Project #3: Aesthetic Selection
I presented an interactive exhibit at the 2025 Envisioning AI at Yale Symposium titled "Aesthetic Selection: Applying Evolutionary Principles to AI Image Generation".
I wrote a Python script that used Stable Diffusion XL to generate three variations on three text prompts (e.g. beach at sunset, scenic mountain vista, futuristic city at night). Then, symposium attendees voted for their favorite within each group using a keyboard. Once an image reached a vote threshold, it was used as an input for img2img Stable Diffusion XL to create three more variations, and the cycle continues.
I had two main goals with this project: to learn more about building Python scripts that use generative AI and to learn how to integrate high-performance computing clusters with my pipelines. My friend and colleague, Sam Friedman, showed me the ropes for running interactive scripts on Yale's McCleary cluster, including a neat X forwarding trick that is the bane of cluster admins everywhere (don't worry, we had permission!).
This project taught me so much about building AI pipelines, version control (dependency hell?), and high-performance computing. Beyond the tech insights, I hope my biologist perspective gave attendees some new ideas on how stochastic image generation allows for the application of selection pressure towards some specified goal. Here are a few discussion questions I included in my exhibit:
How is this similar to evolution? What are the mutations? What is the selection pressure?
How might stochastic methods of creating art change the field?
What does the democratization of art mean for artists?
