AI adoption alone does not guarantee national productivity gains; small open economies need institutional absorptive capacity—skills, data governance, competition policy and social insurance—to convert AI exposure into shared prosperity.

Main Finding

The paper introduces the Dynamic Institutional Absorptive Capacity (DIAC) model to explain why the same AI adoption shock can produce very different productivity outcomes in small open economies. Core finding: AI adoption is a necessary but insufficient condition for aggregate productivity gains — real macroeconomic effects depend on complementary intangible investment, skill alignment, governance, and labour reallocation. When these institutional complements are weak, adoption can produce little or no inclusive productivity growth (or even concentrated gains and distributional harm); when strong, even moderate adoption yields substantial, broad-based productivity improvements.

Key Points

• DIAC concept: absorption = the joint effect of complementary intangible capital, skill alignment, governance capacity, and reallocation efficiency, weakened by distributional stress.
• Multiplicative interaction: AI adoption (A_t) multiplies with DIAC rather than adding to it; marginal productivity of AI rises nonlinearly with DIAC.
• Formal constructs (paper notation):
  • DIAC_t = (C_t^α S_t^β G_t^γ R_t^δ) / (1 + ρ Z_t)
  • Productivity: Δ ln(Y_t / L_t) = θ A_t DIAC_t - λ M_t + η X_t + ε_t
  • Distributional stress dynamics: ΔZ_t = μ A_t (1 - S_t) + ν K_t - ω R_t - ψ B_t
• Five testable propositions:
• Threshold effect: below a DIAC threshold, AI yields low marginal productivity; above it, gains rise nonlinearly.
• Conditional J-curve: an early measured productivity dip (J-curve) occurs only if complementary investments are real and cumulative.
• Labour-market segmentation matters: inclusion depends on institutions that reduce segmentation between high-skill, AI-complementary workers and those in substitutable tasks.
• Adoption-dependence risk in small open economies: imported AI raises adoption speed but may leak rents / learning unless local complementary capability is built.
• Distributional stress feeds back into absorption: inequality/insecurity lowers effective DIAC and thus productivity realisation.
• Three regimes predicted:
  • Adoption without absorption: widespread AI use but weak DIAC → little aggregate gain, high displacement anxiety.
  • Constrained complementarity: gains concentrated in frontier sectors; aggregate unevenness and wage polarization.
  • Adaptive complementarity: strong DIAC everywhere → sustained productivity growth and inclusive outcomes.
• Policy sequencing: maximizing AI adoption alone is suboptimal; priority should be building complementary institutions (skills, data governance, competition policy, portable social protection) to raise DIAC.

Data & Methods

• Methodological approach: analytical theory-building (conceptual model), not primary econometric estimation.
• Suggested empirical dependent variables: sectoral labor productivity growth, total factor productivity growth.
• Candidate proxies for main constructs:
  • AI adoption (A_t): AI-related software investment, AI patents, AI job postings, cloud usage, enterprise AI surveys.
  • Complementary intangible capital (C_t): measures of intangible investment, management-practice scores, digital-process maturity, data architecture indices.
  • Skill alignment (S_t): adult skills surveys, share of AI-complementary occupations, training participation, job-posting skill requirements.
  • Governance capacity (G_t): digital-government indices, competition enforcement metrics, data-protection/regulatory responsiveness scores.
  • Reallocation efficiency (R_t): job-to-job transition rates, active labour-market policy intensity, unemployment duration, credential portability indices.
  • Distributional stress (Z_t): wage polarization, platform/insecure work shares, political resistance indicators.
• Example empirical specification the paper proposes: Δ ln(Productivity_it) = β1 AI_it + β2 DIAC_it + β3 (AI_it × DIAC_it) + β4 Z_it + μ_i + τ_t + e_it
  • DIAC hypothesis predicts β3 > 0 (AI more productive where absorptive capacity is stronger).
• Recommended empirical strategies to address endogeneity and dynamics: dynamic panel methods, difference‑in‑differences around policy shocks, instrumental variables exploiting exogenous variation in AI exposure, and matched employer–employee microdata for wage/mobility effects.

Implications for AI Economics

• Rethink metrics: High rates of AI adoption (subscriptions, tool usage) do not guarantee national productivity gains; measurement must incorporate complementary intangible investment and institutional capacity.
• Policy: Small open economies should prioritise building DIAC — e.g., data governance that preserves learning, management and process redesign subsidies, targeted lifelong learning and credential portability, competition policy to limit data/market concentration, and social insurance that reduces adjustment frictions.
• Timing and sequencing matter: Investments in skills, organisational change, and governance should precede or accompany adoption to avoid adoption-without-absorption traps and politically destabilising distributional stress.
• Distribution is macro-relevant: Inequality and insecurity are not only welfare concerns but reduce the economy’s ability to absorb AI, so social policies can be productivity-enhancing.
• Research agenda: Empirical tests should focus on the AI × DIAC interaction; identify DIAC thresholds; distinguish temporary J-curve dynamics from structural absorption failure; and assess cross-country heterogeneity—particularly in small open economies dependent on imported AI stacks.
• Caution about openness: Access to global AI capabilities is valuable, but policies should balance openness with deliberate investments ensuring domestic capture of learning and non-rival complementarities to avoid rent leakage.

(Concise summary of theoretical model and suggested empirical pathways; the paper is a conceptual contribution that maps mechanisms and testable propositions rather than providing new econometric estimates.)