In this short blog I want to have a look at why social media are so ideal as the battleground for radicalisation and the intervention of outside parties in socio-political processes.
In this post I want to discus a modelling exercise I recently did and which I reported with technical details here. But in this blog I want to leave out all the technicalities and just discuss the ideas behind the model and the basic results of applying those ideas to a simple example.
Radicalisation and populism
The radicalisation of young people in the West to the cause of organisations like Daesh or far-right terror groups has been driven to a significant extent by online materials. Both groups of extremists devote considerable time and resources to polish up and sustain their online presence. Similarly people in early stages of radicalisation contribute to this by sharing posts and site-addresses where such materials are available.
But also the wider populist movements have deployed these tools. In recent elections we have seen the influence of these technologies and the attempts by companies such as Cambridge Analytica and Aggregate IQ to use them for influencing election outcomes. These different players however do of course have some overlap, as they also have with an older ‘underground’ of conspiracy theory and fake-news sites with political agenda’s.
Now typically the assumption in the mainstream media is that this popularity of this medium is the result of its accessibility and the fact that so many users are available there to send information to. However the Cambridge Analytica scandal showed very clearly that although the ease of access is relevant, even more relevant is the fact that these technologies allow targeting of messages. Many commentators have expressed doubt about whether a few Facebook adds would really be sufficient to change someone’s opinion to the degree that it could affect election outcomes. But I believe the error in that argument is the assumption that ‘convincing someone’ is the goal of such interventions.
So why not make this a modelling exercise?
The approach I took is really quite simple. I assumed that the citizens in a community are connected to some of their fellow citizens by links which indicate that they care about that other citizens political opinion. Now suppose this community faces a decision on some kind of a societal issue B. The citizens of my model can have either a positive political position on this (i.e. be in favour) or a negative one (i.e. be against it) but I did not assume that there is any restriction on how much positive or negative their opinion was. I further more made the assumption that each citizen had preferences to
- have their opinion agree with the evidence they held (or believed to hold);
- have an aversion against very extreme positions if it were just up to them individually and have agreement with the citizens about whose opinions they care;
These ideas can be put into a specific mathematical form and, with all the buts and ifs that such modelling brings with it, it is fun to see whether the citizens in the model can agree with one another and what could keep them from achieving consensus.
With the preferences as stated above it is possible to see whether there exists an “optimal” opinion that each citizen could choose so as to maximise their own ‘well-being’. In the simple-minded model I chose this is indeed the case as long as the citizens do not assign to much value to the opinions of the others they care about. If they do assign to much value to the opinion of others then in this model what happens is that the citizens in their community of likeminded citizens become increasingly ‘happy’ the more they share an evermore radical opinion. You might say: if they care to much about the view of others than this will overwhelm their aversion to extreme views.
If they place little value on the opinions of others, then the citizens in my model will simply follow their “evidence” whereas the stronger they value the views of their peers the more they will also take the evidence their peers hold into consideration. It is only when the strength of their preference for consensus exceeds a critical value that they are at risk of ending up in a spiral of radicalisation.
Now in the real-world radicalised individuals are often described as “isolated”. In a future version of my little toy-model I will try to build in which choices my model-citizens make when confronted with people they vehemently disagree with but having the possibility to change their connections. What you might expect is that what makes radicalising people “isolated” is that they cut connections with those of whom they expect that they would not agree with their increasingly radical positions. In addition to that there is of course the effect that online connections are cheap to make and break. So “unfriending” a couple of people that disagree with you does not come at a great cost and these links are easily replaced by adding a few that agree with you.
Another question that you can ask a model like my toy-model is: what happens if the citizens in the community have an opinion that is not optimal for them? Well, if you make the assumption that these citizens might be inclined to revise their views every now and then in a direction that makes them ‘happier’, then you can calculate (in the simplified model-world) whether or not they can reach consensus.
