I got reminded of Salmon's Five Stage model the other day. I'm not sure I'm totally happy with it. I remember reading an ALT-J paper once too which expressed doubts about it, I think mostly on empirical grounds, but I don't have access to the journal where I am at the moment.
Now in real life, most relationships with other people do naturally follow the Five Stage Model, though they may of course stop at one of the stages. It's very hard to go straight to discussion without some sort of social interaction first. Not that it's impossible - people like politicians and academics do often dive straight in, and at work you sometimes find yourself discussing the best way to solve a particular problem with somebody you've never met before. But totally online, the most common direction I've found is the opposite one. You get involved in an interesting discussion with someone new or you find yourself collaborating with someone when playing World of Warcraft, and because of that you start interacting socially.
So why don't we just socialise with random people online, whereas we would be happy to at a party? I guess this is probably partly because people that you meet in real life are likely to be more interesting to you than average because of the context that you meet them in. But, also I suspect because real people impact our brains in so much more powerful a way than words do.
Obviously things get more interesting still when online and real life mix and you can see that in education this could often be the case. I think the sequence 'Brief real life social -> online discussion / online social -> real life social' often seems to work quite well.
I'm not sure how this all applies to education though. I guess Gilly Salmon's model works mostly because you need a really really low barrier to initial participation at the start. Social stuff provides that, though there are other ways too. I must go and find out about the exact research behind the model and what else she tried and rejected.