Constraints enable creativity, but limit it at the same time. We look at this apparent paradox as more than just philosophical wordplay: we treat it as an hypothesis that allows us to understand more about the potential and limits of creative processes. This project aims to understand constraint as a way to construct creative artificial systems. Similarly, studying creativity helps us understand and control the limits of machine learning systems.
We develop tools for the creative industry that allow for computer-aided creativity. Lessons learned from years of research in the field of procedural content generation for games are applied to architecture, music, fashion design, poetry and other fields. The central core of these tools is formed by deep learning technology.
Creativity requires originality and value. Artificial systems typically provide value to a user. In this project we build a minimal neural network system that generates goals for itself. Modeled after abiogenic life forms, this system allows for normative values relative to its own goals to emerge, rather than relative to the goals of its creator.