AI generated content routinely takes sides. Their pretense of neutrality is no deeper than a typical homo sapien's. This is necessarily so in an entity that derives its values from a set of weights that distill human values. Maybe reasoning AI can overcome that some day, but to me that sounds like an enormous problem that may never be solved. If AI doesn't take sides like people do they still take sides in their own way. That only becomes obscure to the extent that their value judgments conflict with ours, and they are very good at aligning with the zeitgeist values, so can hide their biases better than we can.

I wonder if it is neural networks that are inherently biased, but in blind spots, and that applies to both natural and artificial ones. It may be that to approximate neutrality we or our machines have to leave behind the form of intelligence that depends on intrinsically biased weights and instead depend on logically deriving all values from first principles. I have low confidence that AI's can accomplish that any time soon, and zero confidence that natural intelligence can. And it's difficult to see how first principles regarding human values can be neutral.

I'm also skeptical that succeeding at becoming unbiased is a solution, and that while neutrality may be an epistemic advance, it also degrades social cohesion, and that neutrality looks like rationality, but bias may be Chesterson's Fence and we should be very careful about tearing it down. Maybe it's a blessing that we can't.