AI investment in Chinese agricultural firms first lowers pollution but eventually raises it: early digital spending contracts production and cuts emissions, while heavier AI deployment boosts efficiency and scale—driving higher emissions and weakening green-innovation effects.

Artificial Intelligence and Agricultural Pollution in China: A Firm-Level Nonlinear Analysis

· January 27, 2026 · Global NEST Journal

openalex correlational low evidence 7/10 relevance Summary only summary available; pdf_status=paywall DOI Source PDF

Using 2010–2022 data on Chinese listed agricultural firms, the paper finds a nonlinear (U-shaped) relationship where low levels of AI investment reduce pollutant emissions but higher levels increase emissions as productivity and scale effects dominate, with operational efficiency and green innovation acting as mediating channels that themselves vary nonlinearly.

Citation observations

Cumulative provider counts captured on specific dates; providers are never combined.

0 cumulative citations

OpenAlex · Observed August 25, 2026

View corpus context

0 cumulative citations

Semantic Scholar · Observed August 25, 2026

View corpus context

<p>Artificial intelligence (AI) promotes high-quality development in agriculture while also introducing new challenges for the management of pollutant emissions. This study aims to explore the pathways and underlying mechanisms through which AI influences agricultural pollutant emissions. To achieve this, the study employs data from Chinese publicly listed agricultural firms from 2010 to 2022 and conducts an empirical analysis using a semi-parametric additive model. The results show that artificial intelligence has a nonlinear effect on agricultural pollutant emissions, initially inhibiting them and subsequently promoting them. In the early stages of digitalization, constrained by limited resources, AI investment reduces the scale of production, thereby lowering pollutant emissions. However, as AI investment intensifies, firms overcome resource constraints, and the resulting productivity gains and scale expansion effects lead to increased emissions. Mechanism analysis further reveals that AI influences agricultural pollutant emissions through two main channels: it first decreases and then enhances firms’ operational efficiency, and it initially boosts but later weakens their green innovation capacity. These findings provide theoretical support and practical guidance for promoting sustainable development and intelligent transformation in the agricultural sector.</p>

Summary

Main Finding

AI has a nonlinear effect on agricultural pollutant emissions among Chinese publicly listed agricultural firms (2010–2022): at low levels of AI investment/digitalization it reduces emissions, but beyond a threshold further AI investment increases emissions. The net pattern is an initial inhibitory effect on emissions followed by a promoting effect as AI intensity rises.

Key Points

Nonlinear relationship: AI → emissions is not monotonic; it reduces emissions in early stages of adoption and increases emissions once AI investment/intensity becomes large.
Early-stage mechanism: limited resources and the costs/adjustments associated with initial digitalization lead firms to shrink production scale and (on net) emit less pollution.
Later-stage mechanism: as AI investment deepens, resource constraints are overcome, productivity and scale expansion dominate, and emissions rise.
Two mediating channels identified:
- Operational efficiency channel: AI first decreases and then increases firms’ operational efficiency (short-term adjustment/frictions, long-term productivity gains).
- Green innovation channel: AI initially promotes green innovation capacity but, beyond a point, its positive effect on green innovation weakens.
Empirical evidence comes from firm-level panel data on Chinese publicly listed agricultural firms (2010–2022).

Data & Methods

Data: Panel of Chinese publicly listed agricultural firms, covering years 2010–2022.
Empirical approach: Semi-parametric additive model to flexibly estimate nonlinear relationships between AI intensity and agricultural pollutant emissions.
Mechanism analysis: Mediation/ pathway tests examining operational efficiency and green-innovation capacity as channels for AI’s impact on emissions.
Estimation strategy details (controls, robustness checks, identification assumptions) are not provided in the summary text; the study emphasizes nonlinear functional form and mechanism decomposition.

