AI Juries on Evernode

We’d like to introduce our latest project: an innovative approach to automated dispute resolution and moderation for decentralized online communities using AI agents on Evernode. A more substantive paper and video demo of our working toy is included here.

The Core Problem and Solution

Decentralized online communities (like DAOs and dApps) need effective moderation and dispute resolution systems. Current solutions either rely on human moderators, which introduce bias and scaling issues, or use automated tools that lack legitimacy. Our proposed solution involves using AI agents deployed on a specialized blockchain platform called Evernode.

Technical Foundation

In January 2024, Evernode launched a platform that can turn regular applications into decentralized applications (dApps). It works by dividing Linux machines into hosting slots, each with a Byzantine Fault Tolerance consensus engine. This allows developers to create custom blockchains (AppChains) that maintain consistent state across multiple instances.

Evernode’s key advantage is its flexibility – it can perform complex computations and support features not typically available in blockchain applications. This enables what the authors call “middleman-be-gone” applications, such as RoadRush (a decentralized Uber-like service) and TripQ (a decentralized hotel booking platform).

The Need for Dispute Resolution

Effective dispute resolution is crucial for online communities, using eBay as an example. eBay’s success largely depends on its ability to handle tens of millions disputes annually. Decentralized platforms need similar capabilities, but traditional blockchain-based dispute resolution mechanisms have been unsuccessful due to various limitations.

The Proposed Solution

We have developed a proof-of-concept system using Evernode that combines multiple AI agents to form a decentralized jury. Their working prototype includes:

  • Three independent AI agent instances running on CPU-based machines
  • Each agent uses a fine-tuned GPT-4 model
  • The agents reach consensus through a Byzantine Fault Tolerance mechanism
  • The system successfully demonstrated the ability to make collective decisions about blog post moderation, with consensus achieved in approximately 3 minutes

We think having multiple AI agents agree on decisions could make AI Agents more reliable and legitimate and help minimize hallucinations and outliers, though this theory needs more rigorous testing.

Summary

Overall, AI Juries represents an early but promising approach to solving the challenge of decentralized moderation and dispute resolution, combining emerging AI capabilities with blockchain technology in a novel way.

Leave a Reply

Discover more from Lex Automagica

Subscribe now to keep reading and get access to the full archive.

Continue reading