Courtroom

Currently, the game plays out as follows:

1. The player chooses an on-chain crime from a list (only DAO hack currently)
2. Player has an opening statement
3. Player can ask three questions to an imaginary victim (AI Agent) of the on-chain catastrophe. And interact with them, try to shift fault from their client over to the victims (DYOR kinda thing)
4. Player is allowed a closing statement.

The Judge (AI Agent) declares the defendant guilty or not guilty. The judge is very severe.

This is a showcase of what ZK-ML can help accomplish on-chain. What if everyone could play 0.001eth to try and defend the DAO hackers, or the DAO organization in this moral/law lawsuits? Every time the AI model tells the player they are guilty, they loose their money and the treasure chest grows. First winner takes it all!

I want to leverage two interesting innovations: ZK-ML and composability of Mud. My end-goal is to be able to have an on-chain courtroom where people can try and defend the culprits of catastrophic on-chain events. Using a law-trained LLM and ZK-ML, we can commit to a certain model and verify that the verdict rendered is done by some pre-determined model. Using MUD, we can have an ever-evolving game, where new "crimes" can be registered in the world, as crypto history unfolds.

Caveat: the LLM needs to be trained as time goes to have access to historical data about the case, or we must find a way to upload on-chain enough information about the case and prompt-inject it to the model beforehand (tbd).

The main goal is to have fun, see how we can incorporate soft rules into on-chain games, and mix it with fun play-to-earn.

The stack is:

• express server that calls chatgpt3.5 chat completion API
• client in React, with wallet connexion to create games and commit to the end result of the game (no access control implemented yet)
• Solidity smart contracts using MUDv2

MUD is really cool for composability, I want people to be able to add more "on-chain catastrophes" as time continues to go by. And somehow, upgrade the LLM so that the AI judge will be able to "judge" these new cases.