From Here to Chair: Exploring Human Centered Creativity & Generative A.I
Wallis Millar-Blanchaer
Advisor: Simone Salvo
From here to Chair explores the intersection of human creativity and machine learning through chairs—one of humanity’s earliest assistive technologies.
Project Description
***From here to Chair*** explores the intersection of human creativity and machine learning through chairs—one of humanity's earliest assistive technologies.
Using a dataset of 100 original speculative chair drawings, created by the artist, the project involves fine-tuning a vision model and analyzing how the AI interprets, transforms, and regenerates these concepts. This project investigates what happens when we remove chairs from their spatial and functional context, asking: What is a chair if you don't have to sit in it? How does a machine understand a chair if it can't sit?
Through this cyclical process of creation, machine learning, and regeneration, this thesis explores questions about algorithmic homogenization, the transformation of creative labor, and whether AI tools enhance or atrophy human creativity.
Technical Details
From Here to Chair employs a two-stage machine learning process, exploring a recursive relationship inherent in human-computer creative explorations.
The workflow begins by fine-tuning a generative text to image vision model, on a dataset of 100 original speculative chair drawings, created by the artist. The fine-tuned model is then used to generate images of chair, either by the artist or user generated prompts. The resulting image is then passed to a compact vision-language model (Moondream2), which analyzes the AI-generated image to create a text description of what’s in the image.
These AI-generated descriptions then serve as prompts to regenerate new chair designs, creating a feedback loop of creation, interpretation, and recreation.