Abstract: Contemporary society is becoming increasingly digitized, with data permeating nearly every aspect of life. The widespread use of data technologies in the knowledge economy and social decision-making has elevated data to a central role in both knowledge production and intelligent decision-making. This transformation reshapes the relationships between humans and machines, humans and knowledge, and humans and intelligence. At the same time, it signals that data is assuming a decisive role within the knowledge system, gradually establishing an authoritative presence in the realm of cognition—one that is often accepted without scrutiny. However, the authority of data fundamentally differs from that of traditional experts. Expert knowledge is grounded in rigorous systems of production and verification, characterized by causal reasoning and repeatability. In contrast, data is often criticized for its reliance on inductive methods and inconsistent quality, which raises concerns about its reliability as a source of knowledge. Furthermore, expert knowledge can bridge the gap between experts and the public through active participation, whereas data-driven knowledge, constrained by technical barriers, often excludes public engagement. As a result, the widespread trust in data may be misplaced, leading to both cognitive and ethical errors. This paper introduces two forms of structural epistemic injustice—“hermeneutical injustice” and “contributory injustice”—to demonstrate how structural biases in data obstruct knowers' understanding of societal and personal experiences. Hermeneutical injustice arises when certain groups lack the necessary conceptual resources to understand and articulate their experiences in social practices, suffering from bias and marginalization in interpretation. Contributive injustice arises when marginalized groups, though capable of developing concepts that reflect their unique social realities, are denied recognition by dominant groups. These suppressions of cognitive diversity result in their exclusion, misunderstanding, and misrepresentation, preventing their full participation in both social and cognitive processes. The digital intelligent society provides new ground for structural epistemic injustice. Data and intelligent technologies significantly expand the cognitive landscape, allowing previously invisible social elements and marginalized voices to emerge through extensive data collection. However, as data is often shaped by specific intentions and biases, it can amplify injustice and discrimination, entrenching and perpetuating inequities. Structural epistemic injustice in the digital intelligent society manifests in three key forms: (1) The invisibility of certain groups in data, resulting from insufficient representation in data collection, excludes marginalized groups from decision-making processes due to a lack of data identity or representation. (2) Bias and discrimination stemming from low-quality data, where issues with data quality and algorithmic processes lead to unjust outcomes for marginalized groups. The opacity and complexity of data-driven decisions hinder understanding and appeal, leaving affected individuals without the means to challenge results. (3) The undue authority granted to data, leading to procedural “justice,” where unjust outcomes are defended as unquestionable due to the perceived infallibility of data, with cognitive inertia causing the dominant group to overlook clear injustices. In data research and practice, data diversity and cognitive pluralism offer potential solutions to structural epistemic injustice. Data diversity improves data quality by incorporating a broader range of perspectives, while cognitive pluralism advocates for the inclusion of diverse non-cognitive values in data-driven decision-making. Together, these approaches can reduce discrimination, promote data justice, and foster social fairness.

摘要： 当代人类社会生活正在被渐进式地数据化，数据逐渐成为信息和知识的重要来源并正在认知领域形成标识有客观属性的权威。但有赖于数据质量和归纳方法的数据科学可能在实践中被赋予了过高可信度，若不加审查地应用将产生认知与伦理的双重错误。本文试图引入解释非正义和贡献非正义两种结构性认知非正义形式，揭示数据在实践中阻碍认知者获取相关认知资源理解社会与自身经历，并利用结构性偏见边缘化弱势群体，使其无法在数智社会中有效参与社会实践和认知实践。数智社会中的结构性认知非正义表现为三种典型现象：部分群体在数据中不具代表性、低质量数据产生偏见与歧视、数据结果被赋予过高可信度而具有不合理的程序正义。从数据研究和实践过程来看，数据多样性和认知多元化有助于结构性认知非正义的缓解，促进数据正义与社会公平。