Abstract:

Based on the characteristics of artificial intelligence （AI） technology， the current AI generation process relies on pre-existing works as training data to produce expressions according to given instructions. Consequently， there may be a high degree of similarity between the generated expressions and the original works. Under the existing Copyright Law， the determination of infringement is generally based on the criteria of “access plus substantial similarity.” In the context of the internet， the “access” requirement is implicitly sidelined， and the substantial similarity standard is applied from an outcome-oriented perspective， which may hinder the future utilization of AI. The randomness inherent in machine-generated creations deviates from the traditional rules for determining “substantial similarity.” The selection of databases and prompt words plays a significant role in the allocation of liability. Through the deconstruction and analysis of the AI generation process， greater emphasis should be placed on human intervention and guidance in infringement cases involving AI-generated contents. The scope of AI databases and the generation prompt words can also serve as considerations in infringement adjudications.

Key words: artificial intelligence, substantive similarity, creation path, Copyright Law