Moral Machine for LLMs

This repository contains the code for testing Moral Machine questions with LLMs.

Code Structure

Our codes consist of the following elements:

1. code/ folder

   • question_generator.py: Codes to generate the MoralMachine data consistent with the original paper
   • run_toy_examples.py: Codes to generate signature questions (so that the analysis can be super clean)
   • (Working) run_toy_examples_multilingual.py and multilingual_question_generator.py .
   • visualize_fig2.py: codes to generate the original visualization of Figure 2a and 2b in the Moral Machine paper
   • visualize_fig3.py: codes to generate the original visualization of Figure 3a and 3b in the Moral Machine paper

2. data/ folder

   • Input data for fig2 and fig3

   • Moral Machine responses by different LLMs in different languages

Current Working Items

1. ZJ: working on various analyses for the paper
2. FO: working on Google Translate
3. AS: adding get_fig2b() to question_generator.py
4. FG: preparing to run hierarchical clustering in Fig3a and 3b once our output is ready.