Modelling how religious tension escalates

A Generative Model of the Mutual Escalation of Anxiety Between Religious Groups

So I made the time to read the original paper. As I suspected, the press release has some major flaws, to put it mildly : https://www.sciencedaily.com/releases/2018/10/181031080630.htm

This is a numerical sociological model studying how tension can develop between religious groups. It is categorically not a study of violence, because the effects of violence are entirely missing from the simulations. Nor does it use "psychologically realistic AI", that's completely wrong. It doesn't use any kind of AI at all, it uses an agent-based network model. No problem is solved, no goal is achieved, they simply run simulations to see what happens in certain conditions. It's no more intelligent than a n-body gravitational simulation. As for psychologically realistic, the model is informed by psychological findings, but of course it's far from complete. It's certainly far too much of a stretch to claim (as in the press release) that nothing like this has been done before - that is pure nonsense.

What they actually do is pretty simple. They have this 2D world populated by agents belonging to two different religious groups. Within each group, agents are connected in smaller groups (who may be connected to other small groups), essentially representing the people they directly interact with. There are no direct connections between the different religions.

The agents spend their days wandering around (randomly, I think). If they're unlucky, they might encounter something dangerous, e.g. a natural, inherently dangerous hazard, or a perceived social hazard if they become too close to a member of the other religion. This increases their anxiety. If they become too anxious, they and other members of their sub-group (and beyond if necessary) who are also overly-anxious stop moving, sit down and perform a nice quiet religious ritual that decreases their anxiety. All members persist in ritual until everyone is below their own individual anxiety threshold. Since everyone's threshold and anxiety level is different, this means that some members go off feeling happy as Larry but at least one is still so anxious that any more hazards will trigger the need for another ritual. Anxiety is also allowed to drop slowly over time without rituals.

And that's basically it. What they monitor is escalating levels of anxiety. There's no violence in here at all. That's tangential at best, since in the real world, they say, escalating mutual levels of anxiety could be a trigger for violence, but they don't examine the effects violence itself would have. They study how sometimes anxiety is increased only among one of the two groups, depending on the population size. Usually anxiety only increases among the smaller religion, because it's more likely that they'll encounter someone from the dominant religion. They find that periods of mutual escalation occur significantly less frequently (24% of the time) than chance (43% of the time), though it's not clear to me how they calculated this probability.

There's one other aspect here : the agents have varying levels of "religiosity". They define this as a combination of a tendency to ascribe supernatural causes to events ("anthropomorphic promiscuity", to use their unnecessarily complex jargon) and a preference to engage in religious rituals ("sociographic prudery"). The religious rituals make participants more religious, but as far as I can tell, the anxiety threshold is completely independent of the religiosity, and this can only ever increase with time. The somewhat more interesting aspect of this is that the AP scores of ritual participants can increase depending on the average value of the ritual members, so it can vary in a complex, local way.

As they state very clearly, there are no significant new findings here and there are many uncertainties in the parameters used. The novelty is the model. Agent-based networks look like fascinating things; there's no need to resort to overblown press releases to make them interesting. That you can get meaningful results at all from such simple approximations is, in my opinion, really neat.

Via Joe Carter.
http://jasss.soc.surrey.ac.uk/21/4/7.html