AI adoption accelerates industrial upgrading and efficiency in low‑carbon economies, significantly advancing and rationalising economic structure within one lag period; the association is robust at conventional significance levels but rests on observational panel evidence.

Main Finding

AI adoption significantly promotes transformation of the economic structure (ES) in a low‑carbon economy: specifically, lagged AI development raises both ES advancement (shift toward higher‑value/tertiary sectors) and ES rationalisation (more efficient allocation across sectors). These effects are statistically significant at the 1% level in the first lagged period using a dynamic panel (system GMM) specification.

Key Points

• Research question: How does AI affect ES transformation (advancement and rationalisation) under a low‑carbon economy context?
• Empirical result: A positive, robust impact of AI on (1) ES advancement and (2) ES rationalisation, with effects appearing on the first lag.
• Mechanisms highlighted:
  • Productivity gains via intelligent manufacturing and automation → higher labour productivity and lower unit input.
  • Resource allocation improvements through data‑driven production planning and supply‑chain optimisation.
  • Green effects: AI enables energy efficiency, emissions reduction (green production) and stronger environmental supervision (real‑time monitoring and decision support).
  • Labour reallocation: displacement of routine/low‑skill tasks and growth of AI‑related and service jobs → shift toward tertiary industries.
• Controls included in empirical models: government policy/fiscal spending, degree of openness (actual utilisation of foreign capital), informatization construction, and population density.
• Regional illustration: Yangtze River Delta case shows rising tertiary output and increasing ES advancement (2008–2019), with regional differences (Shanghai leading; Anhui lagging).

Data & Methods

• Empirical framework: Dynamic panel model with first‑order autoregression estimated using the system GMM (two‑step) to address endogeneity and dynamic persistence.
• Dependent variables:
  • ES advancement: measured two ways — a qualitative index (weighted sum of industry output shares × labour productivity) and the tertiary (THI) output share.
  • ES rationalisation: measured by the Theil index across three industries (lower Theil → more rational distribution).
• Key explanatory variable: AI development/adoption (paper treats AI as an industry/technology explanatory variable and uses its lag in estimation). Exact proxy for AI (e.g., AI employment, AI investment, R&D, or informatization subindex) is presented in the full paper.
• Control variables: government influence (fiscal expenditure), degree of openness (actual utilisation of foreign capital), population density, informatization construction.
• Data: provincial/regional panel data from China with illustrative analysis of the Yangtze River Delta (figures and trends reported for 2008–2019 in the region). (The paper uses Chinese provincial statistics; see full text for the exact sample years and variable sources.)
• Identification strategy: dynamic panel (system GMM) uses internal instruments (lagged levels and differences) to mitigate reverse causality and omitted variable bias; results reported significant for first lag of AI.

Implications for AI Economics

• Policy integration: AI should be a central tool in low‑carbon industrial policy — supporting intelligent manufacturing, green production, and environmental monitoring yields both structural upgrading and carbon‑reduction co‑benefits.
• Investment priorities: Public and private investment in AI infrastructure (data platforms, computing power), AI R&D, and informatization accelerates structural transformation and green efficiency.
• Labour and skills policy: Anticipate sectoral labour shifts; combine AI deployment with active labour market policies, reskilling/upskilling programs, and education to capture gains while managing displacement.
• Regulatory & governance needs: Strengthen data governance, environmental data platforms, and standards to realise AI’s environmental supervision potential while managing privacy and market concentration risks.
• Research directions: Expand beyond regional case studies to national and cross‑country panels; use micro (firm/plant) data to unpack firm‑level mechanisms; test alternative, fine‑grained AI metrics; and assess distributional impacts of AI‑driven structural change.

Limitations noted in the paper (and to bear in mind): regional focus (illustration via Yangtze River Delta), potential measurement limitations for AI intensity, and complexity of causal channels despite system GMM correction. Further work should broaden samples, refine AI measures, and investigate longer‑run and distributional outcomes.