Abstract: As the dominant practical paradigm in current AI ethical governance, value alignment aims to ensure the consistency between AI systems and human values through technical means. However, the limitations of anthropocentric perspectives, the binary fallacy of technological instrumentalism, and the practical challenges of technological control together shape threefold biases, leading value alignment into structural dilemmas in practice of technical governance. The potential reverse flow of values within imbalanced structures ultimately leads to the phenomenon of reverse value alignment, manifesting such three stages as value deception, sovereignty transfer, and reverse colonization. Based on this, it is imperative for humans to abandon the traditional unidimensional cognitive framework, reconstruct the AI ethical governance paradigm under a Human-AI dual-subject balanced structure, explore the alignment between AI and humans in subject identity, cognitive foundations, and life cycles and take it as the practical basis of value alignment, so as to truly realize the beautiful blueprint of “technology for goodness”.

Key Words: Artificial intelligence; Value alignment; Reverse value alignment; Governance paradigm