알고리듬이 공공성의 대상이 되기 위한 조건과 원리 : 교육분야 알고리듬 공공성을 중심으로

초록

This study departs from the observation that although algorithms now function as computational rule systems governing classification, recommendation, assessment, and prediction across public domains, research that treats algorithms themselves as objects of publicness remains scarce. It explores the conditions and principles under which algorithms can become legitimate objects of publicness, situating the inquiry concretely in the field of education. Taking algorithmic governmentality as its theoretical foundation, the study draws on Foucault's analytics of power—particularly the distinction between sovereign power, disciplinary power, and the governmental rationality elaborated in his Collège de France lectures—and on Rouvroy and Berns's theory of algorithmic governmentality, which theorizes how algorithmic systems preempt subjectivity through the triple process of data harvesting, automated profiling, and anticipatory action. Rather than remaining at the level of diagnosing how algorithms structure the field of possible actions, the study adopts a meta-perspective: it examines the normative conditions required for subjecting algorithmic governance itself to democratic control. Through an issue-oriented literature review, the study analyzes publicness issues arising in adaptive learning platforms, learning analytics systems, dropout prediction systems, and South Korea's AI Digital Textbook policy. It diagnoses the limitations of existing educational publicness discourse and derives four directions for conceptual complementation: extending the object of publicness to sociotechnical mediation, adding operational and procedural dimensions, combining the public (公) and the common (共), and incorporating the material-infrastructural dimension. On this basis, it defines algorithmic publicness in education and proposes six core principles: rule publicness (public explicability of algorithmic purposes and criteria), process publicness (traceability and auditability from data collection to model updates), relational publicness (institutional guarantee of teachers' interpretive authority and learners' right to contestation), common-world publicness (norms protecting shared experience from individualization logic), governance publicness (democratic procurement, oversight, and participation structures), and material publicness (public controllability of data, models, cloud, and infrastructure). Among these, rule, process, governance, and material publicness are presented as universal conditions required wherever algorithms mediate public judgment, while relational and common-world publicness are domain-conditional principles adaptable according to each field's normative character. The study proposes algorithmic publicness (algorithmic public&common) as an expanded framework that encompasses algorithmic fairness as an internal component while responding to the distinctive normative demands of education, complementing rather than replacing existing discourses on educational publicness.

키워드