This is an extremely fascinating article reporting on a couple of research studies by the author (The Wisdom of Partisan Crowds and Networked collective intelligence improves dissemination of scientific information regarding smoking risks) that  - contrary to what you might expect if you follow Eli Pariser's line of reasoning on filter bubbles - show partisan crowds can in fact be pretty wise, converging on more nuanced, more tolerant, less biased views when left to their own devices to discuss the issues about which they are partisan. Rather than amplifying their biases, they actually become less partisan. This happens (apparently reliably and predictably) when - and only when - networks are egalitarian: when there are no clear leaders or privileged voices. When they become more centralized, i.e. when prominent influencers connect to many others, they turn into echo chambers that amplify the influencers' biases and intolerant views. The fairly startling, and heartwarming takeaway is that greater equity leads to greater tolerance and wisdom, even when the groups themselves started out with highly partisan views.

Centola's discoveries help to explain some of the big issues we see in large-scale social networks, with a relatively small number of hubs linking a much larger number of people together and thus amplifying the biases in the ways Centola describes. To pick a fine hair, though technically accurate, I'm not sure about the wisdom of using the term 'centralization' to describe this: it is totally about network centrality in the hubs, but 'centralization' implies a deliberate hierarchy to me (to centralize implies someone doing the centralization), which is not how it works. It is still a distributed network, after all, just one that (on average) follows a power law distribution. However, as Centola tentatively suggests, knowing this provides us with a potential lever to disrupt the harmful effects of echo chambers. The trick, he claims, is not to eliminate the echo chambers, but to do what we can to increase the equity within them. This, as it happens, aligns fairly well with Pariser's recent rather fuzzily formulated and weakly justified call for 'online parks' . I look forward to reading Centola's new book on the subject, due out in January.

How might we use this knowledge?

I think there may be great potential for social media designers to use this knowledge to take the big influencers down a few notches. Indeed, using a very different theoretical basis, I did something rather similar myself when I developed my old CoFIND system (a social bookmarking system using the dynamics of evolutionary and stigmergic systems to evolve structure) in the late 90s and early 2000s. Like others working in the field, I had noticed that a really big problem with my evolving system was that popular resources and fuzzy tags (that I called 'qualities' - they were scalar rather than binary categories) tended to stay that way: it was a scale-free network with a long, long tail. My solution was to give a novelty weighting that brought novel tags and resources up to equal prominence with the most viewed/ranked, and that could be topped up by being used/ranked themselves, but to decrement the value if they were not used. Initially I made the decay rate constant, which was stupid: if the system was not used for a week or two, there would literally be nothing left to see, and it was really hard to tune it right so that new things didn't stick around too long if they were not popular. Later, I made the decay proportional to the overall rate of use of the system or niche within it, so it tuned itself: when the system was used a lot, new resources and fuzzy tags didn't stick around for long but, in less popular systems, they would fall more slowly. The idea behind it was to provide a means for things to 'die' in the system for lack of feeding, and for things that were really no use to make them starve pretty quickly. New resources would have a chance to compete but, if they were not used and rated, they would decay quite rapidly - relative to system use - and drop down into the backwaters of the system where few would ever visit. Later (or maybe it was earlier - my memory is vague) I slightly randomized the initial weighting so introduce a bit of serendipity and to reduce the rewards of gaming it.

In fairness, my mechanism was a bit of a sky-hook of the sort the intelligent design nincompoops invoke when trying to find a role for supernatural beings in evolutionary systems. In natural ecosystems, though novelty can sometimes be beneficial when it allows an organism to occupy an unclaimed niche or to out-compete an incumbent, novelty has no innate value of its own. If it did, it would have evolved from the bottom up, certainly not from the top down. However, I reasoned that I was defining the physics of the system so as to influence its behaviour in the direction I wanted to go (to help people to help one another to learn) and thus could legitimately make novelty a positive selection factor without departing from my general principle of letting evolution and stigmergy do all the work. I was also very aware that the system had to be at least minimally useful and, if I had allowed evolution to do all the work (which I did try, once), given the widespread availability of other well-designed social bookmarking systems, no one would ever use it in the first place: the whole system would have been an evolutionary dead-end. 

I think the principles I followed could be used for pretty much any social network. If we think of the algorithms that choose what, how, where, and in what order things are displayed, as the physics of the social system, then it is quite legitimate to tune the physics to make the network more equitable and egalitarian, while still retaining the filter bubbles that draw people to them. The big question that remains to me, though, is whether anyone would want to use it. I suspect that this kind of flattened social network may thrive in some niches. It would probably be really useful in academia, for instance, research communities, and other vertical markets where the set social form is equal to or more dominant than the network social form but might not be a great competitor to Facebook, LinkedIn, Twitter, and other commercial social networks precisely because of the awful role they play in forming and sustaining identities, and cultivating an exaggerated sense of belonging. Social networks naturally gravitate towards a long-tail distribution so, if we suppress that, they might not form particularly well, if at all. It would be really interesting to try, though.