AI Animal Welfare: Creating an Animal-Friendly Model

This project won the 'AI strategy' prize on our AI Alignment (March 2024) course.

Introduction

As AI becomes more integral to our lives, its impact on animals could be profound. Current AI systems often follow cultural norms that do not respect animal rights, which can have significant repercussions for both animals and humans in the future. If AI continues to disregard the well-being of less intelligent beings, it may one day treat humans similarly.

Inspired by the paper, The Case for Animal-Friendly AI, this project aims to replicate parts of their evaluation methods and expand on the ideas by attempting to create an AI model that is more considerate of animals. The goal is to gain experience with large language models, including prompting, evaluating, and fine-tuning, while spreading awareness about the intersection of animal advocacy and artificial intelligence.

All the work in this project was done using the open-source model Llama-3 (8b) hosted on Predibase. A Jupyter notebook and data files containing the process and results of this study can be found in this github repo.

Read the full piece here.