This is a text-based version of the game Little Alchemy 2 that can be played both by humans and Large Language Models. We have implemented it by extending Wordcraft, a python-based implementation of Little Alchemy 2 that enabled playing the game with reinforcement learning agents.
How did we extend Wordcraft?
- added open-ended tasks. These start with the same items that Little Alchemy 2 starts with and have no target item
- added support for multiple agents. The agents are not in the same environment but receive information about the actions of others
- added ability to encode words into random stringsollama
- engineered prompts for instructing LLMs to play the task
- some bug fixes (eg ensuring that tasks are set deterministically by the seed, dealing with items missing from the data base)
We have tested our code on Linux and Mac with Python version 3.11
To install all necessary package dependencies you can run:
conda env create -f environment.yml
To enable the use of LLMs we are using the ollama library. This is an API to various opensource LLMs, such as LLama 3, Phi 3 Mistral and Gemma 2.
We have provided a script for illustrating how humans and LLMs can play a game. Upon running:
python play.py
you will be asked how many human and LLM players there will be. Human players will need to perform actions through the command line.
If you use this code in your work, please cite our paper:
@article{nisioti_2024,
title={Collective Innovation in Groups of Large Language Models},
author={Eleni Nisioti and Sebastian Risi and Ida Momennejad and Pierre-Yves Oudeyer and Clément Moulin-Frier},
year={2024},
booktitle = {The 2023 {Conference} on {Artificial} {Life}},
publisher = {MIT Press},
}