The Praxis Problem: Alignment, Discord, and the Limits of AI Governance in South Korea and Their Implications for Middle-Power States

Abstract

Praxis refers to the process by which a theory, lesson, or concept is enacted, embodied, and realized in practice. As Artificial Intelligence reshapes the foundations of economic competitiveness and geopolitical power, the question of how nations govern AI has become one of the defining policy challenges of the era. Yet, existing evaluative frameworks share a common limitation: they assess governance inputs and procedures, and where they do examine state–firm dynamics, they treat them as unitary actors, obscuring the intra-sector tensions that often determine whether a nominally coherent national strategy achieves traction in practice. This thesis addresses that gap by proposing and applying a new analytical framework: alignment–discord, evaluating AI governance not by counting laws passed or funds allocated, but by diagnosing the relational coherence between and within the public and private sectors, and identifying strategic decoupling, the condition in which firms formally adopt the vocabulary of national AI rhetoric while capital allocation decisions and investment trajectories orient elsewhere. South Korea serves as the primary case study through which this framework is developed and applied, as it is a particularly instructive case given its status as a middle-power with a historically state-directed industrial economy, a dominant yet structurally asymmetric position in the global AI value chain anchored in memory chips, and an ambitious but nascent AI governance regime. The thesis draws on original expert interviews conducted in Seoul, alongside analysis of Korean-language legislative documents, government press releases, and trade data. Unless otherwise noted, all translations from foreign-language sources and author-conducted interviews are the author’s own, and currencies are inflation-adjusted for current denominations. The analysis identifies four domains of meaningful policy–practice alignment: the government’s recognition of its enabling rather than directive role, its provision of compute infrastructure through GPU distribution and the Independent AI Foundation Model Project, its explicit legislative support for SMEs, and its framing of international engagement as partnership-oriented rather than isolationist. It simultaneously identifies four domains of discord: jurisdictional fragmentation across overlapping regulatory agencies, regulatory opacity stemming from Korea’s positive regulation legal tradition which engenders firm hesitation rather than innovation, intra-sectoral divergence over which path of AI development represents the most viable strategic choice, and ecosystem-level structural deficiencies in data sharing reinforced by chaebol dominance. Discord persists, the thesis argues, due to four structural factors: resource constraints inherent to middle-power status, temporal misalignment between political cycles and long-horizon industrial investments, corporate heterogeneity across the AI value chain, and geopolitical exposure to the U.S.–China technology rivalry. The thesis concludes with three calibrated policy recommendations: establishing a Lead AI Authority with binding inter-ministerial coordination power alongside a legislative harmonization audit of AI-adjacent laws; creating an AI Strategic Council and an AI–Compute Integration Fund to institutionalize public–private alignment; and founding an AI Middle-Powers Forum, a coalition of states holding strategic nodes in the AI value chain, to convert networked leverage into governance influence. Together, these recommendations illustrate the alignment–discord framework’s prescriptive and analytical value, demonstrating that effective AI governance reform must target specific coordination mechanisms and incentive structures rather than simply expanding strategic ambition.