Typically what you find resonates with what I discussed earlier: if they do not place to much weight on the opinion of others, yes then they typically will reach consensus over time. My model will not tell you whether they agree on being in favour or against B, only whether they will agree whichever which of these it is.
In fact in the model you can see that even if there is no evidence whatsoever (or if nobody believes any of the available evidence) these model-citizens can still achieve consensus under normal circumstances.
Another question this model will tell you something about (as long as your question is about the model world!) is whether these agents could also reach a consensus if they seek to optimise the sum of everybody’s ‘happiness’. The answer to that is again yes, under again the same conditions as before. But there is an interesting thing that we can see from comparing the two scenarios: (1) citizens seeking to optimise their individual happiness taking into account they care about the opinions of some others and (2) citizens seeking to optimise the collective happiness. The graph below shows you the distribution of opinions of in such a socio-political network of citizens.
The “Katz centrality” is a measure for the strength of opinion in the presence of some evidence available to all the agents. The dark-grey distribution is for those citizens who only optimise their individual well-being, while the light-grey distribution is for those who optimise the collective well-being.
What you see is that the ‘social-optimisers’ are more narrowly distributed around a slightly higher degree of conviction. The social optimisers are better as avoiding extremes as well as avoiding indecision and somewhat firmer in the conclusions they draw.
Interestingly the model also tells you something about how an outside disruptor could seek to disrupt the process of finding consensus, encourage radicalisation and ultimately destroy the community. The model specifies four goals of ‘attack’ such an outside disruptor could pursue:
- attack or undermine the institutions responsible for the social optimization process of symmetrizing the connectivity between agents;
- ‘spin’ the evidence, the credibility of the evidence and/or the credibility of the evidence providers, because the ‘spun’ evidence can make it impossible for the community to find the consensus;
- stimulate network re-wiring that removes a crucial set of connections so as to make consensus impossible;
- reduce the risk-averseness of the citizens to gambling on extreme political positions;
In order to work towards these aims an outside disruptor would essentially need only three tools, a simple model like mine suggests,
- Easy access to the information-flows reaching every citizen;
- Enough knowledge of the network connecting the agents;
- A ‘normalising’ message concerning politically extreme views;
Real-world interpersonal relationships are costly compared to virtual connections via media such as Twitter, Facebook, Instagram, etc. It is not unlikely that in the past few decades the effective connections in the network among citizens in developed economies has seen a significant addition of ‘cheap’ virtual connections. The loose data-policies of social-media companies as well as the naivete of citizens using these networks has made it comparatively cheap for an exogenous disruptor to actually acquire the three aforementioned tools.
2016 and beyond
With this in mind it seems to me that the two key political disruptions of 2016, the Trump election and the Brexit referendum, were an accident waiting to happen. Goals 2 and 4 are essentially what targeted messaging is all about when deployed by an outside disruptor. The disruptor does not need a coherent message but instead can send out conflicting messages normalising opposite yet radical views as long as they are targeted. Constructing a coherent political message for the global network audience is expensive as it will come under criticism that needs costly defence. But targeted incoherent political messaging is very cheap, especially when you can rely on the network to do part of the targeting for you if citizens engage in message sharing.
Goal 3 also becomes achievable when the most frequent connections accessed by citizens in making their minds up about issue B are the cheap virtual connections on social media. These connections are cheap to acquire but equally cheap to re-wire. The increased radicalisation of citizens that would emerge from such a scenario could very well lead to the disruption of real-world connections between them as well and hence go some way to achieving goal 1. The break-down of real-world connections is frequently reported in connection to Brexit!
The creators of the early platforms of social media shared with many proponents of direct E-democracy the belief that ‘cheap’ connections between citizens would allow for greater participation in democratic and collective decision-making processes and thus enhance the quality of democracy. From my basic and over-simplifying modelling however comes a serious warning: cheap and easily manipulable connections between citizens that might outweigh their real-world links in terms of quantity and leave the collective decision-making process wide open to outside disruptors.