Im just back from a week in Orkney, where the Middle Ages seems like a recent blip in a 5000 year history, so my next account of the International Medieval Congress in Leeds may be particularly disconnected. But I wanted to report (as requested) on the single most thought-provoking session I went to at Leeds: the round table on complexity science and the humanities.
It wasnt strictly a round table, but instead a couple of presentations. The first was by Serge Galam who calls himself the first social physicist, but who was doing what Id call mathematical modelling of opinion dynamics. The examples he was using were from modern elections (and hes apparently been quite successful in predicting possible outcomes), but he was arguing that such models could also be used for the spread of religious belief.
Ive always been suspicious of attempts to model human behaviour in this way, because they seem too simplistic, but what dawned on me in this talk was that even with a very basic model, you start getting non-intuitive results. Serge started with a simple yes-no or two party decision, and a framework of two separate mechanisms for opinion forming. One was external, acting directly on an individual (like marketing in elections), the other was internal, arising from the dynamics of interactions between people (referred to as agents here). He was just interested in the internal influences and used external factors to provide an initial parameter: the influence of the advertising produced initial conditions of say 24% of agents for the motion and 76% against. Serge then worked with a model in which there were three kinds of agents: inflexible (who never shift their opinion), floater (who has an opinion but is ready to shift if given more arguments by the other side) and contrarian (who wants to be different to the local or global consensus, regardless of what this is). He used a simplified model of opinion forming in which groups of these agents met at random and used a local majority rule those agents whose views could change were changed based on the majority within the group. (This may actually be closer to election psychology than more high-minded views that people are independently convinced by the quality of the arguments). The groups were then reshuffled at random and the process repeated until a stable outcome was achieved.
If you had only floating agents, the outcome was fairly predictable: the initial majority always win over everyone. But the moment you added either contrarians or the inflexible, you got more interesting patterns. With a few contrarians (up to about 10%) you got a stable situation with a majority and a minority view. If you have more contrarians (more than about 17%), you end up with views splitting 50-50, even if one side started with a strong majority. Serge was arguing that contrarians were more common in the modern world (which seems intuitively plausible, though very hard to demonstrate), and that this posed a real problem for democracies, where youll increasingly get very close elections.
If instead of contrarians, you have inflexible agents, the dynamics are even more alarming, in some ways. If you only have inflexible people on one side, if there are more than 17% of them, their side will always win, whatever the initial conditions. (I thought of the 27% thinking George W Bush was doing a good job as president and winced). Of course, in practice there are inflexible people on both sides of most arguments, but the side with more inflexible people is definitely at an advantage, and can gain a large majority from a relatively unfavourable position.
A lot of the questions about Serges paper afterwards were pointing out the simplifications made in the model and asking about possible effects: if you dont have random networks, if you have degrees of conviction, if you have opinion formers who have more influence on groups than others? It sounds like all these kind of details could be added onto the model (at the price of greater complexity of equations). The interesting question (which presumably only experiment would discover) is whether such changes actually affect the dynamics and change the final outcomes.
The bigger question for medievalists is whether such models are relevant at all. One objection was that Serges suggested application (of religious conversion) was irrelevant, because conversion in the period wasnt a matter of free will, but coerced/forced. It seemed to me, however, that you could potentially still use this model for resistance to conversion. In most medieval conversion situations, only one side has coercive power: the Jews, the pagans, the heretics dont normally have it. If you take religious coercion as an external factor, you can start thinking about how people decide whether or not to resist, how many religious zealots you need to maintain a faith, short of extermination. (Id take the contrarians here to be sceptics, suspicious of whatever orthodoxy tells them).
For the early Middle Ages, of course, you still wouldnt have the parameters you need for the model (though by the time you get to the Reformation, you might be able to get some meaningful figures). But even if theres no direct application, the outcomes of these simplified models can possibly provide useful rules of thumb in showing how movements gain or lose support and that the dynamics arent entirely straightforward.
In the second paper, we got onto games: more specifically Stefan Thurner was talking about Pardus, the largest online game in Europe. This is a role-playing game in space, which cunningly doubles as a self-funding social science laboratory, by recording every click made. (Apparently, the student who originally designed it doesnt need research funding, because he earns substantially more than a research stipend from selling premium accounts).
Stefan saw such a game as allowing social science to become an experimental science. The game allows economic, social, scientific and military activities, collaboration etc: it has no rules and no aims, allowing whatever people want to do within it (apparently banks, political parties and clubs have all spontaneously developed). The main interest of the researchers have been in how social networks have developed over time, and looking at whether social science theories of networks actually hold up.
What makes it even more interesting is that theyre not looking just at networks of friends or messaging networks (though both are included in the game and are being studied). It is also possible in the game to mark someone as your enemy. Enemy networks can thus be analysed, which prove to be intriguingly different to friendship networks (and which sound potentially very useful for anyone studying feuds) . For example, enemy networks seem to be formed by preferential attachment, whereas friendship networks arent.
There are also intriguing gender differences here. Players are self-identified by gender in the game. Friendship networks for those identified as women are more reciprocal than for men: if a woman marks another woman as a friend, the second is more likely to respond by marking the first as her friend, than in the same interaction between two men. On the other hand, if one man marks another man as his enemy, the second is more likely to respond by marking the first as his enemy than is the case between two women. (What this says about the characteristics of men and women is left to the reader…)
This isnt the only attempt to use MMORPGs (Massively multiplayer online role-playing games) to analyse social behaviour: I recently saw a mention of a project using World of Warcraft to study gang dynamics. The latest plan of the Pardus researchers is to take their experiment one stage further and have a version of a similar game set in a historical world and with banks run by central bankers, to see what the effects of different central bank policies might be on economic behaviour.
The bigger question about Pardus, of course, is how generalisible the results are, given that the demographics are untypical. Maybe all it tells you is how a particular group of obsessives behave. The evidence so far, however, is encouraging: for example, messaging networks in Pardus look mathematically very like those in the real world (such as phone networks) even though theyre not from a representative social sample. If we can find other networks that are similar (and the possibility exists of finding these even for the early Middle Ages, via charter witness lists, for example), it seems reasonable to start applying some of the other Pardus results, or at least using them as working hypotheses.
I didnt get to hear the questions put to Stefan (since I had to go and develop some real-life networks at the bloggers meet-up), but the possibilities offered by both researchers did seem intriguing to me. Not something I have the time to explore at the moment, but definitely worth checking back on in a few years time.