Abstract: Generative Artificial Intelligence (GAI) has become deeply embedded in the socio-economic fabric of human society, and its growing agency has emerged as a focal point of both theoretical debate and practical concern. On the one hand, GAI increasingly exhibits capabilities traditionally regarded as uniquely human, including interactive intelligence, cognitive flexibility, and creative expression. These characteristics not only enhance GAI’s functionality but also open new avenues for augmenting human cognition. On the other hand, the generative processes of GAI are frequently associated with factual inaccuracies, reasoning biases, and ethical dilemmas, leading to a fragile trust relationship between humans and machines. As a result, human reliance on GAI coexists with mounting skepticism, creating a fundamental tension between its empowering potential and the risks of technological unpredictability and moral ambiguity. This paper investigates the question of what paradoxical tensions accompany the rise of GAI’s agency and how these tensions may be addressed. We first identify that GAI’s agency is characterized by three core features—interactivity, autonomy, and creativity—each of which is enabled by distinct forms of technological affordance. Interactivity is shaped by perceptual and communicative affordances, autonomy by operational and situational affordances, and creativity by the interplay of interactive and situational affordances. While these features empower GAI with remarkable generative capabilities, they simultaneously give rise to complex system-level risks and trust challenges. Specifically, the high degree of interactivity leads to a lack of transparency in generative logic, producing a “black-box” effect that impedes user understanding. The rise in autonomy undermines users’ sense of control, as GAI’s decision-making processes and behavioral evolution become increasingly opaque and less predictable. Meanwhile, its expanding creative potential complicates the attribution of accountability, creating asymmetries in responsibility where the more novel the output, the more ambiguous its moral and legal ownership. These contradictions manifest in three interconnected paradoxes: the interaction–transparency paradox, wherein GAI’s interactive flexibility conflicts with the opacity of its generative mechanisms; the autonomy–controllability paradox, where heightened decision-making independence erodes human oversight; and the creativity–accountability paradox, in which increasing innovation introduces diffuse and deferred responsibility. Addressing these challenges requires a shift in governance strategies: enhancing interpretability through explainable AI and knowledge mediation, reinforcing human oversight via participatory design and feedback mechanisms, and clarifying ethical boundaries through shared responsibility and robust accountability frameworks. Only by navigating these tensions thoughtfully can we ensure that the technological advancements of GAI genuinely serve collective human interests and foster a synergistic development of both technological and societal values.

摘要： 人工智能由“分析式”向“生成式”跃升，其主体性愈发凸显。与之伴生的透明度遮蔽、可控感弱化与责任归属模糊，深刻冲击了人机协作的关系范式与信任结构。本文发现，生成式人工智能的主体性主要体现为交互性、自主性和创造性，其演化过程催生了“交互-透明”、“自主-可控”和“创造-责任归属”三组张力，引致黑箱焦虑、失控信任和责任悬置三大现实困境。为化解这些矛盾，本文基于社会技术系统理论，提出技术路径与社会路径的协同治理策略，包括倚赖可解释性AI与知识中介机制重建透明度，借助参与式设计与反馈机制增强可控感，以及藉由责任共享与问责机制明晰责任边界，从而实现技术创新和信任建构的动态平衡。本文深化了对技术与社会系统共构关系的哲学反思，为生成式人工智能的设计优化与治理制度化提供理论支撑与实践指引。