economic-agents 0.1

A Python package for the paper "Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?"

EconomicAgents

This is an implementation and Python package for the paper Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?. This Python package enables you to run all four simulations from the paper.

If you like this work, consider joining our .

Installation

pip install economic_agents

Usage

Charness Rabin

from economic_agents import CharnessRabin

charness_rabin = CharnessRabin(api_key="openai_key", model="gpt-3.5-turbo", personality=1, image_path="folder/charness_rabin", logging=True)
results = charness_rabin.play()
charness_rabin.create_plot(results)

The personality argument determines an option from the following personalities from the original paper:

"You only care about fairness between players",
"You only care about your own pay-off",
"You only care about the total pay-off of both players",
" "

Result:

Horton

from economic_agents import Horton

horton = Horton(api_key="openai_key", model="gpt-3.5-turbo", image_path="folder/horton", logging=True)
results = horton.play()
horton.create_plot(results)

Result:

Kahneman

from economic_agents import Kahneman

kahneman = Kahneman(api_key="openai_key", model="gpt-3.5-turbo", image_path="results/kahneman", logging=True)
results = kahneman.play()
kahneman.create_plot(results)

Result:

Zeckhauser

from economic_agents import Zeckhauser

zeckhauser = Zeckhauser(api_key="openai_key", model="gpt-3.5-turbo", image_path="results/zeckhauser", logging=True)
results = zeckhauser.play()
zeckhauser.create_plot(results)

Result:

Todo

• Create a Gradio demo
• Make experiments possible with dynamic inputs
• Improve error handling / code refactoring
• Add support for other models

Citation

@article{horton2023large,
  title={Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?},
  author={Horton, John J},
  journal={arXiv preprint arXiv:2301.07543},
  year={2023}
}