How can we take the dynamics of a small group discussion and scale it up to a larger group?
This may not be possible. See CommunityMayNotScale and ConnectedGraphSquaringProblem.
In a large discussion, there may be too many comments for everyone to read. There are various techniques for dealing with this.
Each comment can be rated by other people, and then people may choose to filter the comments they view so that only the highly rated ones are shown. The news website SlashDot uses this model.
A collaborative editing process can condense, summarize, and group comments when the volume of comments begins to get large. This is what happens on a wiki.
How should a large group be moderated?
One answer is to distribute the moderation tasks among a large number of users. Examples:
The news site SlashDot distributes the task of numerically rating comments to random subsets of users.
Most wikis rely only on social norms to guide the group. There is nothing to prevent any wiki user from changing any wiki content however they like – but if the change goes against the grain of the community, it will quickly be reverted.