The research paper "Delegations as Adaptive Representation Patterns: Rethinking Influence in Liquid Democracy" was co-authored by Davide Grossi and Andreas Nitsche as part of the PERYCLES project. Nitsche presented the paper at this year's European Network for Digital Democracy conference, held on June 12 at the Centre d’Économie de la Sorbonne in Paris. He opened with a thought-provoking challenge:
“What if everything you fear about liquid democracy is caused by looking at it the wrong way?
Most models treat delegation as a static graph—but in reality, it’s a process. And when we study real-world liquid democracy, we find that transitive influence doesn’t spiral—it shrinks, fast. Exponentially.
That means the guru problem? Mostly fiction. Cycles? Beneficial at most. And many ‘fixes’? Counterproductive. Liquid democracy is a system that learns, adapts, and self-regulates.”
In their paper, Grossi and Nitsche argue that liquid democracy is fundamentally a division of labor system—allowing individuals to decide whether to engage directly in an issue or delegate their influence to someone else. While this influence is often interpreted as voting, they emphasize that delegation can apply to any quantifiable activity—from supporting arguments in deliberation to shaping collective moderation outcomes.
This broader view of delegation is implemented in the LiquidFeedback platform, where delegation affects not only vote tallies but also how input is weighed during deliberation. Yet this dimension has remained largely unexamined in academic literature.
Grossi and Nitsche characterize liquid democracy as a system intended to alleviate the workload on agents. This objective would be unattainable if the delegation structure had to be re-established from the ground up for each new decision, requiring every participant to make a delegation decision for every topic. To address this, LiquidFeedback employs delegation structures that evolve over time.
A hierarchy of delegations, an overlay of default delegations, policy area-specific defaults and topic-specific delegations, gives rise to multiple domain-specific representation patterns on how to address new decisions. Agents can delegate once, have their preferences apply broadly, and refine as they go. This further alleviates the burden on participants and enhances the practicability of liquid democracy.
Instead of requiring users to reconfigure their delegations for every new issue, the system builds domain-specific representation patterns. These patterns reflect the weight of potential agent activities and reveal an epistemic specialization—agents who repeatedly engage in specific domains may acquire greater potential influence. However, this does not constrain other agents’ ability to participate directly at any time.
Grossi and Nitsche introduce a new approach to assessing influence in transitive delegation graphs, recognizing that real-world liquid democracy is a process, where voting and delegation are distinct and increasingly independent activities. They show that transitivity, far from enabling unchecked power, can actually help regulate deliberative influence and decision-making. While maintaining the one-person, one-vote principle for all votes cast, the anticipated influence of an agent—where it results from transitivity—decays along an exponential trajectory.
In this context, Grossi and Nitsche also point out that interventions with commendable intentions can have unintended consequences.
Another key feature explored in the paper is the automatic suspension of delegations. In implementations like LiquidFeedback, when a participant acts directly—by discussing, supporting, or voting—their delegation for that specific activity is suspended, without needing to be explicitly recalled. This design preserves the agent’s autonomy while memorizing the delegation for future use. Suspension is key to decoupling the decisions about active participation from the management of delegations.
Cycles may occur inadvertently, but they are also sometimes intentionally created. These deliberate delegation cycles can serve as a practical safeguard against the risk of forgetting to vote. Contrary to common concerns, cycles do not reduce the number of votes cast—in fact, they increase the likelihood that a proxy vote will be cast on behalf of agents within a cycle, as well as for all agents in the delegation graph leading into it. Viewed this way, cycles are not a problem; if anything, they are beneficial.
Ultimately, Grossi and Nitsche aim to establish a foundation for a theory of liquid democracy that fully embraces its temporal and adaptive nature—opening paths toward richer models of influence, representation, and collective decision-making. By moving beyond static models, their work opens the door to more accurate analyses of influence, representation, and decision-making in participatory systems.
Access paper on arXiv | DOI: 10.48550/arXiv.2506.09789
The authors acknowledge support by the European Union under the Horizon Europe PERYCLES (Participatory Democracy that Scales) grant.
