Dividing humans Reducing useful discourse Empowering those with the wealth to control it
At the very least, attempting to automatically annotate anything AI generated would allow each reader to make their own decisions. The annotation may not be completely accurate, but would prevent human responses being lost due to false positives - as AI becomes more capable of mimicking humans, we run the risk of discussions becoming echo-chambers overrun by AI.
(The other solution harks back to the calls for removing anonymity and requiring some sort of identification mechanism, but that then leads to other problems)
If AI contributes well enough to be acceptable to most readers, perhaps it should be allowed. So long as it is within the rules that humans must also observe. Otherwise, what is being achieved, other than oppression ?