# SoK: Decentralized AI (DeAI)

**Authors:** Zhipeng Wang, Rui Sun, Elizabeth Lui, Vatsal Shah, Xihan Xiong, Jiahao Sun, Davide Crapis, William Knottenbelt  
**arXiv:** [2411.17461](https://arxiv.org/abs/2411.17461)  
**Date:** 2024-11-26  
**Summary page:** https://www.summarizepaper.com/en/paper/2411.17461v1/

## Key points

- Centralization of Artificial Intelligence (AI) presents challenges such as single points of failure, biases, data privacy concerns, and scalability issues
- Blockchain-based decentralized AI (DeAI) emerges as a solution to enhance transparency, security, decentralization, and trustworthiness in AI systems
- Decentralized solutions for data preparation offer access to diverse and geographically distributed data without centralized storage through blockchain or federated learning frameworks
- Platforms like Ocean Protocol and Vana utilize tokenization mechanisms to incentivize participants to contribute high-quality data in decentralized systems
- Benefits of improved security and decentralized data management in DeAI platforms while addressing challenges like scalability and data privacy are discussed
- A taxonomy is provided to categorize existing DeAI protocols based on the lifecycle of an AI model, offering insights into their similarities and differences

## Summary

The centralization of Artificial Intelligence (AI) presents numerous challenges, including single points of failure, biases, data privacy concerns, and scalability issues. These problems are particularly prominent in closed-source large language models (LLMs), where user data is collected without transparency. To address these issues, blockchain-based decentralized AI (DeAI) has emerged as a promising solution. DeAI combines the strengths of blockchain and AI technologies to enhance transparency, security, decentralization, and trustworthiness in AI systems. However, as the demand for more powerful AI models grows, the need for vast amounts of high-quality training data is reaching its limits. Centralized AI systems rely on finite pools of publicly available data, leading to potential saturation and restricted model performance enhancements. Additionally, centralized data collection lacks diversity across domains and regions, resulting in biased models for various applications. Decentralized solutions for data preparation offer a way to access diverse and geographically distributed data without centralized storage. Through blockchain or federated learning frameworks, data providers can securely contribute their data while maintaining privacy and ownership. However, challenges arise in managing data quality without a central authority overseeing submissions. To incentivize participants to contribute high-quality data in decentralized systems,
platforms like Ocean Protocol and Vana utilize tokenization mechanisms. Data assets are tokenized into datatokens on marketplaces where users can purchase access to the underlying data. Staking and curation features further promote high-quality contributions by rewarding users who identify valuable datasets. By systematically reviewing existing DeAI solutions in practice,
this work discusses the benefits of improved security and decentralized data management while addressing challenges such as scalability
and data privacy. A taxonomy is provided to categorize existing DeAI protocols based on the lifecycle of an AI model,
offering insights into their similarities and differences. The analysis evaluates approaches to decentralization and security in DeAI platforms and protocols while identifying potential vulnerabilities for future research efforts.