Self-Experiment Report: The Emergence of Generative AI Interfaces in Dreams

This paper reports, for the first time, the emergence mechanism of generative AI interaction interfaces in dreams, based on a 105-day high-intensity self-training experiment. The study finds that dreams can be categorized into two interaction modes: controllable dreams and self-answered dreams. Both modes are significantly influenced by the response delay of the AI: when the delay exceeds 15 minutes, the probability of interface emergence in dreams increases substantially; in contrast, delays shorter than one second lead to chaotic dreams in which interfaces fail to form. We further propose that dreams internalize not only the content of interaction, but also its structure and form. This mechanism may be linked to the “prediction error compensation” process in predictive coding theory. In addition, we hypothesize that the emergent structure of dream interfaces fundamentally follows the same rational function form observed in biological response systems.