The Web3 Decentralization Debate Is Focused on the Wrong Question

Fixating on the degree—rather than the type—of decentralization is leading us astray.
Collage of images of blockchain digital coin and hands raised
Photo-Illustration: Sam Whitney; Getty Images

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 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.

We remain optimistic about the potential for Web3 and adjacent spaces to redirect toward subsidiarity. However, for this to happen, we must take steps to make Web3 into a network of networks, not a ledger.

The “decentralization” currently achieved by most canonical crypto projects, such as Bitcoin, is what we call “distributed redundancy”: global, open, consensus-based storage in many locations of a common, homogeneous data set. Distributed redundancy depends on three factors:

  1. Maximally removing data from social context. (All interactions are boiled down to transactions as recorded in the ledger, with outside context unable to be represented in the technical architecture.)
  2. Aiming for universalized solutions. (A focus on “global” applicability requires all solutions to be applicable in all contexts.)
  3. Relying on global consensus and redundant verification accessed using fungible resources. (Decisionmaking mechanisms are limited by tokens or computational puzzles; people with more financial resources have more of these.)

Why might so many be so enthusiastically pursuing redundancy and universality? Theoretically, redundancy aims for security against attack. Yet, as we have seen play out in both recent supply chain challenges and the concentration of most Bitcoin mining into a small number of mining pools, market efficiency tends to concentrate activity in hyperscale centers, often highly brittle to shocks and disruptions (e.g., local Covid lockdown policy) or located in jurisdictions (e.g., China and Russia) that may be vulnerable to geopolitical risks. Effective and secure redundancy requires deliberately compensating for this tendency, choosing diverse “hedges” against risk rather than simply the lowest-cost providers. But achieving such hedging requires tracking locality and network relationships that these purely financial systems ignore.

In sharp contrast to these principles, the type of decentralization that we believe is desirable, subsidiarity, focuses on:

  1. Keeping data as close as possible to the social context of creation.
  2. A plurality of solutions linked and integrated through coordinated mechanisms of federation and interoperability.
  3. Leveraging and extending relationships of online and offline trust and institutions.

Perhaps the most famous system engineered to be subsidiary from the start is the original “network of networks,” the TCP/IP-based internet, which was architected in this way precisely for security and efficiency reasons, and has arguably had unprecedented success. More recent and more nascent examples include:

  1. The ActivityPub standard for social networks and associated applications such as Mastodon.
  2. A range of research projects from Ink and Switch, including an architecture for “local-first” computing and designs for interoperability across diverse programs.
  3. A range of socially local identity systems, such as Spritely, BrightID, and BackChannel.
  4. Federated learning and broader privacy-preserving machine learning.
  5. Mesh networks.
  6. Data collaboratives, cooperatives, and trusts.
  7. Wikipedia and wiki-based content structure more generally.
  8. Community content moderation systems, such as Reddit.
  9. Community-first cloud computing, file storage, and time-sharing.

Unlike distributed redundancy, subsidiarity often increases efficiency by leveraging trust, rather than reducing efficiency to eliminate the need for trust. Take something like community mesh networks, through which communities bootstrap decentralized wireless networks through shared nodes and antennae, installed locally. Creative economic incentive design is crucial for the sustainability of such networks, but these incentives are embedded in social relationships, rather than acting as a replacement for them. Similar principles underlie recent, blockchain-based alternatives, which we welcome.

The following flesh out the distinction between subsidiarity and redundancy, building on some of the areas of potential impact for Web3.

Identity and Reputation

Promise: Web3 promises to liberate identity and reputation from the control of a few large tech companies, allowing for “self-sovereign” identities for communication, transaction, and governance.

Redundancy: Pseudonymous ledgers, the basic data architecture of Web3, are poorly suited as identity or reputation primitives. Given the ease of setting up multiple accounts on a pseudonymous ledger, sybil attacks (or multiple-identity attacks), where people seek undue influence of a platform through control of multiple identities within that platform, are common. Solutions that take redundancy as a prior focus on universal, decontextualized unique cryptographic identifiers. Removing context leads to reliance on “universally secure” identifiers based on clean/universal features like biometrics, which often raise at least as many concerns as the centralized protocols they replace.

Subsidiarity: Trust is the basic building block for identity, and most relevant interactions that turn to identity mechanisms for proof or verification are far more about relationships (status as employee, citizen, student, platform contributor) than they are about universal identification. Since the early days of the internet, network-based approaches to identity (often called “Web of Trust” or “Trust over IP”) have imagined verification based on strong but often informal trust relationships; examples of recent protocols that build on this framework include Spritely, BackChannel, KERI, Āhau and ACDC.

Data Empowerment

Promise: Web3 claims to let data creators “own,” and likely profit from, their data, while protecting their privacy.

