Conference Agenda

This paper discusses the benefits of and barriers to the successful creation of decentralized computer systems. The paper will argue that there are substantial benefits, both to society and to open markets, from systems with a decentralized structure. This paper is not intended to report specific research results. This paper is a “background” paper intended to provide a foundation, using insights from different disciplines, for discussions about the evolving shape of applications running on the Internet. These discussions will take on fresh urgency as we see AI systems evolving in different forms, and as they begin to interact with each other over the Internet to constitute new forms of decentralization or centralization. As regulators and policy-makers look at the powerful centralized systems of today, such as Facebook or Google, and contemplate how one or another intervention might reduce their market power or shift their incentives, it is important to understand what forms of these systems can be viable in practice.

In our definition, a decentralized computer system is one in which different parts are controlled by different actors. “Decentralized” is different from “distributed”. A distributed system can have elements on many parts of the Internet, but under the control of one actor. The Internet itself at the packet-carriage layer is a decentralized system, made up of about 77k “autonomous systems.” Early service layer elements like the DNS are decentralized, with many registries, registrars, and owners of top-level domain names. Email, the most important early application, had a decentralized design. The Web is decentralized. However, most of the important applications today, in particular social networking applications like Facebook or X, are centrally controlled.

Among the advantages of a decentralized system, the many actors that make up a decentralized system are a counterforce to the power that the single owner of a centralized system such as Facebook holds. However, centralized systems have many intrinsic advantages, and decentralized systems must overcome serious barriers to be successful.

In this paper, we provide an overview of the advantages that centralized systems naturally gain, and the barriers that decentralized solutions must overcome. We then focus on a particular design challenge, which is the economic viability of a decentralized solution. Technical standards bodies such as the IETF focus on the protocols that allow data to flow within a decentralized system such as email or the Web. But equal attention must be paid to the question of how money flows. This problem, sometimes called the “money-routing” problem, is critical. Anyone who considers being a part of a decentralized system must have an understanding of how they can recover costs and make money. And the rules must be stable enough to justify investment. The rules must either be enforced in some way, or be aligned with the incentives of the participants in the system.

This paper will review example of success and failure, and a range of options for how money can flow. It discusses various options for sources of money: advertising, philanthropy, volunteer efforts such as open source, and direct payment by users. Each of these has its own limitations. Using social networking as an example, we look at actual costs to deliver service, based on blogs and annual reports, to try to scope the sorts of revenue streams that must exist for a decentralized system to be viable. We find that the costs of operation are actually fairly low. To provide a Facebook-like service might cost $2.00 per month per user, and a lean system like Signal costs about $0.10 per month. But the users have become accustomed to paying nothing.

An important component of today’s social networking applications is that they provide (some) compensation to content providers. This compensation is derived either from advertising or from subscriptions from users. A decentralized system that can support payment for content will face additional design challenges.

We make some suggestions for the design of decentralized systems that may simplify the money-routing problem, some strategies for dealing with the small amounts of money involved per user, and a set of proposed design requirements for a stable money-routing system.