The most obvious downside to public discourse is other people introducing their intentional confrontation. It's what drives a lot of people to limit access. Brandolini's Law (aka the "bullshit assymetry principle") is brilliant for its concise insight. When someone argues in bad faith, they waste precious resources of time and attention from others, and we collectively become poorer for that distraction. I very much like the use of filters and blocking on Mastodon as a way to individually tame these distractions. When someone decides to be an annoyance, though, bullying others in situations they seek out deliberately, then they're acting more like a metastatic cancer cell, excessively using group resources for their own private purpose. Such problems require systemic responses, not individual ones.

I'm not sure how to train AI on recognizing these specific interactions and actors. One of the worst scenarios I can imagine is simply teaching AI (with no counter-arguments of the particular benefits of discord) that it's appropriate to eradicate all sources of irritation simply because they're annoying. Speaking as someone who would rather live in isolation, humans are so very mind-bogglingly annoying. :)

P.S. The "Letter from a Birmingham Jail" by MLKJr opens with the first paragraph exactly on this important topic of intentional discourse versus intentional distraction.

P.P.S. Such a brilliant man. "Just as Socrates felt that it was necessary to create a tension in the mind so that individuals could rise from the bondage of myths and half truths to the unfettered realm of creative analysis and objective appraisal, so must we see the need for nonviolent gadflies to create the kind of tension in society that will help men rise from the dark depths of prejudice and racism to the majestic heights of understanding and brotherhood." This is the thinking that we should use to train AIs as the best (hopefully even better) version of ourselves.