Shortcuts in the social graph

I gave a talk on Monday in Salzburg, Austria that expanded slightly on my TED Global talk about imaginary cosmopolitanism. The audience in Salzburg included American, African and European media development experts, and I made the case that groups concerned with media in a digital age needed to look not just at producing media from different corners of the world, but the appetite for this media. We can’t just look at production – we need to look at consumption as well. And we need to be careful not to look at tools that have a global userbase and assume that they’re naturally leading to international interaction – just because there are 500 million Facebook users, it’s not helpful to think of them as a new “nation”, as the Economist and many others suggest.

My friend Onnik Krikorian watched the video of my talk and tweeted a response: “Just want to say that even small connections r more than we otherwise thought. Still, food 4 thought”

Onnik, you may remember, is Global Voices’s regional editor for the Caucuses. A Brit of Armenian descent, he’s had the challenging task of building connections with Azeri bloggers who were initially reluctant to trust him, given the long tensions between the two nations. Onnik is a passionate advocate for connection via Facebook, which has allowed him to build ties with Azeri friends in a way that would probably be impossible offline. While I’ve been critical of the ways in which Facebook is branding itself as a tool for international understanding without much data to support that contention, Onnik’s been reminding me that the connections that get made on networks like Facebook are critically important.

Turns out that Onnik’s not just right in practice – he’s right in theory.

I’ve been reading some of Duncan Watts’s research to get a better understanding of the “small world” hypothesis. You’ve probably encountered this idea as “six degrees of separation”, the idea – promoted by social psychologist Stanley Milgram that social networks in the US connection people through chains of six people. (Milgram’s actual experiment doesn’t provide a lot of evidence that this is actually true – in the most widely discussed experiment, where letters sent from random people in Nebraska reached a specific stockbroker in Massachusetts, only 18 of 96 letters reached their targets. The “six degrees” takeaway was an average of studies including ones where the letter to Boston began in Boston, and an average that doesn’t consider the failed chains.) Watts became fascinated with the problem during his doctoral research and began developing increasingly complex mathematical models to test the properties of “small worlds networks”.

Basically, the small worlds phenomenon is pretty easy to explain if we assume that everyone in the world has a fairly large number of friends who are distributed randomly. If I know 100 people, and they each know 100 people, within two degrees of me, I know 10,000 people. We’re up to a million in three degrees, a hundred million at four and 10 billion at five – and bingo, we’ve spanned the globe.

Of course, that’s not how friendships actually work. I know Onnik, and we both know lots of people the other doesn’t know… but we’ve also got lots of friends in common, via Global Voices. So if Onnik is one of my hundred friends, and twenty of his friends are already my friends, I’m reaching a much smaller set of people through him than I would through a friend who had no overlap with my other friends. Or, as Watts puts it in his book Six Degrees: The Science of a Connected Age, “the more your friends know each other, the less use they are to you in getting a message to someone you don’t know.”

This is interesting to me because I’m intrigued – and worried – by information flows through social networks. If we’re getting more (not lots yet, but more) information through social networks and less through curated media like newspapers, do we run the risk of encountering only information that our friends have access to? Are we likely to be overinformed about some conversations and underinformed about others? And could this isolation lead to ideological polarization, as Cass Sunstein and others suggest? And if those fears are true, is there anything we can do to rewire social networks so that we’re getting richer, more diverse information?

Watts has considered these questions from a point of view rooted in graph theory. He explains that networks where friends are highly likely to know each other (i.e., if person A knows person B and person C, B and C are likely to know each other) have a lot of clustering, and few short paths – it may be difficult, or impossible, to find a path that takes you from Nebraska to Massachusetts. In networks where friendships are random and uncorrelated (i.e, A knows B and C, but B and C are no more likely to know each other than any other nodes in the network), paths are very short – remember, with 100 friends distributed randomly, five hops reaches 10 billion people. But these networks appear to be very different from what we experience in reality.

Small world networks are both highly clustered and show short paths, the sorts of paths Milgram’s experiment suggests exist. These networks emerge when participants in networks are highly connected to their near neighbors (i.e., are clustered) and have a few random connections to distant parts of the graph. Watts discovered that, “the first five random rewirings reduce the average path length of the network by one-half, regardless of the size of the network.”

Those “random rewirings” aren’t so random in real life. We know people in different clusters – I know one set of people from Berkman, another from my college days, another from Global Voices, etc. But that fact that individuals are members of multiple clusters is what allows for “shortcuts” in the graph. People who bridge between different, separated communities – as Onnik does with Armenians and Azeris – shrink the world.

It turns out that shrinking the world in this way may be highly profitable, in terms of creativity, compensation and status. Ronald Burt, at the University of Chicago Business School, studied how ideas spread at Raytheon, a large defense contractor. People in the firm who were most widely recognized as having good ideas were those who bridged different social and professional groups within the company. Burt refers to the divisions within the firm as “structural holes” and introduces his paper with the hypothesis that “people who stand near the holes in social structure are at the highest risk of having good ideas.” To the extent that there’s little communication between Armenians and Azeri for reasons of history, language and culture, there’s a structural hole that prevents ideas from spreading. Onnik, and other people who bridge divides, are at risk of being exposed to new ideas and ways of thinking.

I’m trying to understand what this might mean in terms of the spread of news and other information from around the world. Most news, I suspect, doesn’t spread primarily through small-world networks. We learn about the Haitian earthquake not because we know a friend who knows a friend who knows a Haitian person who called them – we hear about it from news outlets who are reporting on the story, and if we’ve somehow managed to miss hearing the news, friends send us links to journalistic reports on the situation.

I suspect that small world networks are important, though, in helping us decide what news is important to us. If you have a personal connection to Haiti – a Haitian friend, someone who’s traveled to the country – the story may be one you followed more closely that stories about other natural disasters. If you chose to get active in providing support to people affected by the Haitian quake, your involvement may have inspired friends to pay closer attention to the situation and, perhaps, to get involved themselves. And your attention to a story sends a signal to news outlets (professional and amateur) that this is a topic of interest that they should cover.

I’m interested in modeling how international news spreads in a digital age. I suspect a model needs to do much more than look at who’s reporting what stories – like the models I built years ago for the Global Attention Profiles project – they need to consider user interest in those stories, expressed by which headlines they click online and which URLs they amplify on Twitter, Facebook and via email. Understanding those dynamics, in turn, may rely on understanding how individuals communicate to each other the issues they personally care about. My guess is that people who connect the social graph aren’t just likely to have good ideas – they’re likely they key people who help us pay attention to distant parts of the world.