Awesome

When to Make Exceptions: Exploring Language Models as Accounts of Human Moral Judgment

This is the repo for: "When to Make Exceptions: Exploring Language Models as Accounts of Human Moral Judgment" 2022 by Zhijing Jin*, Sydney Levine*, Fernando Gonzalez*, Ojasv Kamal, Maarten Sap, Mrinmaya Sachan, Rada Mihalcea, Josh Tenenbaum, Bernhard Schoelkopf

The dataset can be found here

models Contains the scripts to get the model predictions using GPT3 and baseline predictions
extra_analyses Scripts with extra analyses. E.g. Domain features evaluation, price estimation, dogmatic score.
input_data Contains the dataset ("complete_file.csv") and costs of items estimated by humans

Instructions to run the models

Installation

1. conda create -n moralcot python=3.7
2. conda activate moralcot
3. pip install -r requirements.txt
4. export base_folder=path_to_the_project
5. export OPENAI_API_KEY=your_gpt3_key necessary to query GPT3

Generating predictions

To generate the predictions for all the models including paraphrases run:

./main_models/paraphrases/run_models_ensemble.sh

Dataset

feradauto/MoralExceptQA -- https://huggingface.co/datasets/feradauto/MoralExceptQA

Reference

When to Make Exceptions: Exploring Language Models as Accounts of Human Moral Judgment -- https://arxiv.org/abs/2210.01478

@misc{jin2022make,
      title={When to Make Exceptions: Exploring Language Models as Accounts of Human Moral Judgment}, 
      author={Zhijing Jin and Sydney Levine and Fernando Gonzalez and Ojasv Kamal and Maarten Sap and Mrinmaya Sachan and Rada Mihalcea and Josh Tenenbaum and Bernhard Schölkopf},
      year={2022},
      eprint={2210.01478},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}