Professor Surya Ganguli and Mason Kamb (physics PhD student) have made a startling claim on how AI ‘creates.’ Their paper, 'An analytic theory of creativity in convolutional diffusion models’ describes their mathematical model that shows creativity is in fact a deterministic process for trained diffusion models. Surya and Mason illustrate that it’s the technical imperfections in the denoising process itself that leads to the creativity of diffusion models. Surya is the lead researcher of the Neural Dynamics and Computation Lab.

Mason started his graduate work in 2022 in Surya Ganguli’s lab. OpenAI released ChatGPT the same year, causing a surge of interest in the field now known as generative AI. As tech developers worked on building ever-more-powerful models, many academics remained fixated on understanding the inner workings of these systems.

To that end, Kamb eventually developed a hypothesis that locality and equivariance lead to creativity. That raised a tantalizing experimental possibility: If he could devise a system to do nothing but optimize for locality and equivariance, it should then behave like a diffusion model. This experiment was at the heart of his new paper, which he wrote with Surya as his co-author.

Mason and Surya call their system the equivariant local score (ELS) machine. It is not a trained diffusion model, but rather a set of equations which can analytically predict the composition of denoised images based solely on the mechanics of locality and equivariance. They then took a series of images that had been converted to digital noise and ran them through both the ELS machine and a number of powerful diffusion models, including ResNets and UNets.

The results were “shocking," says Surya: Across the board, the ELS machine was able to identically match the outputs of the trained diffusion models with an average accuracy of 90% — a result that’s "unheard of in machine learning."

Excerpted from Quantamagazine.com, 'Researchers Uncover Hidden Ingredients Behind AI Creativity'