生成式AI价值对齐的法理反思与路径选择

Jurisprudence Reflection and Path Selection on Value Alignment of Generative AI

智能时代,以ChatGPT为代表的生成式AI对人类社会生活产生了深刻影响并带来了道德风险。基于生成式AI的运行逻辑与法律的保守性,化解这些道德风险要求AI的目标和应用与人类价值观保持一致。价值对齐包括共识性道德与路径选择两个不同层面的问题。共识性道德问题回答的是何种道德理论能被恰当地视为对齐的框架,路径选择问题则要说明最有可能实现价值对齐的思路与方法。当前的对齐路径轻视了道德生活的复杂性,并错误地理解了生成式AI在社会结构中所处的位置。生成式AI的特殊性既不足以确立其为主体性的存在,也阻碍了它被视为纯粹客体的可能性。为此,价值对齐路径需要转向人机合作的视角,将生成式AI视为道德接受者,为其建立共识性道德框架并由人类承担一定程度的“注意力”引导。

In the age of intelligence,generative AI represented by ChatGPT has a profound impact on human social life and brings moral hazard.Based on the operational logic of generative AI and the conservative nature of the law,resolving these moral hazards requires that the goals and applications of AI be aligned with human values.Value alignment includes two different problems:consensus morality and path selection.The consensual moral question answers the question of which moral theories can be properly regarded as a framework for alignment.The path selection problem should explain the most likely ideas and methods to achieve value alignment.The current alignment path trivializes the complexity of moral life and misunderstands generative AI's place in the social fabric.The particularity of generative AI is not enough to establish its existence as a subjectivity,and it also hinders the possibility of its being regarded as a pure object.To this end,the value alignment path needs to shift to a human-machine cooperation perspective,which sees generative AI as a moral recipient,establishes a consensual ethical framework for it,and is guided by a degree of"attention"borne by humans.