Conversation

Notices

1. Embed this notice
   Shriram Krishnamurthi (shriramk@mastodon.social)'s status on Saturday, 28-Oct-2023 03:22:29 JST Shriram Krishnamurthi

   A terrific article about how smiling varies across cultures (as all us immigrants recognize) and how AI erases that: "AI and the American Smile: How AI misrepresents culture through a facial expression". HT Chinmay Kulkarni
   https://medium.com/@socialcreature/ai-and-the-american-smile-76d23a0fbfaf

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Saturday, 28-Oct-2023 03:29:53 JST Paul Cantrell

     in reply to

     @shriramk
     I’m interested (and surely somebody out there is doing this work!) in the question of whether we can use the output of large-dataset ML systems as useful measures of social / cultural patterns. For example, if gen AI insists doctors are male and nurses are female, can we treat that not just as empirical evidence that the AI is sexist, but that •society• is sexist? Can such observations be compelling evidence for social theories under dispute?

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Monday, 30-Oct-2023 11:28:52 JST Paul Cantrell

     in reply to

     @shriramk
     Exactly. And I think that’s already plenty compelling as a casual argument in conversation and public debate, but can we make it robust enough to be dispositive evidence in social science research? What are the pitfalls? What methodology avoids them? What hurdles would such research need to clear? What if anything could make it publishable?

   • Embed this notice
     Shriram Krishnamurthi (shriramk@mastodon.social)'s status on Monday, 30-Oct-2023 11:28:54 JST Shriram Krishnamurthi

     in reply to

     • Paul Cantrell

     @inthehands Yeah, really good point. "It just reflects social bias" — oh, okay, great, so that means the bias is real!