Generative artificial intelligence evolves from conception to application through roughly three phases: The first is the collection process, involving access to and replication of existing texts, achieved through acquiring public materials, purchasing corpora, and scraping data to gather foundational materials for AI training. The second phase is the training process, where the aforementioned materials are analyzed and processed to train a big data model that exhibits certain behavioral patterns. The third phase involves responding to user inquiries by generating textual information. These three links are interconnected yet distinct, encompassing various utilizations of “text,” “information,” and “data” by AI “technology developers,” “service providers,” and “users.” This forms a “Three-Three Matrix,” making the copyright infringement issues of generative AI complex and diverse, necessitating a reevaluation of this “Three-Three Matrix” to restore the original usage of works by artificial intelligence.