Glitch as Cognition: Rethinking AI’s Uncertainty as a Creative Force

Current artificial intelligence systems lack self-awareness or subjective agency, relying solely on probabilistic modeling and data-driven generative processes to produce outputs. However, I have been contemplating whether AI can exhibit a form of pseudo-subjectivity or whether it can help us transcend conventional cognitive boundaries. The mainstream perspective suggests that AI merely maps statistical correlations, but through my research on diffusion models (DMs) and generative adversarial networks (GANs) in image inpainting, I have found that the phenomenon of "glitch" may offer an entirely new perspective on AI’s cognitive potential.

The concept of glitch is difficult to define, while technical glitch generally refers to unexpected errors or abnormal behaviors occurring in technological systems, electronic devices, or software during operation. However, from the perspectives of information theory and artistic creation, glitches can reveal underlying structures within a system’s running process and even serve as sources of creative inspiration. For instance, in a documentary about 20th-century calligrapher Inoue Yuichi, I observed that the pixelation artifacts caused by low-resolution video did not obscure his artistic expression; instead, they unexpectedly highlighted the timeless essence of his brushwork. This accidental visual distortion led me to consider whether the uncertainty present in AI’s generative processes could serve a similar function.

My work utilizes AI-based image outpainting, expanding visuals outward based on glitches found in the documentary, such as pixelation and mosaic distortions. Rather than simply restoring missing parts, I use "glitch" as a guiding keyword to drive AI’s generative process, allowing it to extend the original footage beyond its intended frame. This approach transforms glitches from mere data loss into creative triggers, introducing new visual narratives shaped by AI-generated uncertainty. By leveraging implicit completion mechanisms, my work reframes AI-generated distortions as moments of creative potential, rather than errors, enabling a new form of storytelling that emerges through AI-driven expansion.