This is an entry in our Keyword series, where we try to define the terms you’ll often hear when people talk about building a better Internet and put those keywords in their current context.
The term decentralization gets thrown around a lot today, often referring to a paradigm shift in Internet technologies that’s just around the corner. You may have encountered it in a conversation or documentation about something weedy and technical like a blockchain technology or a federated social network.
Compared to the other jargon that tends to flood those conversations, decentralization seems like a relatively intuitive concept. If something is decentralized, that means it’s not centralized. Ah, what a perfectly clear and specific definition.
Decentralization, much like the term “punk,” is one of those slippery terms whose meaning is almost entirely context-dependent. You’ll get a very different idea about what punk connotes from a New York Fashion Week runway show than you would from an anarchist-run Food Not Bombs drive.
The first notable use of the term decentralized as a description of communication networks came in a paper written by RAND Corporation researcher Paul Baran in 1962, called “On Distributed Communications.”1 Of course, there was no Internet at the time. Baran was hypothesizing methods for making military communications resilient should a nuclear attack destroy communication centers.
In the report, Baran provided a diagram that has really stuck in the imagination of people who theorize the structure of the Internet. It depicts three possible structures for communication networks: centralized, decentralized, and distributed.
In the centralized version, one node transmits information to many. In the decentralized example, a central node transmits to other nodes, which in turn become central nodes for transmitting information to other nodes, and so on. In the distributed version, there is no central node.
One reason that the meaning of “decentralization” is so messy is that a lot of times people are actually referring to what Baran identifies as a distributed system. It helps a lot to pick apart various technologies, and interrogate whether they are built on decentralized or distributed networks. So let’s look at a few technologies often deemed decentralized.
Peer to Peer or p2p technology like the BitTorrent or Secure Scuttlebutt protocols are often called decentralized because the most important data is not shared on a central server, but instead transmitted directly between users via some sort of client. The only data that is centralized is a record of such data existing.
So, for example, if you accessed the now-defunct Swedish BitTorrent site The Pirate Bay and wanted to download all of the seasons of The Sopranos, you wouldn’t be able to download the actual media from The Pirate Bay, just the BitTorrent file itself which directs your torrenting client to download the media from the peers connected to The Pirate Bay’s network.
The data was distributed, but the index of that data was centralized on The Pirate Bay’s servers. That’s why when Swedish authorities shut down The Pirate Bay in 2009, users could no longer access and download that distributed data.
Federated networks like Mastodon or other software using the Activity Pub protocol are great examples of decentralized networks, à la Baran: anyone can run their own instance of the social network on their own servers, acting as independent nodes. Mastodon looks a lot like Twitter, but unlike Twitter, Mastodon isn’t operated and hosted entirely by a central corporation. Instead, anyone who wants to spin up an instance of Mastodon on their own servers can. It means that you can never, for example, ban members of a violent white supremacist group from using Mastodon or Activity Pub in toto. You can only ban members from a specific Mastodon instance. That group can go form its own Mastodon instance on its own servers.
Blockchain technology forms an entirely distributed ledger of every transaction, which every participant on the network downloads and hosts. That ledger isn’t stored on a central server, but by all the participants who agree to host a copy of it. In some cases, like that of DogeCoin, the code mandates that every user store a complete copy of the ledger. In practice, the blockchain ecosystem often relies on a whole swath of centralized services, whether that’s the OpenSea NFT marketplace storing images associated with tokens or the cloud-based Metamask wallet that many use to store their coins.
In all of these cases, decentralization is effectively a metaphor, defined by a specific point of view of a given network. Of the three, federated networks are the only ones that meet Baran’s definition of decentralized by design—both P2P and blockchain match the distributed definition—but even federated networks only remain truly decentralized if that federation ecosystem is robust enough.
Let’s take the case of Mastodon to one logical extreme. If for some reason membership on a Mastodon instance ballooned to rival Twitter’s, and the majority of other instances withered from relative inactivity and ultimately disappeared, it would be pretty hard to claim that the Mastodon network is, in practice, decentralized any longer.
Researcher Francis Tseng does a fantastic job in his 2019 article “Decentralize What?” demonstrating how vulnerable the current web is given the fact that much of its hosting is centralized on Amazon’s AWS servers, deeming this “infrastructural centralization.” In that piece he also outlines his concept of “political centralization,” where state regimes have enough influence over domestic Internet networks that they can stifle dissent or enact service blackouts during periods of protest.
