Delegation and Participation in Decentralized Governance: An Epistemic View

TL;DR

This paper poses a fundamental question: can decentralized governance achieve high epistemic performance comparable to centralized systems? It develops two epistemic tests—one for independent voter competencies and one for dependent competencies—to evaluate methods like partial abstention, direct participation, and transfer delegation. It finds partial abstention is often the most robust approach under independence, while transfer delegation faces serious coordination and dependency challenges; direct participation can harm epistemic accuracy unless placed in an near-optimal weight environment. To address these challenges, the work explores supplementary mechanisms—prediction markets (futarchy), Condorcet AI agents, and contestable control—arguing that hybrid systems may better align decentralization with effective decision-making in competitive environments. The findings offer a framework for assessing DAO governance and highlight practical avenues to improve outcomes without sacrificing decentralization.

Abstract

We develop and apply epistemic tests to various decentralized governance methods as well as to study the impact of participation. These tests probe the ability to reach a correct outcome when there is one. We find that partial abstention is a strong governance method from an epistemic standpoint compared to alternatives such as various forms of ``transfer delegation" in which voters explicitly transfer some or all of their voting rights to others. We make a stronger case for multi-step transfer delegation than is present in previous work but also demonstrate that transfer delegation has inherent epistemic weaknesses. We show that enhanced direct participation, voters exercising their own voting rights, can have a variety of epistemic impacts, some very negative. We identify governance conditions under which additional direct participation is guaranteed to do no epistemic harm and is likely to increase the probability of making correct decisions. In light of the epistemic challenges of voting-based decentralized governance, we consider the possible supplementary use of prediction markets, auctions, and AI agents to improve outcomes. All these results are significant because epistemic performance matters if entities such as DAOs (decentralized autonomous organizations) wish to compete with organizations that are more centralized.