Copyright Expectancy Right: Paradigm Reconstruction of AI Training Data Governance

Abstract

This paper focuses on the intricate challenges in AI training data governance and innovatively introduces the theory of copyright expectancy right as a potential solution. It first dissects the existing "trilemma dilemma" in AI copyright governance, encompassing the ambiguous rights of data sources, the paradox in determining copyright for AI-generated content, and the lag in regulatory frameworks. Subsequently, the study conducts in-depth jurisprudential validation of the expectancy right theory, exploring its legal philosophical foundations—drawing on Locke's labor property theory (framing data contributions as "digital labor") and Rawls' principle of justice (ensuring fair data distribution under the "curtain of ignorance")—and highlighting its institutional comparative advantages, such as breaking the "all-or-nothing" logic of traditional copyright, compensating for the passivity of the unjust enrichment system, and compatibility with the EU Text Mining Exception. A three-stage governance model (technology, institution, and ethics) is constructed to propose a gradient implementation path, including blockchain traceability, Shapley value-based dynamic distribution, obligation configurations for different scenarios, and ethical guidelines like Habermasian interaction rationality and the "glass box" principle of algorithm transparency. Additionally, a formula (ER=(Q×0.6+V×0.4)×C) is developed for quantifying the realization of copyright expectancy right. Finally, the paper returns to the humanistic value of intellectual property law, reinterpreting the incentive theory and constructing a ternary balance paradigm of technological innovation, institutional protection, and humanistic care. The research aims to provide a new paradigm for AI training data governance, balance the relationships between technological progress, institutional fairness, and humanistic care, and contribute to improving the global AI governance system.