Researchers Uncover Hidden Ingredients Behind AI Creativity
1 min read
Summary
A paper by two physicists, due to be presented at the International Conference on Machine Learning in 2025, has claimed that technical imperfections in the AI creative process are to blame for the creativity of diffusion models.
The models work by converting images into digital noise, or an incoherent collection of pixels, before reassembling them.
The researchers developed a mathematical model that shows an 80-90% correlation between the ELS machine - which optimises the technical imperfections - and powerful diffusion models.
This demonstrates that the creativity displayed by diffusion models is a deterministic process – a direct consequence of their architecture.
Experts agree that much remains unknown about AI creativity, and it is not clear whether the discovery of these links will have any practical applications in the short term.
