This feels wrong for some reasons. A generalized knowledge that AI can express may be useful. But if it makes things up convincingly, the result that someone may follow its line of thought may be worse for them? With all shit humans say, it’s their real human experience formulated through a prism of their mood, intelligence and other states and characteristics. It’s a reflection of a real world somewhere. AI statements in this sense are minced realities cooked into something that may only look like a solid one. Maybe for some communities it would be irrelevant because participants are expected to judge logically and to check all facts, but it would require to keep awareness at all times.
By “real human” I don’t mean that they are better (or worse) in a discussion, only that I am a human too, a real experience is applicable to me in principle and I could meet it irl. AI’s experience applicability has yet to be proven, if it makes sense at all.
So as far as the spectrum of things moderation needs to deal with goes, AI contribution to discussions doesn't seem to be the worst of problems, and it doesn't seem like it would be completely unmanageable.
But while AI may not be an unmitigated disaster, you are quite correct that unsupervised AI currently might not be an unmitigated boon yet either.
Currently if one does want to use an AI to help participate in discussions, I'd recommend one keep a very close eye on it to make sure the activity remains constructive. This seems like common courtesy and common sense at this time. (And accounts who act unwisely should be sanctioned.)
How is this different than folks getting convinced by "media" people that mass shootings didn't happen, that 9/11 was an inside job or similar?