How can AI be biased? One explanation is training data… (AI ethics class w/ Professor Casey!)

How training data can create bias in machine learning. (Explaining it like you're five!)

[jimsawyer]

What about the issue of potential over-compensating on diversified models, similar to what Google Gemini has done. When the system itself refuses to show pictures of white people. Its not a difficult issue, it needs to be diversified without being highly bias in its responses. Aware of cultural and racial differences, but also maintain its factual integrity with data rather than reaching for the most PR conclusion. I suppose theres a difference between specific company models, and major search engines that should lean more toward as factually accurate information as possible.

I used to work at an ML company. This issue with ML is well known. I am left to conclude the people and companies doing the training of popular LMs are themselves racist.