Thoughts on "Acceptable Ads"

Privacy and Security Discussions
22

Yes, it was a dumb mistake for Google to hard code diversity parameters. However, the idea of encouraging diversity in AI isn’t necessarily a bad idea. AI has been shown to be biased[1][2], especially racially, and it often lacks training data on minorities. Trying to correct for the drought in representation isn’t by itself a bad idea, just a misguided solution that often fails to address the root problem.

  1. A study by Cornell has shown that LLM’s are more likely to recommend criminal conviction if African-American English is used. ↩︎

  2. Facial Recognition algorithms have shown to be less accurate and worse at detecting people of color because of lapses in training data. ↩︎

3 Likes