Skip to content

README.md

Latest commit

 

History

History
52 lines (28 loc) · 1.83 KB

README.md

File metadata and controls

52 lines (28 loc) · 1.83 KB

AI Moot Court Project

This project is forked from Chatarena, with some modifications, to simulate a moot court environment using three AI agents.

Getting Started

Note that Python >= 3.7 is required.
To install and run the project:

pip install chatarena[all]  
git clone https://github.com/HKUGenAI/AIMootCourt.git  
cd AIMootCourt  
./start.sh

Additionally note that the original OpenAI API call was changed to Azure OpenAI, in

AIMootCourt/chatarena/backends/openai.py

You can convert it back to OpenAI API by following the original Chatarena code, if necessary.

To use Azure OpenAI in your project, modify the following in start.sh :

export AZURE_OPENAI_API_KEY="your-api-key-here"  
export AZURE_OPENAI_API_ENDPOINT="your-api-endpoint-here"  
export AZURE_OPENAI_API_DEPLOYMENT_NAME="your-api-deployment-name-here"

Modifying the case

4 prompts are used to control the moot court simulation:

  1. A global prompt, which consists of the court rules for the AI to follow, as well as details and facts of the case.
  2. A role description for the Plaintiff.
  3. A role description for the Defendant.
  4. A role description for the Judge.

To use our pre-built case, on the Web UI, select "NLP Classroom 3 Players" from the dropdown menu "Environments".

To use your own test case and prompts, you can modify the text directly on the Web UI for a single trial, but the changes will not be saved. To change the environment directly so the prompts are saved, you can modify the following file:

AIMootCourt/examples/nlp-classroom-3players.json

You may use the python file at AIMootCourt/convert_to_string.py to help convert a text prompt to the required json format.

Finally, click "Run" on the Web UI to start the simulation.