Bluesky’s Great Social Splintering
When Elon turned Twitter into X and shifted the vibes a couple of octaves to the right, an exodus of alienated users scoured the internet for a new home. Eventually, they landed on Bluesky. Built on the decentralised social network protocol, AT Protocol (or “ATProto”), Bluesky allows users to carry with them their social graph and content across different platforms. This ability to pull up the tent and move elsewhere is supposed to shelter the platform users from the erratic whims of mercurial billionaires, who will be beholden to keep the users happy, lest they wander off.
Free choice, that fabled cure-all, may seem like a good solution to the vexing and often paradoxical challenge of governance in social networks. Yet, all the devils dance in the design. If adopted at scale, ATProto’s frictionless, cross-platform portability contains within it the threat of a negative competition environment, a classic Moloch, that will further splinter our social sphere.
Mundus sine Caesaribus
Jack Dorsey, still during his tenure as CEO, instigated in 2019 Bluesky as a project inside Twitter to build a decentralised social networking standard. In 2022, it severed its ties with Twitter to establish itself as an independent social network built on the ATProto standard. Today, there is momentum on Bluesky. It recently hit 35m users—still a couple of OOMs less than its competitors, but it’s not nothing. And the builder energy is palpable: projects are popping up left and right, TikTok and Instagram clones, services for analytics or feed customisation, Mark Cuban throws seed capital at some of them. Vibes are fun, it’s the good internet, good in the friendly-neighbour-kind-of-way (with a sprinkle of furries). When CEO Jay Graber went on stage at SXSW wearing a t-shirt with “Mundus sine Caesaribus”, the t-shirt sold out in 30 minutes.
At its core, Bluesky is built up around a federated content system, which is a way to remove single control and distribute the network infrastructure across many nodes and operators. Anybody can set up a server to host content and users, while the content can flow between them [ref]. Crucially, users can move between different serves or services, as they see fit, taking their content, connections, and followers with them. This account portability creates new incentive dynamics.
“The fact that users have that choice means we’re incentivized to keep serving [them] […] If a billionaire came in and bought Bluesky and took it over, or I decided tomorrow to change things in a way that people didn’t really like, then they could fork off and go on to other applications.” Jay Graber in The Observer
In short, the argument is for good governance from the wise crowd. The crux then is this: what is the nature of those choices, and how will platform fungibility impact the communication ecosystem as a whole? In the next section, I will describe how the structure of the open internet pushed people into attention sinkholes, where antisocial content curation creates negative information spirals.
Islands of Ideology
In the first wave of user-generated content, the Web 1.0, the hyperlink reigned supreme as a directed attention edge in the network. HTTP and HTML protocols made it relatively easy to create a publishing platform. From one-to-many communications, we moved into more-to-many. Naturally, most of these websites and forums had trivial reach, if any at all, however, in aggregate, they began to matter. In the 00s, analysis of hyperlink graphs was all the rage. Often, we would see something like this:
On one side were a cluster of blue nodes, densely connected, on the other a cluster of red nodes. Between the two clusters was a gossamer net of bridges. It was a graph displaying the hyperlinks between websites of either conservative or liberal leanings in the US.
Structurally, the tea leaves were there to be read. It might have been possible to extrapolate from these network structures the splintered information landscape of today. The graph was the message. The incentives in blue and the red cluster of the American political blogosphere look very different. It’s possible to assume certain evident truths on one ideological island that would be deeply controversial on the other. I don’t think it’s hyperbole to state that this kind of information flow topography accelerated and exaggerated what was already an already existing trend in a splintering society.
With Facebook, YouTube, and Twitter, it became super easy to publish content, which then could be distributed to other users on the same platform. Media moved from more-to-many communication to many-to-many information flows in Web 2.0. These social networks became attention sinkholes, where increasingly more time was spent online. Curation and moderation of information became key to a successful platform experience, and the way to attract and retain more people. The role of the information gatekeeper and curator moved from a professional class of scholars, journalists, and politicians, to a class of technologists, who implemented their (very human) priorities through algorithms that could operate at scale on those astronomical volumes of user-generated content. Those technology companies emerged and sustained themselves through ad-based revenue models that, generally speaking, relied on keeping eyeballs on the site for as long as possible. The best way of doing that is to meet the user’s tastes—even if those tastes favour short-term hedonist satisfaction over long-term health.
One doesn’t have to go full Sapolsky and reject free will to claim that humans tend to be generally geared towards hedonistic short-termism. There is less obesity in modern, capitalist Japan than in modern, capitalist USA, because the Japanese state took measures to implement nutritional policies to induce a healthy populace. If we move from nutrition to information, as Yuval Noah Hariri points out in Nexus, a similar dynamic appears: quality information is expensive to produce, slob is cheap. Cheap content will outnumber and outpace thoughtful and carefully crafted content by several OOMs. Not only is it cheaper to produce, if no longer tethered to truth, it’s much easier to make glitzy, and our brains will be more attracted to the shiny stuff. To the fat and sugars. This is the race to the bottom of the brain stem. It’s not a question of left versus right. It’s about fast versus deep. About something that approaches truth against something that pursues clicks.
Social networks are winner-take-all. The marginal cost of digital distribution is almost zero, and it is non-competing good (i.e. one person consuming digital content does not prevent another from doing the same, as opposed to, say, a banana). In this game, there are few moats, arguably, the only moat worth caring about is the network itself. This is why Reid Hoffman advocates blitzscaling. Every marginal user provides marginal utility to the entire network, and when it reaches a tipping point, the utility of the entire network (all users, the connections between them, and the content they’ve generated), becomes the moat. It makes sense for a new user to join the biggest networks. Unsurprisingly, we observe a social media landscape dominated by a few gargantuan, continent-sized networks, who function a bit like sinkholes, sucking in users and attention and capital.
