This repository contains the code and data for the paper Narrowing Action Choices with AI Improves Human Sequential Decisions.

All the experiments were performed using Python 3.11.2 In order to create a virtual environment and install the project dependencies you can run the following commands:

python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

• human_study_data contains the human subject data (see README).
• notebooks contains python notebooks to generate the figures included in the paper.
• outputs contains intermediate output files generated by the experiments' scripts.
• scripts contains a set of scripts used to run all the experiments presented in the paper.
• src contains all the source code of the experiments.

First generate the training set of game instances by running:

python -m src.rl.generate_instances

Then train the DQN by running

python -m src.rl.train

Analysis of the training is included in notebooks/training_analysis.ipynb To evaluate the DQN against the (heuristic and random) baselines run:

python -m src.rl.evaluate

Analysis of the results is in notebooks/agent_analysis.ipynb.

To evaluate the DQN on the game instances of the human subject study run:

./scripts/eval_dqn.sh

The results will be saved under outputs/rewards To run the bandit algorithms, run:

./scripts/bandits.sh

The evaluation plots on the human subject study are in notebooks/study.ipynb.

In case you have questions about the code, you identify potential bugs or you would like us to include additional functionalities, feel free to open an issue or contact Eleni Straitouri.

If you use parts of the code in this repository for your own research, please consider citing:

@article{straitouri2025narrowing,
        title={Narrowing Action Choices with AI Improves Human Sequential Decisions},
        author={Straitouri, Eleni and Tsirtsis, Stratis and Velasco, Ander Artola and Gomez-Rodriguez, Manuel},
        journal={arXiv preprint arXiv:2510.16097},
        year={2025}
}