Extreme beliefs can easily lead to conflicts between groups. But how do such beliefs arise? Are certain people simply bound to be extremists? Or can significant pockets of extremism arise out of social interactions among moderate individuals?
In other words, the question we are after is how people are radicalized within a society.
Alizadeh et al. (2014) investigate this question using an agent-based model and some well-known theories from psychology.
The idea is to build a model of social interactions in which agents hold various beliefs, and see whether large extremist clusters can arise in such a model without many individuals being predisposed to it.
Let me briefly describe the model first. Suppose there are various agents who randomly interact with each other. Each agent also has a belief on two separate issues. When two agents meet, they can sort of influence or induce each other to change/adjust these beliefs.
In order to understand how such adjustments take place, we need to take a short detour to the world of psychology. Specifically, the phenomenon of cognitive dissonance. Cognitive dissonance arises when a person holds two contradictory beliefs. This results in a sort of mental stress, no one likes to contradict themselves.
Given that humans don’t like mental stress or discomfort, they seek to avoid such situations. So they may reduce dissonance by changing one of their beliefs in order to make it better “fit” with the other. Now, whether two beliefs are contradictory is relatively subjective. So dissonance may be triggered if a person’s opinion on how contradictory their own beliefs are suddenly changes for instance.
At this point, it’s best to provide an example. Suppose there is a rural politically conservative person, and an urban politically liberal one. Both of them think it’s great to support local farmers by buying produce directly from them as opposed to bigger grocery stores. Thus they have opposing beliefs on politics, but similar beliefs on supporting local farmers.
Now, imagine the rural person moves to the city and meets the liberal guy. They will realize that their political opinions are totally different. Then when the rural guy hears that the liberal guy also likes to support local farmers, he might associate this belief with liberal politics and thus dissonance may be triggered. He will feel that being conservative and supporting local farmers may be somehow contradictory. So he may stop going to farmer’s markets to reduce dissonance. A similar process can happen with the liberal person as well.
Let’s get back to the model now. Recall that agents have beliefs on two different issues and that they meet each other randomly. So how do they influence each other’s beliefs when they meet? Basically, the tree below summarizes the answer to this question (constructed by me, not the authors).
The key point here is that if agents are far from each other on one belief, but close on another one, then dissonance may be triggered. It is, however, only triggered if the agents are farther from each other on the differing belief than a “dissonance threshold”. E.g. if the differing belief is your answer to the question whether god exists, then an agnostic and an atheist hold differing beliefs, but they may not consider the other’s belief to be far enough from their own to trigger dissonance. But if one person is an atheist, the other a theist, then distance may be bigger, especially if the theist is heavily religious.
So what are the results from the model? Indeed, it seems extremism is an emergent phenomenon. In other words, extremism can arise from random social encounters of moderate individuals if those interactions follow the rules above.
Somewhat obviously, the higher group differentiation is (i.e. the farther away individuals move from each other when they experience dissonance), the more polarized agents’ beliefs will become. And hence there will be more extremists. Also, the higher the dissonance threshold is (i.e. the more dissonance individuals can tolerate), the lower extremism is (because then fewer people will feel dissonance and move away from each other).
Caveat # 1: the model had an initial (low) number of extremists in the population. So this assumption was used to derive all the results above.
Technically speaking, initial beliefs were uniformly distributed on [-1, 1], and extremists are people whose belief is above .9 in absolute value. As the authors mention, it would be interesting to see how the model fares with different initial belief distributions, especially one where there are (virtually) no extremists. On a side note, I absolutely don’t understand why the authors didn’t try this themselves, it seems like something that can be accomplished by a trivial modification of their code.
Caveat #2: the assumption of random encounters. Basically, in all specifications agents randomly meet each other. But if instead agents would have a social network that exhibits homophily, then they would be more likely to interact with people who hold similar beliefs. This would reduce the probability of meeting an individual with different beliefs compared to a random encounter or random social network situation. And thus extremism would probably arise more slowly, depending on the extent of homophily.
To sum up, extremism is not necessarily an inherent characteristic of an individual, neither do people need to socialize with lots of extremists in order to become one. Instead, extremists can be the product of social interactions among regular, moderate people. This is what the theory is implying. While the model may have its limitations, its predictions seem plausible. And while some of the caveats mentioned – if corrected – can weaken the results, I see no reason why this would make the main results disappear altogether.