From a pure design standpoint, the more centralized a network is, the more vulnerable it is to censorship and disruption. Elon Musk has been running a case study in this principle since November of last year. He has shut down microservices that help people use two-factor authentication, temporarily banned journalists who are critical of him, and even attempted to block links to perceived competitors like Substack. Musk has demonstrated that Twitter is a vulnerable communication network because it’s so centralized that one powerful figure can shape it.
In theory, distributed networks are the most resistant to censorship and disruption, but they’re incredibly expensive and inefficient. For example, in 2020, crypto.com shared a infographic summarizing the inefficiency of blockchain pretty well. The financial transactions run by Visa could process 24,000 transactions a second, while the Bitcoin blockchain could only process seven transactions per second. One estimate claims that one Bitcoin transaction demands the same amount of energy as over 1,000,000 Visa transactions, as of this post’s publishing date.
For communication networks, decentralization must strike a balance. The Fediverse makes social networking relatively easy for people to access with minimal technical know-how, and also gives people with the skills and interest the opportunity to have a more hands-on relationship with how their instance is run. In both the best and worst case scenarios, that means that people have a more direct opportunity to shape their communication network to reflect their values.
Indeed, this is why we’re developing Smalltown. We hope that by spinning up discrete instances of our Mastodon fork, people have a dedicated space for local civic discussions that is values-driven, closely moderated, trusted & local. We think that the flat playing field of Nextdoor and the poorly governed hinterlands of Facebook Groups lead to context collapse, harassment, and the rapid spread of misinformation into discussions about local politics.
Similarly, it’s why Truth.Social and Gab also got their start as Mastodon forks: the people running those platforms believe the only way for them to have the types of discussions they would like to have is to build their own spaces. Or, in many cases, they were simply deplatformed and kicked off other networks, and these spaces welcomed them.
Both our project and these right wing projects view the centralization of social media as part of the problem. When the privilege to cultivate social spaces is entirely relegated to incredibly powerful private actors who are effectively unaccountable to the public or any kind of independent regulatory body, centralization looks pretty unappealing.
But beware, centralization isn’t defacto a bad thing, and often the loudest people online pushing for decentralization have very explicit libertarian motives. Its biggest proponents in Silicon Valley–such as the venture capitalist Balaji Srinivasan–are often ideologues agitating for a digital free market, insulated from regulator action.
For those who may be curious about what centralization can offer us, I’d direct you to read a fantastic little book called The People’s Republic of Walmart. Its authors Leigh Phillips and Michal Rozworski make the case that Walmart’s global, cybernetic logistics operation, and ability to manipulate and temper entire supply chains, is a true technological marvel of the 21st century that need not be used simply to make the Waltons one of the world’s richest families. Instead, they argue, were such a huge, centralized coordination system to be taken over by a socialist state, it could be used to distribute resources and eliminate hunger in an efficient, ambitious manner not yet seen by humankind. Already, Walmart uses its bargaining power and distribution structure to provide insulin at some of the lowest prices in America.
I end on this note to say that centralization and decentralization in and of themselves are neither good nor bad, virtuous nor evil, advantageous nor wasteful. Since the definition of these things is so slippery as is, I invite you to meet their usage with thoughtfulness and curiosity.
There are some questions you can always ask when people invoke the term. What is being decentralized, and from what center? What do the people doing that decentralizing want? What is the problem those people are identifying, and would whatever they’re calling decentralization solve that problem adequately without creating worse complications?
1 In his 1964 follow-up paper “On Distributed Communications II”, Baran would propose his most consequential contribution to the Internet’s architecture, something he called “hot potato switching.” This came to be known as packet switching and is an essential component of the HTTP protocol.
2 iDPI’s Ethan Zuckerman and Chand Rajendra-Nicolucci wrote a piece for the Social Science Research Institute outlining the so-called “local paradox,” that simply making social networks smaller and more geographically local for users does not necessarily reduce the amount of misinformation on social media, and in some cases, can make it move faster and become much more difficult t. o mitigate. Quality moderation and solid community norms are crucial for the quality of information on these platforms.