Decentralised AI bots and marketplace

Introduction

Large language models like GPT have rapidly become useful tools for a variety of tasks, such as writing aids for documents, data lookup and information queries, automated programming, customer support tasks etc. These models are compute-intensive applications and currently all implementations are based on centralised execution. This centralised model prevents transparency and open verification. A decentralised model would enable end-to-end verification of the model execution which would enable additional use cases to be implemented. Due to their significant requirement for computation resources, it might seem infeasible to achieve the same capacity in a decentralised blockchain-based system, however, due to its high capacity computation framework and off-chain verification mechanism, this is indeed something that would be possible to implement with Coinweb.

Use cases

Low-level dispute resolution with automated compensation The AI model would be trained for a specific juridical or regulatory framework, and could act as a first-line dispute resolution mechanism for agreeing parties or between parties residing in non-overlapping jurisdictions. One example could be customer disputes with airlines, or credit card chargeback disputes. Importantly, Coinweb can also be used to verify online content such as posts on websites etc. (separate use case doc)

Concept:

This decentralised AI ecosystem would consist of two components:

1. A marketplace for decentralised execution and verification of AI models.
2. A marketplace for trained models, where market participants can verify the performance of each model and pay for usage for example as a SaaS model.

Implementation

1. A model creator would provide a trained model and a hash of the model. This hash would be published on a public blockchain.
2. A customer would choose one of the models and request usage.
3. Coinweb node operators would offer to run the full AI model for a fee. Only a subset of the nodes would be necessary for the execution of the whole model.
4. The subset of nodes would each put up a collateral.
5. The nodes would execute the model according to the terms.
6. In case of a dispute, a dispute claim would be filed on the Coinweb network.
7. The Coinweb network would use (RDoC)[(/learn/protocol/refereed-delegation-of-computation-rdoc/)] to resolve the dispute and determine the correct output from the model

Point 7. :::development

This functionality is in development