GIP-63: Should GnosisDAO award a Grant to Algovera to develop a DAO framework for Decentralized AI Teams?

GIP-63: Should GnosisDAO award a Grant to Algovera to develop a DAO framework for Decentralized AI Teams?

  • Let’s do this!
  • Make no changes

0 voters

GIP: 63
title: A DAO framework for Decentralized AI Teams
author: Silent Spring (silentspring30@gmail.com), Richard Blythman (@richardblythman), Hithesh Shaji (@hithesh98)
status: Draft
type: Funding
created: 2022-08-25

Simple Summary

Algovera is requesting a grant of $50,000 to develop a DAO framework (on top of Gnosis smart contracts) for decentralized AI teams. This will involve building a Proof of Concept for publishing and co-owning AI assets in a trustless manner, receiving profits generated by assets to the treasury, and sharing profits generated with team members.

Algovera

Algovera is a community of independent AI teams developing user-centric Web3 AI applications. We’re building an AI creator platform where teams and their community can keep ownership of and monetize their creations. We provide support to teams including funding, community events/workshops and our decentralized AI framework (DAID) provides guidance in how to build a decentralized AI DAO that is based on a vision of cooperation, sustainability and ethics. Algovera Grants has funded 28 independent AI teams (called Squads) to date. We also train squads to use our decentralized AI infrastructure, which consists of decentralized storage, compute and marketplace (developed through grants from Ocean Protocol, 12+ OceanDAO grants to date and 1 Ocean Shipyard grant, and 1 Filecoin grant).

Abstract

DAO tooling can help to onboard AI developers to Web3, and improve coordination and governance within distributed AI teams. We have been setting up Squads that we fund within DAOs (currently DAOhaus) and have run into a number of problems. It is not possible for AI squads to publish and co-own IP related to data, algorithms and apps in a trustless manner, and there are no suitable tools for sharing profits generated by assets. Furthermore, existing tools are targeted at generic Web3 users, rather than data and AI teams. A UX study that we conducted showed that ownership, monetization of datasets and AI models are the most important at this time for our squads and communities.

The aims of this project proposal are to (i) build a Proof of Concept for using Gnosis to enable the publishing and co-owning of data and AI assets, as well as sharing of profits generated by assets and (ii) onboarding Algovera Squads to Gnosis and ML teams.

Motivation

AI development is currently controlled and owned by large tech companies within the singular vision of human-competition, autonomy and centralization, without consideration for society at large. We experienced this ourselves and would like to change this. We believe that AI is a common good and communities should have ownership. Although there are tools to develop AI models such as Kaggle or Huggingface, it is difficult for AI teams to keep ownership and create sustainable revenue for AI DAO communities.

AI teams have hardly any experience with Web3, Gnosis and DAO frameworks. Through our research and qualitative interviews, we found there is some awareness in the value of decentralized AI specific tools and architecture but most tools do not exist yet. Some of the biggest pain points for AI developers are lack of ownership around what they create in universities and tech companies, and lack of infrastructure for coordinating and monetizing their creations online.

Specification

Work Package 1: Proof of Concept to publish data sets and AI models for monetization of marketplaces

During this work package, we will implement a Proof of Concept for trustlessly publishing and co-owning IP NFTs for datasets and algorithms through a Gnosis Safe app for AI DAOs, receiving earnings generated by assets into the treasury, and distributing earnings back to the contributors/members of the team. Publishing of the asset will be performed by the multisig, with the transaction being executed after the policy defined by the multisig has been satisfied. The asset will be published with the multisig wallet address as the author and any fees associated with the consumption of the asset will be transferred to the multisig treasury. The final step to the solution is implementing profit-sharing tools to distribute earnings back to the contributors/members of the team.

Overview of goals:

  • Build a Gnosis Safe app for AI DAOs
  • Publish datasets and algorithms to Algovera/Ocean Marketplace through app
  • Receive earnings from marketplace into the Gnosis treasury
  • Distribute amongst DAO members
  • Documentation and video guide to onboard users

Work Package 2: Onboarding of Algovera Squads to Gnosis and decentralized AI strategy awareness and education

In this work package, we will run regular Decentralized AI DAO (DAID) workshops to not only develop Web3 awareness e.g. onboard our squads to Gnosis safe and monetization tools but also to expose them to pluralistic strategies of how to develop decentralized AI organizations. We believe that bringing together GnosisDAO and AlgoveraDAO squads would be mutually beneficial to the community.

Current centralized AI systems do not actively consider community ownership or how to integrate gender, racial or sustainability factors in developing AI models. Furthermore, researchers have highlighted the harm of AI models that are benchmarked in human-competition, autonomy and centralization [How AI Fails Us: Divya Siddarth, Glen Weyl et al. 2021]. They propose a pluralistic vision of developing AI models that complement, cooperate and support society rather than compete with humans.

