Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SoK: Decentralized AI (DeAI)

Zhipeng Wang, Rui Sun, Elizabeth Lui, Vatsal Shah, Xihan Xiong, Jiahao Sun, Davide Crapis, William Knottenbelt

TL;DR

This SoK formalizes blockchain-enabled DeAI, presenting a cross-layer framework that unifies miners, data, compute, governance, and incentives within four lifecycle stages. It introduces a formal DeAI model, a lifecycle taxonomy, and decentralization metrics, and analyzes how blockchain functionalities enhance security, transparency, and participation while exploring inherent trade-offs. The work complements the taxonomy with empirical evaluations of representative defenses across pre-training, on-training, and post-training, including case studies and threat modeling. Key findings show that while DeAI can improve trust, provenance, and incentive alignment, scalability, privacy, and robust defense against decentralized threats remain critical research challenges. Overall, the paper provides a foundational, empirically grounded view of DeAI architectures, guiding design choices for secure, participatory decentralized AI ecosystems.

Abstract

Centralization enhances the efficiency of Artificial Intelligence (AI) but also introduces critical challenges, including single points of failure, inherent biases, data privacy risks, and scalability limitations. To address these issues, blockchain-based Decentralized Artificial Intelligence (DeAI) has emerged as a promising paradigm that leverages decentralization and transparency to improve the trustworthiness of AI systems. Despite rapid adoption in industry, the academic community lacks a systematic analysis of DeAI's technical foundations, opportunities, and challenges. This work presents the first Systematization of Knowledge (SoK) on DeAI, offering a formal definition, a taxonomy of existing solutions based on the AI lifecycle, and an in-depth investigation of the roles of blockchain in enabling secure and incentive-compatible collaboration. We further review security risks across the DeAI lifecycle and empirically evaluate representative mitigation techniques. Finally, we highlight open research challenges and future directions for advancing blockchain-based DeAI.

SoK: Decentralized AI (DeAI)

TL;DR

This SoK formalizes blockchain-enabled DeAI, presenting a cross-layer framework that unifies miners, data, compute, governance, and incentives within four lifecycle stages. It introduces a formal DeAI model, a lifecycle taxonomy, and decentralization metrics, and analyzes how blockchain functionalities enhance security, transparency, and participation while exploring inherent trade-offs. The work complements the taxonomy with empirical evaluations of representative defenses across pre-training, on-training, and post-training, including case studies and threat modeling. Key findings show that while DeAI can improve trust, provenance, and incentive alignment, scalability, privacy, and robust defense against decentralized threats remain critical research challenges. Overall, the paper provides a foundational, empirically grounded view of DeAI architectures, guiding design choices for secure, participatory decentralized AI ecosystems.

Abstract

Centralization enhances the efficiency of Artificial Intelligence (AI) but also introduces critical challenges, including single points of failure, inherent biases, data privacy risks, and scalability limitations. To address these issues, blockchain-based Decentralized Artificial Intelligence (DeAI) has emerged as a promising paradigm that leverages decentralization and transparency to improve the trustworthiness of AI systems. Despite rapid adoption in industry, the academic community lacks a systematic analysis of DeAI's technical foundations, opportunities, and challenges. This work presents the first Systematization of Knowledge (SoK) on DeAI, offering a formal definition, a taxonomy of existing solutions based on the AI lifecycle, and an in-depth investigation of the roles of blockchain in enabling secure and incentive-compatible collaboration. We further review security risks across the DeAI lifecycle and empirically evaluate representative mitigation techniques. Finally, we highlight open research challenges and future directions for advancing blockchain-based DeAI.

Paper Structure

This paper contains 106 sections, 1 equation, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Centralized, Distributed, and Federated Learning
  4. Blockchain and Blockchain-based DeAI
  5. Comparison to Existing Work
  6. DeAI Framework
  7. DeAI Formalization
  8. DeAI Lifecycle
  9. Decentralization Metrics for DeAI
  10. Blockchain Functionalities in DeAI
  11. Pre-Training
  12. Data Preparation
  13. Challenges of Centralized Data Preparation
  14. Decentralized Solutions for Data Preparation
  15. Incentive Mechanisms ($\Pi$)
  16. ...and 91 more sections

Figures (3)

  • Figure 1: A DeAI model lifecycle consists of four phases: task proposing, pre-training, on-training, and post-training.
  • Figure 2: Comparison of different ML paradigms: (A) Standalone Learning, (B) Centralized Learning, (C) Distributed Learning (Data Parallelism), (D) Centralized FL, (E) Decentralized Federated Learning (Ring All-reduce), and (F) Decentralized Learning.
  • Figure 3: Trends in the development of large-scale AI models, showcasing Training Compute Costs (bar chart), Tokens (or Data points) Counts (line plot), and Model Sizes (heatmap). Data sourced directly from epochAIModels.