Defining Decentralization: An Ontological Perspective

This paper, Defining Decentralization: An Ontological Perspective, addresses the lack of a universally accepted, formal definition of decentralization in computer science, a problem it terms the Decentralization Problem. The authors argue that existing definitions are often context-dependent, technology-specific, or conflated with the related but distinct concept of distribution. To solve this, they introduce a graph-based ontology that formally defines decentralization as a structural, relational, and subject-specific property.

The paper identifies four core research problems: (1) Definitional Ambiguity, where the same system can be considered both centralized and decentralized; (2) Conceptual Confusion with Distribution, where the two terms are used interchangeably despite referring to different properties; (3) Lack of Structural Formalism, where existing approaches focus on node-centric quantities rather than the relationships between entities; and (4) Absence of Transferable Quantification, where metrics are often tied to specific architectures like blockchain.

The proposed solution is an ontology built on a graph-theoretic realism, where a system is a non-empty graph G = (V, E). The core ontology includes primitive concepts like System, Topology, and Subject, and primitive relations like hasTopology, hasSubject, and hasProjection. A key element is the projection particular (p), which links a system, a subject, and a topology. The framework defines two crucial functions: δ(pu), which counts the distinct realizations of a subject (its multiplicity), and λ(pu), which counts the distinct vertices supporting those realizations. This distinction is fundamental: Centralization and decentralization depend on δ, while distribution depends on λ, preserving the intended separation between multiplicity and spatial dispersion.

Based on this, the ontology formally defines:

• Centralized Subject: A subject where δ(pu) = 1.

• Decentralized Subject: A subject where δ(pu) > 1.

• Distributed Subject: A subject where λ(pu) > 1.

• Centralized System: A system where all subjects are centralized, which is the degenerate zero-dimensional case.

• Fully Decentralized System: A system with at least one subject and where every subject is decentralized.

• Partially Decentralized System: A system with at least one centralized and one decentralized subject.

The paper also introduces two novel graph-based metrics for quantifying decentralization: Void Tolerance (TL) and Imperviousness (IL). Void Tolerance measures vertex-focused decentralization by assessing the resilience of the subject-induced subgraph to vertex removal. Imperviousness measures edge-focused decentralization by assessing the minimum number of edge deletions required to compromise the connectivity of any subject-specific vertex. These metrics produce a two-dimensional decentralization vector, d⃗u = [TL, IL], for each subject, allowing for a more nuanced comparison than scalar measures.

The framework is validated through instantiations on Federated Learning and Blockchain architectures. For example, in a vanilla Federated Learning system with a star topology, the ontology classifies the system as PARTIALLY DECENTRALIZED because while data storage and model training are decentralized (δ > 1), the aggregation authority is centralized (δ = 1). In contrast, a decentralized federated learning setup with multiple aggregation authorities is classified as FULLY DECENTRALIZED. The analytical metrics further differentiate these systems, showing that the second setup has a higher aggregate decentralization score.

The paper concludes by discussing the limitations of the framework, including its focus on structural properties rather than behavioral or economic ones, and its reliance on graph-realistic representations. It also acknowledges that the specific functional forms of the analytical metrics are ad hoc design choices, though they are ontologically constrained. Ultimately, the work establishes decentralization as a formally definable relational property rather than an application-dependent or informally interpreted concept.

Improvements to AI Systems:

1. Ontology-Driven Decentralization Auditing: AI systems can be enhanced with a formal reasoning module that uses the proposed graph-based ontology to automatically classify any multi-agent or distributed system (e.g., federated learning networks, blockchain consensus, or microservice architectures) as CENTRALIZED, PARTIALLY DECENTRALIZED, or FULLY DECENTRALIZED based on the δ and λ functions. This enables AI to generate verifiable compliance reports for regulatory or governance purposes.

2. Resilience-Aware System Design: By integrating the Void Tolerance (TL) and Imperviousness (IL) metrics into AI-driven architecture optimization, AI can now propose topology modifications that maximize decentralization under adversarial conditions. For example, an AI system can simulate vertex/edge removals and recommend adding redundant aggregation nodes or alternative communication paths to improve TL and IL scores, directly improving fault tolerance and censorship resistance.

3. Context-Sensitive Decentralization Quantification: Instead of relying on scalar or blockchain-specific metrics, AI systems can now compute a two-dimensional decentralization vector (d⃗u = [TL, IL]) for each subject. This allows AI to compare heterogeneous systems (e.g., a permissioned ledger vs. a peer-to-peer file storage) on equal footing, enabling better decision-making in system selection or migration scenarios.

4. Disambiguation of Distribution vs. Decentralization: AI models can be trained to distinguish between these concepts using the formal separation (δ for multiplicity, λ for spatial dispersion). This improves natural language understanding in technical documentation, code comments, or governance texts, reducing misinterpretation in automated legal or policy analysis.

5. Subject-Specific Governance and Control: AI governance systems can use the ontology to identify which specific subjects (e.g., data storage, model training, aggregation) are centralized or decentralized within a system. This allows AI to enforce targeted policies—e.g., automatically flagging a system as non-compliant if any critical subject has δ = 1, or dynamically adjusting trust levels based on the decentralization vector of each subject.

6. Transferable Benchmarking for AI Research: The formal metrics enable AI systems to benchmark decentralization across different architectures without redefining metrics per technology. An AI meta-learner can use TL and IL as universal features to predict system robustness, scalability, or attack surface, improving transfer learning across domains like IoT, cloud computing, and decentralized finance.

What the improved AI system can do:

• Automatically audit and classify any networked system’s decentralization status in real time.

• Optimize system topologies to meet user-defined decentralization thresholds (e.g., ensure all subjects have δ ≥ 3).

• Generate explainable, formal proofs of decentralization properties for certification.

• Compare and rank candidate architectures (e.g., in a multi-cloud or blockchain selection task) using the same mathematical foundation.

• Detect and correct conceptual conflations in technical literature or user queries, improving AI’s reasoning about distributed systems.

Decentralization as a concept in computer science has existed for over half a century. Despite its fundamental role across domains such as security, distributed computing, artificial intelligence, cloud infrastructures, and Internet of Things (IoT) architectures, there remains no universally accepted definition of decentralization applicable across computer communication systems. This has become increasingly problematic with the emergence of decentralized AI and machine learning paradigms, including collaborative training, distributed inference, blockchain-based, and agentic AI, where decentralization is often treated as a core design objective. Meanwhile, existing approaches frequently conflate decentralization with related notions such as distribution of trust or specific implementation paradigms. Such ambiguity creates inconsistencies in system analysis, limits comparability between works, and weakens the rigor of formal reasoning surrounding communication architectures and protocol design. In this work, we define this research gap as the Decentralization Problem. We analyze the formal-semantic, epistemological, and pragmatic foundations of decentralization and introduce a graph-based ontology defining it as both relational and subject-specific property of computer communication systems. The framework formally distinguishes decentralization from distribution and supports evaluation through two novel metrics: Void Tolerance and Imperviousness. We also provide a browser-based implementation that enables automated classification and metric computation of arbitrary systems. Instantiations to federated learning and blockchain architectures show consistent, comparable assessments where existing definitions produce incomplete or contradictory conclusions, providing a domain-independent foundation for analysing decentralization across heterogeneous systems.