# Libraries, Integrations and Hubs for Decentralized AI using IPFS

**Authors:** Richard Blythman, Mohamed Arshath, Jakub Smékal, Hithesh Shaji, Salvatore Vivona, Tyrone Dunmore  
**arXiv:** [2210.16651](https://arxiv.org/abs/2210.16651)  
**Date:** 2022-10-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.16651v1/

## Key points

- AI developers rely heavily on centralized cloud services and compute environments
- Limitations of centralized systems include high costs, lack of monetization and reward systems, lack of control over data access, and difficulty in reproducing results
- Decentralized technologies such as Web3 wallets, peer-to-peer marketplaces, decentralized storage (IPFS and Filecoin), and compute power sharing platforms like DAOs can address these limitations
- Decentralized systems provide users with greater control over their data while reducing costs associated with centralization
- Challenges of decentralized systems include governance issues related to dispute resolution among gatekeepers who may not always be unbiased, and difficulties in managing fine-grained access control over assets
- Solutions proposed using blockchain-based technologies that enable immutable versions of code or models stored on IPFS networks
- Decentralized technologies offer potential benefits for the field of AI and could revolutionize the way data scientists interact with AI assets.

## Summary

The field of artificial intelligence (AI) requires significant amounts of storage and compute power, leading AI developers to rely heavily on centralized cloud services such as AWS, GCP, and Azure, as well as compute environments like Jupyter and Colab notebooks, and AI hubs such as HuggingFace and ActiveLoop. However, these services come with limitations such as high costs, lack of monetization and reward systems, lack of control over data access, and difficulty in reproducing results. To address these limitations, this report explores the potential of decentralized technologies such as Web3 wallets, peer-to-peer marketplaces, decentralized storage (IPFS and Filecoin), compute power sharing platforms like DAOs. The report showcases some libraries and integrations built to tackle these issues along with a proof-of-concept for a decentralized AI hub app that uses IPFS as its core infrastructural component. Decentralized technologies offer several advantages over centralized systems. They provide users with greater control over their data while reducing costs associated with centralization. Decentralized systems also enable fine-grained access management for owners who need to manage access to their assets while ensuring privacy. However, decentralized systems have their own set of challenges such as governance issues related to dispute resolution among gatekeepers who may not always be unbiased. It is also challenging for owners to manage fine-grained access control over their assets since traditional access tokens like OAuth 2.0 or API keys can be widely shared without permission. Existing hubs like GitHub for AI make reproducibility difficult due to limitations in versioning models or datasets. This report proposes solutions using blockchain-based technologies that enable immutable versions of code or models stored on IPFS networks. In conclusion, this report highlights the potential benefits of using decentralized technologies in the field of AI. It proposes solutions to address the limitations of centralized systems and offers a glimpse into the future of decentralized AI hubs that could revolutionize the way data scientists interact with AI assets.