Targeted AI adoption could materially boost Kazakhstan's value added and employment: simulations attribute 16.8 percentage points of a projected 35.3 p.p. GVA increase (2025–2035) to AI, lift the GDP share of priority sectors by 6.3 p.p., and create roughly 1.3 million jobs, as task-creation effects outweigh automation.

Main Finding

The authors develop a calibrated, integrated dynamic model for a resource-dependent developing economy and find that targeted AI diffusion (2025–2035) can materially accelerate structural transformation in Kazakhstan: cumulative gross value added (GVA) in the analysed industries rises by 35.3 percentage points (p.p.), of which 16.8 p.p. is attributable to AI. Concentrating public support on a small set of priority sectors yields faster AI adoption (86.8–93.8% adoption in priority sectors by 2035), increases the priority-sector share of GDP by 6.3 p.p., and expands total employment by 22.4% (≈ +1.3 million jobs). Across sectors, task-creation effects dominate automation effects, reflecting a shortage of qualified personnel in a resource-dependent economy.

Reference: Varavin, Y. V., Kozlova, M. V., Rakhimberdinova, M. U., & Ozpence, O. (2026). The Impact of Artificial Intelligence on the Structural Transformation of Kazakhstan's Economy. Economy: strategy and practice, 21(1), 110–127. https://doi.org/10.51176/1997-9967-2026-1-110-127. JEL: J24, O11, O33.

Key Points

• Purpose and novelty: Builds the first calibrated integrated model for a resource‑dependent developing economy that merges innovation diffusion, endogenous technological progress and a task‑oriented labor framework, with sectoral prioritization.
• Model architecture: three linked modules
  • Diffusion: extended Bass model with sectoral p, q parameters (captures S-shaped adoption + stronger network/scale effects for AI).
  • Productivity/GVA: expanded production function with endogenous tech progress that translates adoption into GVA gains.
  • Employment: Acemoglu–Restrepo task framework separating automation vs. new task creation effects.
• Sector selection: Ten industries (from 19 OKED branches) were modelled; mining was excluded as not relevant for diversification. The ten sectors include: agriculture/forestry/fishing; manufacturing; construction; wholesale & retail; transport & storage; information & communication; finance & insurance; real estate activities; professional/scientific/technical activities; administrative services.
• Prioritization: a multi‑criteria scoring system with five criteria and weights — K1: diversification potential (30%), K2: AI application potential (25%), K3: job creation potential (20%), K4: digital/infrastructure readiness (15%), K5: export/global value chain potential (10%). Threshold of 8/10 yields four priority sectors.
• Key calibrated parameter ranges reported:
  • Bass diffusion: p ∈ [0.013, 0.025]; q ∈ [0.70, 0.88]
  • Productivity channel: multiplicative effect includes α ∈ [0.60, 0.90] and β ∈ [1.0%, 2.0%] (per-period tech effect scalar)
  • Employment (task model): δc = 0.35 (task-creation intensity), δa = 0.15 (automation intensity) for priority sectors; employment evolves with Li,t = Li,t-1 · [1 + γ·A·(δc − δa)].
• Main quantitative outcomes (2025–2035, baseline):
  • Aggregate cumulative GVA growth in modelled industries: +35.3 p.p.
  • Contribution of AI to that growth: +16.8 p.p.
  • AI adoption in priority sectors by 2035: 86.8–93.8 percentage points (13–32 p.p. higher than non‑priority sectors)
  • Priority-sector GDP share increases by +6.3 p.p.; mining share declines by −1.3 p.p.
  • Employment: aggregate employment increases by +22.4% (~+1.3 million jobs). Net effect dominated by new task creation over automation across all sectors.
• Sensitivity/scenario results: baseline GVA +35.3%; optimistic/ conservative scenarios ±3.8 p.p.
• Policy recommendation: concentrate government support on a limited number of high‑priority sectors to maximize diversification and structural change.

Data & Methods

• Data sources: official sectoral statistics from the Bureau of National Statistics (Agency for Strategic Planning and Reforms of the Republic of Kazakhstan) for 2020–2024 (GVA, employment, physical volume index), supplemented by international empirical estimates and national strategic documents (Kazakhstan AI Concept 2024–2029, labor market strategy).
• Sample: 10 selected industries out of 19 OKED branches (mining excluded). Baseline 2024 sector-level GVA shares and employment used for calibration (table of sector GVA, shares, employment and average physical volume growth provided for 2020–2024).
• Model details:
  • Diffusion module: Ai,t = Ai,t-1 + [p + q·Ai,t-1]·[1 − Ai,t-1] (Bass equation) with sector‑specific p and q calibrated within reported ranges to reflect faster AI diffusion (strong q due to network/data effects).
  • Productivity module: VASi,t = VASi,t-1 · (1 + g) · (1 + α·A·β), where A is adoption level; α and β capture adoption‑to‑productivity elasticity and per‑unit tech effect.
  • Employment module (task approach): Li,t = Li,t-1 · [1 + γ·A·(δc − δa)], separating task‑creation (δc) and automation (δa) intensities.
  • Prioritization: a two‑stage industry selection — first filter by share, public dominance, and diversification relevance; second stage multi‑criteria scoring (Si = 0.30·K1 + 0.25·K2 + 0.20·K3 + 0.15·K4 + 0.10·K5). Threshold 8/10 for priority designation.
• Time horizon and resolution: 2025–2035, annual steps.
• Calibration strategy: combined use of Kazakhstan sectoral time‑series (2020–2024) and international micro/firms/experiment evidence (Brynjolfsson, Dell’Acqua, IMF, McKinsey, Eloundou et al., etc.) to set plausible parameter ranges; scenario analysis around central calibration.

Implications for AI Economics

• Methodological contribution: provides an operational integrated framework that merges diffusion (Bass), endogenous productivity, and task‑based labor reallocation for a resource‑dependent developing economy — filling a gap where most AI impact studies focus on advanced economies.
• Policy inference for developing/resource economies:
  • AI can be a credible lever for diversification when adoption is concentrated in a few sectors with high synergy and export potential.
  • Targeted public support (to accelerate diffusion and build digital/institutional readiness) is likely more effective than uniform policy across all sectors.
  • Skills, human capital and institutional bottlenecks matter: in low‑skill/resource contexts, AI tends to generate new tasks more than substitute labor, implying strong complementarities that require upskilling and training policies.
• Labor‑market insight: the Acemoglu–Restrepo task decomposition can yield net employment gains in contexts with labor shortages and low initial automation, reversing fears of mass disemployment in some developing settings.
• For future research:
  • The model suggests firm‑level and task‑level microdata from developing countries would improve parameter identification (e.g., sectoral δc and δa, adoption elasticity α).
  • Extensions could integrate trade/complexity measures explicitly (to trace how AI affects export complexity) and spatial/institutional heterogeneity across regions.
  • Cross‑country comparative calibration would test external validity for other resource‑dependent economies.
• Limitations to note for economic interpretation: results depend on calibration choices and assumed policy intensity; mining excluded (so GDP‑level impacts interact with commodity dynamics outside the model); the model abstracts from firm‑level market structures and potential distributional/institutional frictions that could dampen expected gains.

If you want, I can: - extract the four priority sectors identified in the paper and summarize their sector‑specific projections; - produce a concise policy brief (1 page) based on the paper’s recommendations; - outline model equations and parameter values in a compact table for replication.