In the AI Era, We Are Still Asking ''Who''
We have long been accustomed to one question: "Who infringed?" This question was long valid — creators were clear, infringers were identifiable, and responsibility was attributable to a single subject. Law was also designed on this premise. However, AI circumvents this question. Infringement is no longer explained as a single act — it occurs within a flow, distributes through processes, and only appears as a "problem" at the moment of distribution. A signal starting from data passes through algorithms, receives direction through prompts, and through platforms appears as one "issue" at the moment of distribution. Numerous subjects are involved in this process, but it''s difficult to definitively identify any single one as the infringer alone. So the question must change: not "who did it?" but "where did it begin?" Three metacognitive observations: (1) Is training memory or infringement? AI passes through vast flows of text and images — reading, forming patterns, then reconstructing them. This process is not dissimilar to human learning. The question: is AI''s training "memory" or "copying"? When a human reads extensively and writes, we call that "influence." When AI trains on data and generates content, the legal framework calls it potential infringement. The asymmetry reveals that copyright law is designed around human intentionality and market displacement — categories AI''s mechanical processing doesn''t fit neatly into; (2) Is generation creation or simulation? AI-generated output is statistically plausible continuation of patterns in training data — is this "creation" or "sophisticated pattern simulation"? The distinction matters because copyright requires originality, and originality requires a human author making creative choices; (3) Whose responsibility at distribution? The training company? The platform enabling deployment? The user whose prompt directed the output? The legal system''s subject-centered liability framework needs redesign for distributed AI systems where harm emerges from the combination of training decisions, deployment context, and user direction.