Fake news is in the news, part of the nauseating task of dressing the ugly wound of the presidential election. The OED has named “post-truth” its word of the year, and the Republican candidate for president aligned himself with the kind of wilful confusion that is known to be a deliberate tool of autocratic regimes.

Some of my published research is about seeding informative or persuasive messages in social networks, and I have a recent working paper on the vulnerability of network structures to exploitation by outsiders. What have I learned from studying these phenomena that might help us to understand a bit more about fake news?

In that working paper, I ask how the structure of a network might make it more or less vulnerable to exploitation. The original version of that paper is many years old, originally about what places in a network a snake oil salesman would hit first to get in, make a quick buck, and get out. So there are some painful parallels.

The idea of the paper in its current iteration is a network of people who interact with an outsider. Every day one person is selected to play a trust game with the outsider. Good long-term relationships are mutually beneficial, but the outsider faces a short-term incentive to rip off a consumer and run. Think something like a mechanic: your car breaks down and they want you as a customer in the future, but they might be tempted to overcharge you.

The idea is that your neighbors in the network can help to protect you: if you rip me off, I’ll shun you, but I’ll also tell my friends to shun you. My research question is how the game changes when the outsider knows who lives where in the network—how many friends I have and who they are—versus when they do not.

The information is a mixed blessing. It can expose people who have poor social protection and so erode trust. If on average we all have lots of friends, we might all be safe, but if you are “lonelier” than average in the network, once your identity is revealed you become vulnerable. Your small number of connections may not be enough to discipline the outsider. It might then sometimes pay for the outsider to stay “strategically ignorant” about the network structure, to avoid the temptation to cheat people.

What structures are vulnerable? Fragmentation and isolation in the network create vulnerability. The problem is that it’s recursive: vulnerability is contagious. Your vulnerability takes you out of the equation, and so your neighbors become less protected than they may have thought. In the hub-and-spoke here, if the peripheral people are vulnerable to being ripped off, the central person has no protection at all.

So where does this leave us with the fake news problem? The model in my paper is not a perfect match to the fake news issue, but there are some possible linkages.

1. The fragmentation of media means that there is no “firewall” of truth. Niche websites or Facebook groups can begin to erode trust around the margins, but then leave the center exposed, feeling isolated and destabilized in its understanding of the world.
2. Knowledge about the structure of social networks is entirely available, and individuals and groups can be surgically targeted for their vulnerability to fake news. The analogy is that in the paper connections help you to socially punish an outsider that rips you off; in the fake news application, connections help to keep you in touch with mainstream news. It may not pay to shoot in the dark, hoping to hit a few vulnerable folks, but when you can surgically target them, fake news becomes a viable, profitable enterprise.
3. The destabilizing effect of post-truth confusion politics may leave people feeling like they live in a Twilight Zone where those around them are suckered by news that seems obviously ridiculous. This is like the recursive vulnerability in the paper: you thought you were safely insulated in a reality-based community, but once your vulnerable neighbors are picked off, you are left on an island.
4. Networks with redundancy and cliques are the least vulnerable to exploitation in my model. In the fake news case, the implication would be that isolated people and communities would be more vulnerable to being suckered by misinformation, while those with many mutually reinforcing safety valves, in densely packed areas of the network and with rich in-groups, are the least vulnerable. The fringes truly are the fringes of the network.

Another strand of my research is about targeting. I wrote about that here before so I won’t repeat myself too much.  A key idea from that research is that if you are trying to spread a message or persuade a group of people, it pays to use dispersed, irredundant seeding rather than concentrating on one neighborhood or group. In the network to the left, the cheapest way to reach everyone or their neighbor is to send three messages to the shaded consumers. Notice the dispersion among the different clusters in the network.

This is a little complementary to the previous model. A minute ago I said that there is safety in numbers: being part of mutually reinforcing cliques helps. For a messenger, cliques are also helpful since they cut down on the number of people you have to reach to get the word out.

But the whole world is not one big clique. When faced with a big population that clusters around different media, ideas, or neighborhoods, the right way to do cost-effective messaging is to reach a small number of people in each neighborhood. Social networks like Facebook and Twitter make this strategy quite simple and viable relative to legacy media, even other internet technologies. Cost-effective blanket seeding is relatively easy in a world in which social media platforms are only too happy to sell you tailor-made baskets of targets in the network.

So what lessons can we draw for the fake news case?

1. The fact that “mainstream” media consumers are clustered together gives a catch-22. That internally well-connected cluster is over-targeted by the true news and over-sampled when we try to figure out how many people are influenced by that true news. It is both a hard cluster to penetrate with fake news, but a cluster that will naturally be complacent about fake news precisely because they are optimally infrequently targeted by a smart algorithm. Note this is not because they’re somehow too smart to be duped, but because that fake news algorithm will try to avoid over-targeting any one cluster.
2. Emissaries to the fringe clusters are probably needed if any headway is to be made against fake news. This is where the self-segregation possibilities afforded by social media really create a tough feedback loop. The “echo chamber” is self-constructed but it also facilitates easy targeting strategies on the part of the cynical fake news machine. We have to put ourselves in the way.
3. I show also in this research that competitive targeting is complicated, but in a way that might help us strategically undermine fake news. Exploiting word-of-mouth gets hard when you are competing with another organization to seed the network: your opponent can “fill in the gaps” in your seeding strategy, creating a patchwork, dispersed firewall. Maybe I’m crazy to advocate an escalation of the confusion in the face of confusion, but putting roadblocks to the lie getting around the world before the truth gets its boots on is, I think, not a lost cause. Media messaging budgets could usefully be spent replicating the fake news strategy with the truth. When conspiracies become truth, emulate a conspiracy.

The telephone game teaches us well that everything gets garbled in the retelling—truth and lies alike. Centralized media no longer has the reach to preclude the telephone game. The fact that word-of-mouth is the most trusted of all forms of messaging combines to devastating effect with this decentralization and with the rise of targeting in social media. We must either rebuild “grand cliques”—a fool’s errand, perhaps, to turn back the clock—or build new firewalls. This calls for a fundamental rethink of how mainstream news disseminates its messages, and perhaps calls for more citizen activism to consciously emulate the same tools that have enabled fake news to run amok. Localness and smallness and dispersion are the kindling for fake news, and we should fight fire with fire.