We need to have user-generated labels attached to comments: sarcastic, flamebait, domain expert, etc. And then go even more meta and have algorithmically and/or privileged users affirm the label. Lets throw in some labels for common cognitive biases, statistical errors, logical fallacies and eagerly slap them on all popular offending comments. And then go all in on multivariate testing of mutations of algorithms optimizing for best (your definition may very) discourse.

If such thing would happen and became popular, someone-is-wrong-on-the-internet crowd would be puking rainbows all day long.

I guess it is a question of time this becoming feasible.

BTW, I wasnt sarcastic.