讲座 | Can AI Be Creative? Examples from Research in Computer Graphics

Recent advances in GenAI have revived an old question: can machines be creative, or do they merely reproduce and recombine what humans have created before? Computer graphics provides an interesting experimental ground for studying this question, since computational systems can now participate in several stages of the artistic process from perception and abstraction to conceptualization and physical production. I will examine machine creativity through examples from our research. Our painting machines initially translated digital images into physical paintings through computationally generated brush strokes and visual feedback. More recent approaches use semantic information and visual saliency to decide what should be preserved, simplified, or emphasized when an image is transformed into a robotic painting. This already raises a fundamental question: when does algorithmic optimization become an aesthetic decision? I will then discuss recent work, which moves from visual appearance toward conceptual creativity. Rather than simply transferring style or image content, our system extracts the abstract structure underlying a visual metaphor and transfers this structure to a new semantic domain using multiple cooperating AI agents. These examples illustrate a progression from machines that execute artistic decisions, to systems that make aesthetic decisions, and finally to systems that attempt to construct new visual concepts.