Feedback Loop Collapse of Artificial Intelligence

This paper examines the problem of feedback loop collapse in artificial intelligence systems, a phenomenon that occurs when AI models are repeatedly trained on data generated by other AI models. As generative systems become widely used for producing text, images, and other digital content, the proportion of synthetic data in training datasets is steadily increasing. Over time, this can lead to recursive training cycles where models learn from outputs that no longer reflect original human-created distributions. This paper also introduces and formalizes the concept of Corpus Entropy Decay (CED), defined as the measurable degradation of diversity, correctness, creativity, and robustness in outputs produced by models trained on recursively generated synthetic data.