I think the general view is it’s impossible to not hold biases, but biases that lead to harm are bad. I don’t consider myself progressive but I do believe biases that are harmful to people should not be reflected in the tools we build, in so far as we can. I don’t feel people shouldn’t be allowed to hold biases that are harmful to others, nor that they should be disenfranchised for it. But it’s not a good thing to be biased negatively, and having a bias that is positive is a good thing. The definitions of of good and bad for me are harmfulness to others. A bias that gay people should be excluded from work and social institutions is harmful to a gay person, and I’m constructing that as a negative bias due to the harm. A bias that people shouldn’t exclude gay people from work or social institutions is not harmful to the person who holds the negative bias because it doesn’t involve them and it only affirms a positive right. When constructing a tool, I think it’s probably a good thing that it support positive biases and diminish negative biases. This also doesn’t hurt the people who hold a negative bias. It just doesn’t support them in that bias.

But, it’s just computer code and thus can be customized and adapted by anyone with the technical ability, and soon we will see the antisemitism chatgpt and the XYZ bot the espouses all sorts of ideologies, some that are very harmful to many people.