Let's imagine we have a group of 100 people and we want them to reach consensus on a controverted topic. Let's apply a divide and conquer approach: pair people 1 on 1, and if they can agree (even if the agreement involves different resolutions depending on personal preferences), you join them with another group of 2 and repeat the process.
Now, why doesn't this work? Well, this could work if people's opinions were static and perfectly stated on the first try. But in practice, and even under an ideal setup, people will be changing opinions too fast to ever be able to reach an equilibrium. They might realize that what they previously stated doesn't really match what they believe, or they didn't consider certain things that, when revealed, will now be considered important.
And even more, this is if they were actively trying to reach a consensus while being patient, listening carefully, properly considering other people thoughts, expressing their own in a very precise and faithful way and being able to keep all that information in their heads and prioritize without bias. Which is not realistic.
At this point, we might as well say that the problem is that the more people and points of view you introduce in a discussion, the more precise the consensus needs to be, and precise consensus is unachievable in a continually changing scenario. So, when we believe there's consensus on something, then it must be because either the problem is simple or vague enough, or the consensus itself is vague and imprecise. Which is fine too: vague consensus is not useless, nor it does make sense to reach perfect consensus for everything each time, but that's what we would be lead towards if we all tried to have "better arguments online"... which is never reached anyway at a big enough scale due to the issues previously described.
(yeah, we don't need arguments to reach consensus, you can learn from what others say without agreeing with them, but... I don't even agree with myself, I'm gonna delete th-