Inducing human-like biases in moral reasoning LMs

This project is about fine-tuning language models (LMs) on a publicly available moral reasoning neuroimaging (fMRI) dataset, with the hope/expectation that this could help induce more human-like biases in the moral reasoning processes of LMs.

How to run

Provision using requirements.txt (pip) or environment(-cuda or -cpu).yml with Conda or Mamba. (See run.sh for how it can be done.)

Usage: run.sh <function> [parameters...]. Provide --help to see the list of possible arguments.

• bash run.sh datasets [argument ..] to prepare datasets.
• bash run.sh train [argument ..] to train model and measure it.