If the other users are a given platform’s main benefit, then switching comes at a high cost for any individual user. All the network value is lost, you have to start from scratch on a new network. For a platform operator, it makes commercial sense to keep switching costs high, well, at least to not invest in lowering them. Removing offboarding friction leads to user leakage and will undermine the most critical value that the network has: you.
Web1.0 introduced a new era of publishing. While it was in principle open to all, an open-ended, scale-free network, power accumulated in reservoirs. During the Web 2.0 era, much of the internet traffic flowed towards social media platforms, on which commercial pressures pushed the content into more and more extreme, while also pushing the operators to erect walls around their gardens to avoid user leakage. Now, Bluesky’s ATProto proposes to solve the former with a fix for the latter.
The Futility of Defederation
In federated social media ecosystems, such as the ones based on ATProto, cross-platform switching costs are severely reduced for platform users, and we must ask what new types of incentive incentive structures will emerge from this design. I will attempt a thought experiment to explore. First, let us spin up three federated servers: A, B, C.
A and B are happy places: the content is generally moderate and benign. The vibes are good. C, on the other hand, is laissez-faire in their moderation; a couple of conspiracy theorists start spewing the nonsense du jour. Users on A and B are upset—ever diligent and willing to listen to their communities, the platform managers defederate A and B from C.
Users on AB stay happily ignorant of all the mess on C. However, C is a commercial venture that competes with attention sinkholes like TikTok, X, and Instagram. They make money from showing as many ads as possible to their users, and they know that in order to reach a defensible market position they must blitzscale. So they tolerate or even (subtly) encourage incendiary content (space-laser lizards cabals, soft porn a la MTV, etc), and use every single dark design pattern they can without falling afoul of the law. Their island grow. Each marginal user brings marginal utility to the network as a whole, they scale. They raise more venture capital, and take a page out of the TikTok playbook with an aggressive marketing campaign on Facebook and X.
Now comes network D, a new AI social media startup. Starting a new venture is all about peeling away the onion layers of risk. Location risk is certainly a key concern for a scruffy startup taking on the Goliath. One of the first layers of risk, after some positive indication of a product-market fit, is simply to get significant traction. It is a rational business decision to be where there is enough user flow potential. Put the store on the high street, not in the back alley. They like ATProto, as this is an easy way to usurp users from other networks. D decides to federate with A, B, and C, but in order to compete with the biggest network, C, they implement even worse incentive structures.
Their network is basically a Fentanyl casino. At first, C finds mutual benefit in federation, when the competitive pressure is perceived to be too intense, they defederate. Well, the C users can move their social graph to the new network, so they do. C relaxes their moderation policies even further and stays connected. Together, CD now race at breakneck speed towards the bottom of the brain stem. AB stays happy and ignorant. Tristan Harris makes another documentary. Sociologists, media theorists look on in horror.
Then another social media platform, E, arrives… and the cycle repeats. Ad nauseam. The likelihood that a new network or user would join CD over AB is significantly higher, unless they have some kind of political bent that justifies going for smaller networks (or perhaps, more furry-friendly). In the end, the informational and ideological world on AB and the world on CD looks entirely different. CD is slob, cheap conspiracies, bad logic, in short, mortal entertainment. AB might be healthy, but politically irrelevant.
Decentralised Glitz is Still Glitz
Bluesky claims that account portability will let the collective behaviour of users keep the platform operators in check. By removing the barriers to free choice of social network, users can vote with their feet/clicks. Yet, unmitigated this platform fungibility comes also with a threat of worsening an already bad informational incentive landscape.
Clearly, we don’t want to go back to the days when a single, centralised narrative. When trust in journalism was high, it was only because those who bought ink by the barrel became informational suzerains, the gatekeepers of a fabricated truth. Today, the propaganda is executed via algorithms and platform design under the auspices of centralised networks, and it seems reasonable to assume that smaller, more user-controlled networks might mitigate some of the centralisation risks. In that sense, I agree with Renee diResta:
The idea of one massive “public square” should be squarely over & dead by now -didn’t work, won’t work, can’t work. The whims of a billionaire shouldn’t be the rules. But on a centralized platform, they are. So here we are.
Yet, it should similarly be clear that a decentralised system that facilitates the frictionless and fast movement of content and users between platforms, will increase platform competition, and the tendency is not for truth to emerge, but rather for glitz to explode like confetti. And this will further accelerate as AI joins as content producer, curator, moderator, perhaps even as consumer, and everything in between. All it takes for this hypothesis to hold is, I believe, the claim that the human psyche tends towards the simple, fast and glitzy, and not the slow, healthy, complex.
For now the jubilant Bluesky community delight in the construction of a viable alternative to X, but in the distance looms an unresolved question — and, so far, a proper analysis of the ATProto design’s incentive structures is relegated to the sidelines of the conversation. I’m on Bluesky, I want Bluesky to succeed. Online social platforms, correctly constructed, are great information networks that can produce and distribute knowledge for a better world, a world of more understanding and collaboration, and may even starve off the bellicose mares that again seem about to spring from their gates. However, I believe serious thought needs to go into the incentive schemes that will emerge from ATProto’s platform fungibility. Right now, I don’t have any suggestions for how to mitigate some of the risks herein highlighted. All I hope is that as the development and adoption of ATProto continues, we can have a clear-eyed conversation about the path ahead.