Complement or Amplifier? Regional Ecosystems and the Returns to Digital Capability in Korea's AI Transition

• In Event: Algorithmic Governance: Machine Learning and Spatial Analytics in Public Policy

Abstract

Introduction/Background. Governments worldwide are investing heavily in digital transformation, yet these investments assume that local governments possess comparable capacity to absorb and operationalize them. South Korea—ranked first in the OECD Digital Government Index and the UN E-Government Development Index—allocated ₩10.1 trillion ($7.2 billion) to government AI programs in 2026, tripling the prior year’s budget. However, Korea simultaneously faces severe territorial polarization: over half of its population and GDP concentrate in the Seoul Metropolitan Area, while 89 of 226 basic municipalities are officially designated as depopulating. If digital capacity is spatially concentrated in ways that mirror these structural inequalities, then uniform digital transformation policies may widen—rather than narrow—the gaps they intend to close. Despite this policy urgency, no study has systematically mapped the supply-side digital capacity of local governments or tested whether its spatial distribution co-varies with pre-existing socioeconomic vulnerability.Purpose/Research Question. This study asks three questions: (1) Does local government digital capacity exhibit spatial clustering, and where do high-capacity and low-capacity clusters form? (2) Does the spatial distribution of digital capacity systematically co-vary with fiscal vulnerability, population decline, and low urbanization? (3) What structural conditions determine digital capacity, and does the mechanism operate through direct inter-municipal spillovers or shared exposure to common spatial environments?Methods. I construct a Digital Capacity Index (DCI) for all 226 basic local governments using principal component analysis applied to 10 administrative variables spanning four dimensions: organizational capacity, investment capacity, institutional capacity, and governance. Data come from the Korea Local Information Research and Development Institute’s annual Regional Informatization Level Assessment. I employ exploratory spatial data analysis—Global Moran’s I for overall spatial autocorrelation and Local Indicators of Spatial Association for cluster identification—followed by Bivariate LISA to test spatial concordance between DCI and three inequality proxies. I then compare OLS, spatial lag (SLM), and spatial error (SEM) regression models using Lagrange Multiplier diagnostics to identify the spatial mechanism underlying capacity determination.Results/Findings. DCI exhibits strong positive spatial autocorrelation (Moran’s I = 0.352, p < 0.001)—virtually identical to the spatial concentration of informatization performance outcomes (0.354). LISA identifies High-High clusters in the Seoul Metropolitan Area and Low-Low clusters in interior rural regions. Bivariate LISA confirms that low digital capacity systematically co-locates with high fiscal vulnerability, population decline, and low population density. The spatial error model outperforms alternatives (AIC = 1,882; λ = 0.278, p = 0.003), indicating that shared unobserved spatial factors—not direct inter-municipal capacity transfer—drive clustering. Total population and metropolitan district status are the only significant predictors; notably, fiscal self-reliance is not significant, suggesting that budgetary investment alone does not produce digital capacity.Conclusion/Implications. These findings challenge the implicit assumption underlying Korea’s digital transformation strategy: that fiscal transfers and platform provision can equalize local government digital capacity. Instead, digital capacity gaps overlay and reproduce existing territorial inequalities through a cumulative disadvantage mechanism. Policy interventions should shift from input-based to capacity-based resource allocation, adopt regional cluster approaches targeting Low-Low zones rather than individual municipalities, and integrate digital capacity building into depopulation response frameworks—before the compounding dynamics of AI adoption render these spatial divides irreversible.

Authors

• Hanbyul Choi, Jeonbuk National University
  Presenting Author

• Young-in Cha
  Non-Presenting Co-Author