Art and artificial intelligence find their intersection in the work of Audrey Zhang

Written by
Allison Gasparini
April 29, 2024

Audrey Zhang, Class of 2025

While she was still in high school, Audrey Zhang began to become concerned about the question of how artificial intelligence might be affecting our world. Senior year, she took an advanced computer science class and began coding on her own time. “I got this book, Make Your Own Neural Network,” said Zhang. “That was cool because I ended up building a model.”

However, in spite of her blossoming interest in machine learning systems, when she arrived on the Princeton University campus, Zhang, who illustrated and wrote her own webcomic, decided to go for a major in Art and Archaeology. Then, by chance in 2022, Zhang had lunch with a friend who encouraged her to take a course with Center for Statistics and Machine Learning lecturer Daisy Yan Huang.

Zhang enjoyed her experience taking Statistics and Machine Learning 201, Introduction to Data Science with Huang so much that she decided to fully pursue a Statistics and Machine Learning (SML) minor. Now a junior, Zhang is wrapping up her independent work project which intertwines her interest in AI and art. She decided to train a text-to-image deep learning model called Stable Diffusion on art from one of her webcomics titled Goddess of Chaos. The model is pre-trained on data taken from around the internet, but by adding her own, new data, Zhang is practicing “fine-tuning” a neural network to see just how specialized she could make the outputs. “I wanted to see if the AI could generate images in my art style,” said Zhang. 

side by side comparison of Zhang's art and the art generated by the Stable Diffusion model

On the left, artwork from Zhang's original webcomic Goddess of Chaos. On the right, an image generated from the Stable Diffusion model fine-tuned by Zhang for her independent work. Images courtesy of Audrey Zhang

After feeding Stable Diffusion just ten examples of her artwork, Zhang found the model could already generate images very similar to her own (though, as of yet, she can not be sure if the similarities are purely due to fine-tuning or random chance). “That was pretty exciting,” she said. “And also a little concerning.”

Models like Stable Diffusion are trained on images taken from all around the internet, and, controversially, the original artists aren’t being compensated for this use of their work. This blurry territory at the intersection of art and AI is of particular interest to Zhang, as an artist herself. She’s hesitant to say she’d use AI in creating her own art. “If I ever made enough art to train a network I would want to make sure that the model was trained on truly copyright-free material or trained on my art only,” said Zhang. “Because, then, I could fine-tune it without worrying about the model accidentally copying its training data directly.” 

This interest in the ethics of art and AI is what ultimately inspired Zhang to use her independent work project exploring just how good an AI algorithm can be in recreating unique art styles. “The rate at which AI is accelerating and improving is something we haven’t seen before in other technologies,” Zhang said. “A lot of the laws might not be catching up in time.”

Though Zhang is currently the only student at the University pursuing an SML minor alongside an art degree, it’s a combination that makes more sense than one might expect. In Zhang’s perspective, SML isn’t solely about knowing how to use machine learning models — it’s also focused on teaching students how to think about and approach problem-solving.  “Doing an SML minor has really improved my logical thinking skills,” said Zhang. “That’s invaluable for any major.”