This page is a permanent link to the reply below and its nested replies. See all post replies »
ninalanyon · 70-79, T
I suspect that much of the training data is written by men and also that quite a lot of it is technical in nature. Not to mention that far more text is available in English than any other language so that introduces subtle biases as well. It's not so much missing humanity as missing parts of humanity, the parts that haven't written texts that ended up in the training set.
For instance, mothers-in-law will be represented in AI not by what they say and do but by the jokes that men make about them.
It reflects the texts that it was trained on and will have all the biases that the training data had in addition to extra bias introduced by those who program and control it.
For instance, mothers-in-law will be represented in AI not by what they say and do but by the jokes that men make about them.
It reflects the texts that it was trained on and will have all the biases that the training data had in addition to extra bias introduced by those who program and control it.


