Researchers created an artificial society to find the causes of religious conflict

To understand religious warfare, you could study the hundreds of historical or ongoing world conflicts that center on religion. Or you could program an AI to mimic human psychology and generate artificial societies, and then run it millions of times under different variables.

That’s what a team of researchers from Oxford University, Boston University, and the University of Agder-Norway did, in a study published last week in the Journal of Artificial Societies and Social Simulation. The group drew on theories from cognitive psychology to create AI agents that would mimic human behavior, including identification with a particular religion and alignment with group identity. After programing the agents with different ages and ethnicities, and running it millions of times, they found that conflict most often occurs when a majority religious group constitutes around 60% of the population, and a minority group makes up 40%.

“Our work shows that given two xenophobic groups that are close to equal in size, and where individuals from one group can regularly encounter others from the opposite group, prolonged periods of mutually escalating anxiety can occur,” says Ross Gore, a professor at the Virginia Modeling, Analysis & Simulation Center at Old Dominion University. In these anxious periods, people tend to “use religion as a calming mechanism,” Gore says, which leads communities to become more religious. And people are more prone to violence and anxiety when their sense of individual identity is tied up with a group, and that group is then insulted or attacked by outsiders. 

One of the researchers, LeRon Shults, a professor at the University of Agder, says he was surprised at how little violence there was. Overall, the model led to conflict in just under 25% of the scenarios.

Of course, even though the researchers ran their experiment millions of times with millions of different conditions, they can’t possibly include every possible variable. “There’s always some improbable person or event you can’t predict,” says Shults. “By doing millions of runs under different combinations of variables, including environmental stressors such as natural hazards and even predation, you may not be able to have a Trump or somebody, but you can explore huge numbers of possible types of interactions and all kinds of people, and say ‘here are the most probable conditions under which you get inter-group conflict.”

Overall, Gore says, independent of large-scale environmental factors, such as natural hazards, conditions such as the size of the two groups and the level of interaction, are statistically significant predictors of conflict.

“This is significant because it means that there can be implications in terms of policies to reduce mutually escalating anxiety at an individual level,” he says. For example, creating barriers between those in different groups would reduce encounters, mean anxiety and conflict is less likely to occur. Gore suggests both physical barriers between groups and greater distance between them would reduce interactions and so the likelihood of conflict. Of course, in practice, such a policy would create moral concerns about separating and confining groups based on identity, as well as whether dividing groups based on religion should be a goal in any society. But according to the artificial simulation, at least, this alone would help prevent conflict.