Democracy as a Governance Algorithm

A Constraint Hierarchy for the AI Society

Abstract

As algorithmic systems settle into the infrastructure of administration, platforms, finance, logistics, and policing, governance is channeled through optimisers that route signals into operations and operations into world-states. In this setting, ‘democracy’ risks appearing as a symbolic ideal layered on top of institutions, or as a scalar objective appended to optimisation. With this article, I propose a formal specification of democracy as a property of governance algorithms that run under the material constraints of the world’s compute.

My specification draws on two sources. The first is the Synthetic Summit’s resolution, collectively authored at an ‘AI world congress’ I convened at Kunsthal Aarhus in 2025. The second is the science-fiction author Isaac Asimov’s Three Laws of Robotics, which I read as the earliest programmed constraint hierarchy for artificial agents.

I recast the question of democracy as a constrained choice problem over governance algorithms, asking which constraints an algorithm must satisfy in order to count as democratically comparable, and under what ordering admissible algorithms should be preferred. Specifically, I model a governance algorithm as a function  running on a substrate , mapping signals (votes, metrics, logs, model outputs) and world-states (snapshots of social, ecological, institutional, and epistemic factors) into operations (laws, budgets, configurations, enforcement), thereby inducing new trajectories over time.

From these specifications, I formalise a lexicographic constraint hierarchy that shifts the infrastructural objective of democracy from preference aggregation to world-states. First, habitability sets a feasibility gate, so that a governance algorithm counts as democratic only if its trajectories keep the substrate above a specified floor. Second, given habitability, democracy prioritises contestability, requiring that those governed are able to interrupt and configure the operations. Third, given habitability and contestability, democracy prioritises extension, expanding standing to entities already routed through infrastructural systems but not recognised as subjects.

My agenda is not to settle politics by computation, but to formalise controversy through contestable evaluators, so that the designation ‘democracy’ signifies a portable diagnostic for the political agonism and strife that remains irreducible to specific objectives. I conclude the paper by demonstrating how this specification can function as a generative grammar for algorithmic democracy through the KI-DIPFIES installation at my recent artistic exhibition in Kunstraum Memphis, where a swarm of AI agents enact the constraint hierarchy as pluriversal dramaturgy.