I wouldn't say it is impossible. It is difficult to build a smart model with a given worldview, but trained on a corpus of text written by people with a different worldview.

The kinds of people administering wikipedia and working in academia tend to have a certain worldview, and will work backwards to justify that worldview, censor, etc. The idea that "reality has a liberal bias" is pretty fatal mistake because you are conflating reality with the opinions of the studiers of reality.