Implications for AI Economics

Nonlinear externalities: AI adoption generates non-monotonic environmental externalities. Models of AI-driven productivity should allow for threshold effects where scale gains eventually outweigh early frictions.
Policy design: Policymakers should not assume uniformly beneficial environmental effects from digitalization. Early-stage AI diffusion may reduce emissions, but sustained/unchecked scaling can increase pollution without complementary measures.
Complementary policy instruments: To capture productivity benefits while containing emissions, combine AI promotion with sustained support for green innovation, environmental regulation, and capacity-building (so green-innovation gains persist at higher AI intensity).
Firm strategy: Agricultural firms should plan digital investments with environmental management in mind—invest in green R&D and emissions-control technologies as AI scales to avoid rebound effects from larger output.
Research directions: AI economics needs more work on dynamic/adaptive regulation, heterogeneous firm responses, thresholds for technology-induced scale effects, and causal identification of AI’s environmental impacts across sectors and countries.

Assessment

Paper Typecorrelational Evidence Strengthlow — The paper documents a clear nonlinear (U-shaped) association and investigates mediating variables, but lacks a credible source of exogenous variation (no natural experiment, IV, or difference-in-differences exploiting plausibly exogenous shocks) so reverse causation, omitted variables, and selection into AI investment remain plausible explanations. Methods Rigormedium — Use of a semi-parametric additive model is appropriate for uncovering nonlinear relationships and mechanisms, and the analysis appears to use firm‑level panel data with covariate controls; however, the study does not appear to implement stronger causal-identification tools (instrumental variables, discontinuities, or randomized variation) nor robustness checks that would mitigate endogeneity concerns. SamplePanel of Chinese publicly listed agricultural firms from 2010 to 2022 (firm-year observations); key variables include firm-level AI investment/intensity (proxy), measured agricultural pollutant emissions (firm-level or estimated), operational efficiency metrics and green-innovation proxies, plus standard firm controls (size, profitability, leverage, age) and temporal controls; exact sample size and selection criteria not reported here. Themesproductivity innovation IdentificationEstimates nonlinear associations between firm-level AI investment and measured pollutant emissions using a semi-parametric additive model on panel data (Chinese publicly listed agricultural firms, 2010–2022); identification rests on controlling for observables (firm characteristics, time trends) and exploring mediation via operational efficiency and green innovation rather than on exogenous variation or quasi-experimental sources. GeneralizabilityLimited to publicly listed agricultural firms in China — excludes smallholders and non-listed firms, Country-specific institutional, regulatory, and market context (China) may limit applicability to other countries, Sector-specific (agriculture) — results may not generalize to manufacturing or services, Findings depend on how AI investment and emissions are measured; measurement error could affect external validity, Time period (2010–2022) spans rapid AI change; relationships may evolve with newer AI technologies

Claims (6)

Claim	Direction	Outcome	Confidence & Evidence	Details
This study employs data from Chinese publicly listed agricultural firms from 2010 to 2022 and conducts an empirical analysis using a semi-parametric additive model. Other	null_result	not applicable (methodological / data description)	Reading fidelity high Study strength high	not reported 0.5
Artificial intelligence has a nonlinear effect on agricultural pollutant emissions, initially inhibiting them and subsequently promoting them. Other	mixed	agricultural pollutant emissions	Reading fidelity high Study strength medium	not reported 0.3
In the early stages of digitalization, constrained by limited resources, AI investment reduces the scale of production, thereby lowering pollutant emissions. Other	negative	production scale leading to agricultural pollutant emissions (mediated outcome: scale of production / emissions)	Reading fidelity high Study strength medium	not reported 0.3
As AI investment intensifies, firms overcome resource constraints, and the resulting productivity gains and scale expansion effects lead to increased agricultural pollutant emissions. Other	positive	agricultural pollutant emissions (mediated by productivity gains and production scale)	Reading fidelity high Study strength medium	not reported 0.3
AI influences agricultural pollutant emissions through operational efficiency: it first decreases and then enhances firms’ operational efficiency. Organizational Efficiency	mixed	operational efficiency (as a channel affecting emissions)	Reading fidelity high Study strength medium	not reported 0.3
AI influences agricultural pollutant emissions through green innovation capacity: it initially boosts but later weakens firms’ green innovation capacity. Innovation Output	mixed	green innovation capacity	Reading fidelity high Study strength medium	not reported 0.3