" without a broad understanding of the culture and a comprehensive list of past racial slurs and grievances. Witness the KFC ad (https://www.youtube.com/watch?v=ZaIhf41ctkM https://www.youtube.com/watch?v=ZaIhf41ctkM) which was broadly labeled as "racist", despite the fact that the 'black people love fried chicken' stereotype is (as far as I know) only a U.S. construction.

jnky 11y ago
> The issue here is that race relations in the U.S. are so convoluted and have so much history that it's literally impossible to tell what would be considered "racist" without a broad understanding of the culture and a comprehensive list of past racial slurs and grievances. I understand, but isn't the logical conclusion that one cannot make a piece of technology (like an image caption bot) that is unaware of, for instance, such complex racial relations? And if so, should technological progress really be hampered by people's sense of outrage, even in the absence of malicious intent?
- taneq 11y ago
  The logical (to me) conclusion is that people need to unbunch their panties and stop looking for things to be offended about. I don't think it's reasonable for anyone to expect an AI system, at our current level of technology, to have a complete understanding of every nuance of human pique to the degree where it will never do anything which could be interpreted as offensive by anyone. Hell, that's a far higher bar than we humans can hope to meet in today's 'outrage culture'.

gengkev 11y ago

Yes, I think people overreacted a bit. But part of the backlash was because if these algorithms were better trained on a diverse set of faces, they might not have made that mistake. I think that's a fair criticism.

colllectorof 11y ago
> if these algorithms were better trained on a diverse set of faces, they might not have made that mistake. Is this assumption based on something specific? I.e. are there good reasons to believe that they weren't trained on a diverse set of people and that such training would prevent this error from happening?