Web3 advocates promise decentralization on an unprecedented scale. Excessive centralization can stymie coordination and erode freedom, democracy, and economic dynamism — decentralization is supposed to be the remedy. But the term on its own is too vague to be a coherent end goal. Getting the job done takes the right kind of decentralization, and we worry that Web3 is thus far heading down the wrong track.
In particular, we worry about the focus on degree, rather than type, of decentralization. Focusing on degree — whether we want more or less decentralization — can lead Web3 advocates to mischaracterize both the reality of existing centralization, as well as the possibility of pure decentralization. On the one hand, existing “centralized” systems are not nearly as centralized as Web3 advocates commonly describe. “Legacy” banks delegate many activities to local branches, and even central banks are often consortia. Architecturally, “centralized” clouds are rarely so centralized in practice; they are usually scattered around a range of geographies and train large machine-learning models in a distributed fashion.
On the other hand, many Web3 critics have pointed out the extreme inefficiencies that accompany proposed proposed decentralized architectures, as well as the inevitable re-emergence of “centers” in Web3 (NFT platforms, currency exchanges, wallet providers). In addition, there are important limits and trade-offs involved in broadly aiming at greater decentralization. Narrow technical decentralization, for instance, faces contradictions between resisting censorship and embedding values that often results in either worse functionality or some centralized decisionmaking in the end, as shown by content moderation on decentralized social networks.
Thus, there are (soft) limits to the degree of centralization and to the decentralization feasible in a functional system. Rather than pursuing a false debate over whether next-generation technology should be centralized or decentralized, we should ask how to best arrange the pattern of desirable decentralization. Such a debate requires precisely articulating what we want from decentralization.
We believe decentralization’s value is in genuinely empowering people to act decisively within their social contexts, while providing mechanisms of necessary coordination across contexts. This is in contrast to the current technical landscape, where decisionmaking agency over information, computation, moderation, and so on is increasingly in the hands of authorities “distant” from the relevant groups — for example, platform content moderation processes try to be cross- community and cross-cultural, and largely fail at both. In this situation, decisions are removed from the context of application and made by people with little direct interest in the matters, who are then unable to take advantage of rich distributed information.
Our view of decentralization is about coordination. It emphasizes solving problems through the federation of “local” units, clustered around the social contexts most relevant to the decision at hand. This is not a new idea: US federalism, with local, state, and national governments, essentially pulls from this principle of subsidiarity, as does the setup of open source code repositories and wiki-like structures for information aggregation. The key is that these local units are composable—Modular and interoperable with each other, essentially “stackable” to a more global scale — to enable decentralized systems to efficiently solve problems that may at first blush seem to require centralization for coordination. We call this model composable local control.
Composable local control would distribute decisionmaking, leveraging a core principle of both markets and democracy: Those closest to a problem usually have the most knowledge and the greatest stake in its resolution, and it is by aggregating, federating, and filtering this knowledge that the best collective decisions are made.
Subsidiarity is the architecture and type of decentralization that makes composable local control possible. But the dominant trajectory of Web3 is unlikely to deliver, and may even run contrary to, subsidiarity. Permissionless blockchains are built as a distributed redundant ledger, where storage and authority are allocated by anonymous economic mechanisms and accessed via fungible, transactable resources such as computation and tokens. This architecture is optimized for a highly narrow set of problems, and thus by its very nature is unable to interface with the rich economic and social networks in which problem-solving coordination is actually needed. Such purely financial systems have a well-documented history of concentrating wealth, information, and power, properties that the current Web3 ecosystem is already taking to extremes. Thus, redundant distributed ledgers are in tension with subsidiary networks and the benefits of the form of decentralization we advocate.