This is why we have developed a Decentralized AI DAO framework (DAID) which accelerates our squads with technical tools and strategies that are rooted in interdisciplinary team work, human-centered design, ethics, sustainability and community commercial models.

Our goal is to host 2x interdisciplinary workshops with members of the GnosisDAO, Ocean Protocol, IPFS and Algovera communities for six months, create 6 videos and 5 tutorials to onboard our Squads to Gnosis, and to foster interdisciplinary teams to push towards the development of AI datasets and models for Web3 projects and DAOs.

DAID consists of 4 key areas:

  1. Primary - Design Thinking: who is the user and who is impacted by my AI model?

  2. Community Value Generators: How to achieve sustainable ethical, environmental and commercial value creation.

  3. Decentralized AI activities: from data sourcing, storage, training, deploy to publishing AI models to market places.

  4. DAO activities: covers how to set up a DAO, governance, proposals, treasury management and member protection and monetization of data sets and AI models

Implementation

The implementations must be completed before any GIP is given status “Final”, but it need not be completed before the GIP is accepted.

GnosisDAO Snapshot

Phase 2 Proposals: Please ignore this section, and leave as is. It is used for Phase 3 proposals.
Phase 3 Proposals: Add a link to the corresponding GnosisDAO Snapshot poll you’ve created.

10 Likes

Hi Richard - Welcome to the Gnosis Forum :slight_smile: When you’re ready to edit your post to Phase-2 you can reference the README: GnosisDAO Governance Process

You shouldn’t have a problem editing your post. If you run into any problems please feel free to message me.

3 Likes

Great organization with vision and a track record of producing results. As an academic researcher in machine learning I’m proud to work with them. Please help them take the next step!

4 Likes

Richard & team are absolutely brilliant & are able to attract the best & brightest across the entire ecosystem. This is a crystal clear win-win

4 Likes

yeaaa super excited to see Algovera here, in Eden :deciduous_tree: and Developer DAO we are obsessed with you guys!

2 Likes

As I understand your proposal besides educating AI Teams about web3 and DAOs you will build contracts which can be used to keep intellectual property (and the ability to monetize) within the DAOs or even single DAO members? Does this mean something like patent protection for the algorithms? Or will the contracts only protect the generated data? I am not too familiar with AI and the needs, so apologies if this is a stupid question.

3 Likes

Thank you! Richard just mentioned you the other day. Could I reach you on our Algovera’s discord?

1 Like

I’m intrigued by this too. The only way to protect IP is with patent or keep it secret. Maybe the future is one where IP has less value and instead it is the implementation that has value (but can be copied). Maybe the solution requires a trusted marketplace+infrastructure where I can run someone else’s model remotely and I don’t get direct access to their model.

It seems that AI is probably going to get very good at copying AI e.g. if I can train a model by running data through another model then is the new model breaking any IP/copyright protection ?

3 Likes

Hey @refri. It is a good question and certainly not a stupid question. We are handling IP of AI assets using Ocean Protocol. This section of the docs outlines and goes into depth how the contract handles IP & data- https://docs.oceanprotocol.com/core-concepts/datanft-and-datatoken. This section outlines algorithms and compute to data which is similar to what @MarkNZed mentioned - “ I can run someone else’s model remotely and I don’t get direct access to their model.” - https://docs.oceanprotocol.com/building-with-ocean/compute-to-data

However Ocean Market doesn’t directly integrate with safe apps. We are working on our own fork of Ocean Market that is tailored towards AI team (supported by the Ocean Protocol Team). Ocean does have a wallet connect provider but the publishing flow did not work when we tested it. Our fork of the market will include Safe App wrappers so that the marketplace can integrate more smoothly with Gnosis and has an easier UX for our users who have little experience with Web3.

Hope that answers your questions. Let me know if you have any further questions or want more clarifications.

3 Likes

It will be possible to protect many different types of assets like datasets, features, algorithms, apps, and notebooks (like interactive publications).

Another form of IP (other than patents) is copyright, and protection in this form is what I’m personally more interested in. I do think patents can be effective (and even necessary) in other fields such as biotech, which require large amounts of capital investment. VitaDAO is working on IP NFTs for this. However, I haven’t seen patents being productive in AI.

At the same time, there are many issues to be solved just around copyright. Licenses for code get broken all the time in my experience, and there are grey areas around training language models on copyright data (e.g. does interpolating between different copyright pieces of data - like DL models can do - count as infringement, should the owners of the training data be rewarded?). I tend to agree with @MarkNZed in having ownership and tracking provenance around “implementations” rather than ideas.

I think we likely need to re-write the rules around ownership of AI property from the ground up, and this DAO framework will help us to experiment with that in our community of real AI teams.

4 Likes

yes of course feel free too :smiley: BluePanda#5982

1 Like

I think we should definitely give Richard a chance !