Abstract: Algorithmic decision-making is deeply embedded in socioeconomic activities, creating substantial value. Major global tech giants like Google, Amazon, Meituan, and ByteDance have established extensive platform ecosystems within their respective industries, leveraging their technological advantages. The widespread influence of these platforms on socioeconomic activities far surpasses typical corporate behavior. However, the improper use of algorithms has led to phenomena such as algorithmic discrimination, collusion, and monopolization, significantly distorting the initial purpose of algorithms in serving socioeconomic development. The root cause of improper algorithm use lies in the asynchronous development of AI-related technologies and the gradual nature of human cognition and its adherence to social norms—a phenomenon referred to as the “social disembedding” of algorithmic decisions. Traditional regulatory measures and legal norms tend to lag, unable to ly address the technological variations brought about by the self-evolving nature of algorithms. This lag in regulation and the rapid pace of technological change create a disconnect that can lead to significant social and economic issues. This paper identifies the core issue of social disembedding in algorithmic decision-making as the asynchronous development between iterative advancements in artificial intelligence technology and the gradual nature of human cognition. To address this, the paper explains the social disembedding problem of algorithmic decisions through the paradoxical relationship between formal rationality and substantive rationality. Specifically, the paper re-examines algorithmic decisions based on human cognitive abilities, focusing on the intrinsic connections between algorithms and human cognition. This approach aims to enable algorithmic decisions to more accurately perceive human needs, thereby facilitating the proper resolution of real-world issues. On one hand, human cognitive abilities can compensate for the purely data-driven nature of algorithmic decisions and improve the interpretability of algorithmic outcomes. On the other hand, algorithms can overcome the limitations of theoretical logic induction and empirical deduction in human decision-making processes, alleviating cognitive pressure. Building on this foundation, the paper develops a cognitive enhancement algorithm framework. This framework introduces the concepts of algorithmic space and cognitive space. The cognitive enhancement algorithms within the algorithmic space manifest in two primary functions: first, by utilizing advanced algorithmic logic to map and analyze complex real-world problems, translating these issues into meaningful patterns and relationships through digital representation. Second, through algorithmic optimization and artificial intelligence technologies, these patterns and relationships are transformed into forms of knowledge that are comprehensible and usable by humans. The cognitive space, on the other hand, evolves around different stages of human cognitive development, constructing an ethical and value system surrounding algorithm application. Within the cognitive enhancement algorithm decision-making framework, algorithmic space and cognitive space develop both independently and interdependently. The cognitive enhancement algorithm framework proposed in this paper organically integrates technological capabilities with human cognitive abilities to reconcile the paradox between formal rationality and substantive rationality. This integration is crucial for creating a harmonious relationship between advanced algorithms and the society they serve. By addressing the disconnect between the rapid iteration of AI technologies and the slower pace of human cognitive and normative development, this framework aims to foster a more balanced and ethical application of algorithms in socioeconomic contexts. Ultimately, this integration facilitates the social re-embedding of algorithmic decision-making, ensuring that these technologies contribute positively to societal development while mitigating the risks associated with their misuse.

摘要： 算法决策深度嵌入于社会经济活动而创造价值，但因其自我学习与自适应能力的强化以及人类认知能力的相对弱化，愈发展现出脱离于社会规范体系的趋势，致使算法失当现象频发。传统技术规制和法律规范的解决措施具有一定程度的滞后性，难以实时应对因算法自生长性所带来的技术变异。本文发现算法决策的社会脱嵌根源在于人工智能技术迭代式发展与人类认知渐进性的非同步性。对此，本文基于形式理性和实质理性之间的矛盾关系来解释算法决策的社会脱嵌问题，并发展了一个认知增强算法框架。该框架提出了算法空间与认知空间的概念，将技术能力和人类认知能力有机统合以纾解形式理性和实质理性的矛盾，最终实现了算法决策的社会再嵌入。