The potential web3 provides for this particular domain of work is that you can have more, and more precise, layers of separation. Centralized work tends to bake AI deeply into whatever platform is built, whereas web3 generally allows for open source cores that implement very simple logic (you often can't put complicated models entirely on-chain) and then more complex AI alogrithms exist on "the outside". The key aim of our work is to make sure that:

The aim of these weekly hacking sessions is to go through the whole process of developing a useful decentralized AI algorithm owned by a community of individuals rather than one centralized party. This may include crowdsourcing additional data, developing new model architectures and training procedures, training on private datasets, using cloud compute, using new coordination and governance tools, publishing the model on the Ocean marketplace and creating a web app to demo it. At the end we will share ownership of everything that we create, so that the value generated by the algorithm flows back to the contributors.

Take some time to watch the conversation below with Emad Mostaque to get a sense of someone deeply embedded in the industry who has a similar approach to open source and the ethics it implies.

We're also indebted to Trent McConaghy, Gene Kogan, and Simon de la Rouviere for various thoughts which have helped us hone our initial thinking here. These can be found listed below and serve as a great place to get started if you're not familiar with what decentralized environments for artificial intelligence might look like.

This guild does focus on practical skills sharing, so the references here are largely just conceptual background rather than core context for the work presented in each session.

We're incredibly grateful to Richard Blythman and Algovera for the pioneering work done here. Algovera is a community of individuals working to facilitate and accelerate the development of decentralised AI products and research.