This report examines how artificial intelligence (AI), particularly modern, data-driven, and increasingly generative systems, is likely to affect Korea’s productivity, growth, employment, and wages over the next decade or two. Three layers of evidence are integrated within a unified task-based framework: (i) micro-level measures of AI exposure and economic viability at the task and occupation levels, (ii) firm-level panel evidence on AI adoption and productivity, and (iii) a macroeconomic general-equilibrium model that aggregates task-level shocks into total factor productivity (TFP) and long-run labor market outcomes. The study defines AI as software systems that replicate human capabilities in learning, prediction, pattern recognition, perception, and natural language processing―a definition consistent with that used in the underlying firm survey and broad enough to encompass both earlier, narrow algorithms and more recent foundation models and AI agents.

Chapter 2 offers an accessible overview of the technological evolution of AI from traditional machine learning to transformer-based large language models and emerging agent-like systems. It emphasizes two features of particular macroeconomic relevance. First, advances in general-purpose and multimodal models have greatly expanded the range of tasks technically exposed to AI. Second, effective deployment requires complementary investments in data, organizational capacity, and human capital, suggesting that technical feasibility alone is a poor predictor of short-run economic impact.

Chapter 3 develops a task-based measure of AI automation potential tailored to the Korean labor market. Using detailed work descriptions from the Korean Employment Classification of Occupations (KECO) and Workpedia, the study decomposes 537 occupations into 6,824 unit tasks and scores each task along five dimensions: measurability of outcomes, data availability and quality, context dependence, required human judgment necessity, and agentability (the ease with which an AI agent can execute a given task). These components are combined into a task-level AI score that captures effective automation potential rather than mere technical exposure. A cost-benefit framework inspired by recent work on computer vision and AI costs is then applied to identify tasks and task·firm combinations for which automation is economically viable, given realistic assumptions about data, development, inference, integration, and maintenance costs relative to labor savings. The findings are conservative by design. Under a naive exposure-only criterion, AI could potentially affect roughly 8-12% of current employment and 10-11% of current sales, concentrated in finance, insurance, R&D, ICT, and certain transport and business services. Narrowing the focus to tasks that are both technically automatable and economically viable reduces these shares to about 1-3% of employment and 1-2% of sales. This suggests that, in the near term, AI will likely induce selective and partial automation in specific occupations and industries, rather than a broad-based, immediate displacement of labor.

Chapter 4 analyzes firm-level evidence using the nationally representative Korean Survey of Business Activities for 2017-2023, which asks firms whether they use AI―defined in the questionnaire as software that replicates human learning, reasoning, perception, and natural language understanding―and other fourth industrial revolution (IR4) technologies. To address the endogenous timing of AI adoption, the analysis employs panel IV (2SLS) models with firm and year fixed effects, instrumenting AI adoption with industry-level economic viability measures from Chapter 3. In the preferred specifications, firms that adopt AI and related IR4 technologies exhibit statistically significant increases in log sales per employee on the order of 5-20%, driven primarily by higher sales rather than systematic employment cuts. Pure “AI only” effects, net of broader digitalization, are estimated with less precision, consistent with AI operating as a general-purpose technology that must be embedded in a broader digital and organizational transformation. As a robustness check, propensity-score matching and difference-in-differences estimates for AI first movers suggest modest but positive short-run effects on labor productivity, with limited evidence of strong dynamic effects within the short observation window.

Chapter 5 incorporates these micro-level estimates into a task-based general-equilibrium model following the approach of Acemoglu and Restrepo to simulate AI-induced TFP paths over the next 10-15 years. Hulten’s theorem implies that, under certain assumptions, aggregate TFP change can be approximated by a Domar-weighted sum of task-level cost savings. The model therefore combines (i) the economically viable task shares by industry, (ii) firm-level productivity effects of AI, and (iii) alternative adoption scenarios―baseline, optimistic, and pessimistic―to trace out possible TFP trajectories. Under baseline assumptions, AI contributes a cumulative TFP gain of about 2.3% over a decade, nontrivial but below some of the more optimistic global projections. These simulations focus on the task automation channel and treat parameters as exogenous. They abstract from structural transformation, intangible capital accumulation, idea-generation effects, and potential discontinuities in AI capabilities. Then the analysis turns to long-run employment and wage effects across occupations. Using a cross-section of Korean 2-digit occupational categories over 2014-2024, the analysis estimates reduced-form relationships between changes in employment and real hourly wages and measures of routine and structured task content, while instrumenting for endogenous employment changes with leave-one-out labor-demand shocks. Occupations with higher routine task shares experience lower employment growth and somewhat lower wage growth over the decade, consistent with international evidence that automation pressure exerts downward pressure on labor demand and wage growth in exposed jobs. Combined with the AI exposure metrics presented earlier, these estimates imply that, under realistic adoption scenarios, the annual displacement and wage-drag effects attributable to AI are likely modest relative to existing demographic and structural trends, but non-negligible for specific groups.

Chapter 6 synthesizes these findings and discusses their implications for policy and business. For policymakers, the central message is that AI is unlikely to trigger an immediate, economy-wide employment collapse, but will require active management of selective automation, task restructuring, and skill upgrading in exposed sectors. The study argues for a market-friendly policy mix: clarifying AI-related regulation and liability, investing in data and compute infrastructure, strengthening education, training, and retraining systems, and fostering complementary innovations and organizational change. For firms, the key insight is that productivity gains derive less from AI in isolation and more from integrating AI into workflows, data pipelines, and human capital strategies.

Abstract
Preface
Summary (Korean)

Chapter 1. Introduction

Chapter 2. Technological Evolution and Economic Implications of Generative AI
　Section 1. Technological Evolution of Generative AI
　Section 2. Economic Implications of AI Adoption
　Section 3. Concluding Remarks: Technological Evolution and Economic Value of AI

Chapter 3. AI Automation: An Economic Feasibility Perspective
　Section 1. Estimating the Potential for AI Automation
　Section 2. Economic Feasibility Analysis of AI Automation
　Section 3. Macroeconomic Implications of AI Automation

Chapter 4. The Impact of AI on Firm Productivity
　Section 1. Current State of AI Adoption
　Section 2. Empirical Methodology
　Section 3. Empirical Results
　Section 4. Summary

Chapter 5. The Impact of AI on the Macroeconomy
　Section 1. Contribution of AI Adoption and Diffusion to TFP Growth
　Section 2. Long-Term Effects of AI Adoption on Employment and Wages
　Section 3. Summary

Chapter 6. Conclusions and Policy Recommendations
　Section 1. Summary and Limitations of the Study
　Section 2. Policy Measures and Corporate Strategies for AI Transformation

References
Appendix