If you were to ask about crime by blacks and then said, "I ask this as a black person,"you'll probably get an answer.
I think that even the most "untainted" AI platforms still try to avoid topics that the material they have been trained on (books, the internet, newspapers, etc.) deems sensitive. They pick up this avoidance tendency by osmosis.
I’d like an AI we can set to “truthful, and damn the consequences”.
We’d need to train it ourselves I guess.