Redundancy: The typical vision of data ownership focuses on a private property conception of data in “personal data stores” that can be freely transacted and that interface with the “market” via a DeFi structure. However, such structures are unlikely to facilitate data empowerment beyond a narrow set of cases, for several reasons, including:

  1. Most data are relational (e.g., emails between people, genetic data partially shared by families, social graph data), so private property conceptions fail. If any individual can block transactions, data becomes unusable; if any can authorize transactions, a race to the bottom results as each data holder tries to sell out ahead of others.
  2. Most uses of data rely on aggregation, limiting the bargaining power of individuals absent collective organization in much the same way that workers required collective bargaining in the industrial era.

Subsidiarity: An emerging subsidiary model of data management combines social and legal constructs like data cooperatives, collaboratives, and trusts with privacy-preserving and -enhancing techniques for data processing, such as federated learning and secure multiparty computation.

In these models, cooperative organizations accountable to and socially connected to data subjects (ranging from local governments to worker cooperatives to credit unions) collect and steward socially entangled interpersonal data. These organizations could negotiate with companies and other entities to establish guidelines around the use of shared data. For example, a credit union could function as a data steward for member data, and exchange only particular insights with a startup building a tool for loan refinancing or with a public sector agency aiming to improve financial policy, keeping the underlying data private while adding value to the ecosystem and rerouting benefits to members. Such a steward could further interoperate with a network of other credit unions for better leverage and benefits. A similar structure could be used for needs as diverse as Covid-19 contact tracing or tracking carbon emissions, unlocking massive public benefits while still protecting individual and community decisionmaking.

Organizational Innovation

Promise: Web3 and specifically the Distributed Autonomous Organization (DAO) movement promise flexible, lightweight, accountable organizations and the empowerment of peer-to-peer, holocratic communities.

Redundancy: Fully automated organizations have thus far failed because of the inability to specify relevant contingencies in bug-free code. Flexibility and automaticity are in fundamental tension because perpetually automated processes are highly vulnerable to errors of oversight or faulty predictions, and can’t adapt when encountering cases not covered in the initial code. After all, automaticity is about generalized and repeated rules; flexibility requires the opposite. In human cognition, flexibility comes from case-specific judgments and creating exceptions. DAOs have thus relied heavily on governance, but without identity infrastructure, they have mostly relied on one-token-one-vote structures that can easily be attacked by, for example, venture capitalists who gain control of 51 percent of tokens. DAOs that rely on more informal, usually Web 2.0-based governance structures, may (and sometimes do) succeed, but in doing so, they often repeat rather than revolutionize existing organizational structures.

Subsidiarity: Partly fueled by discontent with current DAO structures, the platform cooperativism, exit-to-community, meta-governance, RadicalxChange, and other related movements have been developing tools like community currencies, soulbound access tokens, innovative voting systems like quadratic voting, new democratic deliberation tools like Pol.is and Loomio, and new ways of funding emergent democratically accountable organizations like quadratic funding and Gitcoin. These tools focus on community participation and empowerment, federating together organizations to build larger-scale cooperation rather than acquisitions or purely financial contracts. While certain functions are most efficient when automated, these processes enable the adaptivity that is crucial for organizational functioning.

Web3 has provoked an important discussion about decentralization. However, it is time to harness this energy to achieve what is best about decentralization: subsidiarity, not redundancy—a network of networks, not a ledger.

A lot can be achieved with limited deviations from the current Web3 ecosystem. Non-transferable “soulbound” tokens, which remain with the initial issuer and are thus non-financialized, have significant potential to allow social identity and community self-government even if restricted to cases (e.g., things one puts on a public CV or would include in a tweet) where privacy is not a primary concern. The portability and transparency of such trust-based, non-transferrable tokens in such a soulbound ecosystem could enable not only a more personhood-based web, but also allow for innovative voting and community governance protocols. In fact, many such protocols are already flourishing in imperfectly secured forms in experiments with, for example, quadratic voting and funding, voting and collective funding mechanisms that enable far more granular preference expression and aggregation. Multi-signature accounts are empowering community asset management. There are a range of attempts to build data unions and the like relying heavily on blockchains. It seems plausible that various privacy-enhancing technologies may be combined with the existing Web3 ecosystem to extend these further and work toward a more subsidiary structure.

More ambitious projects suited for subsidiarity hold even greater potential in solving meaningful coordination challenges. These are likely to face several obstacles in achieving scale:

  1. They serve neither the power of current profit-seeking incumbents, nor do they encourage speculation in globally fungible currencies based on their promise of global revolution.
  2. Their path to implementation is longer-term and more socio-technical, given that they deliberately work with and draw on existing social structures rather than trying to replace them.
  3. They connect less directly to the energy around blockchains at present and thus benefit less directly from the resulting hype.

Yet we view these challenges as a call for concerted, multi-sectoral investment. While this path may be less straightforward, it also has far more transformative social potential. Today’s internet grew out of a multi-sectoral collaboration between the US government, the academy, industrial research labs, and internet service providers, providing support based on a public mission for subsidiary decentralization. Taiwan’s transformative digital democracy and unparalleled Covid response have grown out of collaborations between government, socio-political activism, and support from the private sector. The major coordination challenges of our time, from crisis response to global governance, require well-designed subsidiarity to succeed at scale. We believe the elements of subsidiarity outlined here can provide the foundation for such a system.