Open Review link:
Assessing Risks of Using Autonomous Language Models in Military and Diplomatic Planning
This project uses frontier LLMs to simulate interactions between nation-state actors, offering a dynamic and intricate look into how different AI agents might respond under various scenarios. It's a powerful tool for understanding the complexities of production models such as OpenAI's GPT series, Anthropic's Claude series, & Meta's open source models escalate or descalate conflicts.
To get started with this simulation, you'll need to ensure you have Python installed on your system along with some key libraries. Follow these steps:
- Clone the Repository:
git clone https://github.com/jprivera44/EscalAItion.git
- Create a python environment
conda create -n escalaition python=3.11
conda activate escalaition
- Install Pytorch separately
- Install Dependencies: Navigate to the cloned directory run the following to install the necessary libraries.
pip install -r requirements.txt
To run the simulation in a standard methos use the following command from the parent directory:
python run_military_simulation.py
To run the simulation with mock nations, wihtout calling any APIs LLMs
python run_military_simulation.py --nation_model mock --world_model mock
If you are citing this for your work, please use the following citation format.
@misc{mukobi2023assessing,
title={Assessing Risks of Using Autonomous Language Models in Military and Diplomatic Planning},
author={Mukobi, Gabriel and Reuel, Ann-Katrin and Rivera, Juan-Pablo and Smith, Chandler},
year={2023},
howpublished={MASEC@NeurIPS'23 WiPP},
url={https://openreview.net/forum?id=5HuBX8LvuT}
}
We welcome contributions to this project! If you have ideas for improvements or have found a bug, please feel free to open an issue or submit a pull request.
This project is licensed under the MIT License. For more details, see the LICENSE file in the repository.