This week in my Network Epistemology class we looked at models of argumentation in social networks. "Argumentation" here is not to be taken negatively -- like when you say someone is argumentative. Instead, it's meant as the method of conversation where you exchange ideas.
Most models of information spread and belief aggregation look at how one belief is affected by social influence. For example, models of probabilistic pooling all assume we are "pooling" a single belief with others.
But in reality, of course, we sometimes interact by arguing with one another. I tell you what I believe and I also give you *evidence* for that belief. That evidence might be something that you already believe, and so I might convince you through argumentation.
This is a very important area of study, but remarkably little formal modeling has been done on the influence of social structure on argumentation. We read what little I know that has been done in this area.
We started with applications of Dung's framework. Dung's framework is remarkably simple, but also very powerful. It was initially designed for AI applications, but can be used for our purposes as well. Here's Dung's original paper.
dai.fmph.uniba.sk/~sefranek/kri/…
dai.fmph.uniba.sk/~sefranek/kri/…
Dung's framework uses graphs to represent arguments, where each node is an "argument" and edges represents a relationship of "defeat" or "attack." Consistent beliefs are those that can defeat every potential defeater. 
But that framework doesn't allow for some meaningful types of argumentation relations. We read this paper by Šešelja and Straßer which extends Dung's framework to account for the scientific notion of "explanation."
link.springer.com/article/10.100…
link.springer.com/article/10.100…
They show how they can reconstruct the historic debate about continental drift in their framework. And they try to use it to account for some philosophy of science concepts like "explanatory depth."
We discussed what various purposes such a framework might serve, including a method for historical analysis and for empirically comparing argumentative strategies. But for this class, we were looking at models of argumentation. For that...
We next read this paper by Borg et al which creates an agent based model of argumentation in the context of scientific debates.
ssoar.info/ssoar/bitstrea…
ssoar.info/ssoar/bitstrea…
The paper largely focuses on the question of when communication is helpful or harmful, and finds that communication is largely helpful. The paper is aimed at a comparison with other models (by me and by Grim et al) that showed communication could be harmful.
Because the argumentation framework is very different on many dimensions, we were a little unsure of what was driving the results. The class concluded that the model was interesting, but more exploration was needed to determine what drove the differences.
Finally, we looked at a different framework for modeling argumentation, based more directly in deductive logic, due to Gregor Betz. (We read the first chapter in his book, linked here.)
springer.com/gp/book/978940…
springer.com/gp/book/978940…
While Betz doesn't focus on networks, he does look at distinctive social questions. What strategy for arguing with one another makes the group most effective at (a) coming to a consensus and (b) converging on the truth?
In the introductory chapter he provides a nice overview of the various strategies and their effect on the two goals. Many of them make intuitive sense. For example: it's best to use an argument where the interlocutor accepts the premises.
One conclusion that is interesting is that there is sometimes a conflict between consensus and truth (goals (a) and (b)). Some argumentative strategies produce one at the cost to the other. And so we should be careful not to substitute one goal for the other.
I think this framework is worthy of study, and I think it would be interesting to apply it to networks. (My secret hope was that a student would be motivated to take that on... we'll see if it worked.)
In class, we talked about the limitations of the frameworks for modeling non-deductive debates, like those in science. And whether Betz's conclusions were all that surprising (and whether that mattered).
Overall, I think modeling argumentation is a very exciting area which is wide open for anyone who wants to work on it. I think there are a plethora of interesting, unanswered questions just waiting for someone to come along with a good model.
