Dynamic Linkages among Carbon Emissions, Artificial Intelligence, Economic Policy Uncertainty, and Renewable Energy Consumption: Evidence from East Asia and Pacific Countries

Abstract

A growing number of countries are concerned about the reliability of environmental indicators; as a result, there is a pressing need to find ways to improve ecological welfare on a global scale. This study investigates the dynamic linkages among CO₂ emissions, AI, economic policy uncertainty (EPU), and renewable energy consumption. To analyze these relationships empirically, this study used panel data for East Asian and Pacific countries from 2000 to 2023. This study used fully modified ordinary least squares (FMOLSs), dynamic ordinary least squares (DOLSs), Hausman fixed effects (FEs) and random effects (REs), the generalized method of moments (GMM), and variance decomposition tests. This study’s results show that AI has a positive relationship with CO₂ emissions in terms of the benchmark regression, while it shows minimal impact on CO₂ emissions according to the variance decomposition test. Similarly, economic policy uncertainty shows a strong positive relationship with CO₂ emissions through benchmark regression FEs and REs, GMM, and the variance decomposition test. An increase in EPU will positively affect CO₂ emissions. Renewable energy consumption has a strong negative impact on CO₂ emissions in East Asian and Pacific countries. These findings reveal that a unit increase in renewable energy consumption will decrease CO₂ emissions. Based on the results of this study, it is suggested that policy certainty and an upsurge in renewable energy consumption are essential for environmental upgrading. In contrast, adopting AI has no robust effect on ecological degradation (CO₂ emissions). East Asian and Pacific countries need to focus on the adoption of renewables, as well as the control of economic policy uncertainty. While AI in East Asian and Pacific countries is still in the initial stage of adoption, policy formation is essential to overcome the possible carbon footprint of AI in the short term.

1. Introduction

Climate change-related devastation is still progressing, even though the economy has slowed down because of the COVID-19 pandemic and its aftermath. Emissions temporarily decreased as a result of a reduction in human activity during the pandemic. Around 95% of pollution emissions come from greenhouse gases that are generated by humans, which condense in the atmosphere [1] As sustainable development aims to meet “the needs of the present without comprising the ability of future generations to meet their needs”, it is a critical global concern. According to the 17 sustainable development goals (SDGs) set out by the United Nations (UN) as part of the 2030 agenda, a better world must be created. The primary focus of all 17 objectives is prosperity and well-being, with 169 targets and subgoals set out to achieve these objectives. The UN SDGs demand significant action in all spheres of life, including all possible applications of technological innovation [2]. The other objectives cannot be attained without industry, innovation, and infrastructure, which are emphasized in SDG 9. Similarly, the Paris Agreement clarifies how crucial cutting-edge climate technology is to a sustainable future. A solution to climate change that might aid in promoting economic growth and easing environmental burdens is to accelerate and encourage innovation; however, it is difficult to achieve sustainable development in the early stages of growth and development. When meeting basic human needs is prioritized over the environment, there appears to be a clear tradeoff between economic development and environmental security.

Ref. [3] explained the fourth industrial revolution, in which technological dependence is crucial. Still, it also requires a dual shift to digital and green practices. This dual shift will impact every facet of people’s lives. For instance, to promote growth and development, industries with low energy efficiency must increasingly rely on green energy and energy-efficient technologies, as detailed by [4]. Given the essential roles that green growth, connectivity, infrastructure, digitalization, and the Internet of Things play in the twin transition, this shift is key to decarbonizing the economy. In recent years, researchers have advanced the research in this field, adding a new Industry 5.0 phase centered on sustainability, the green economy, and the human–technology partnership [5]. One study explored the concept of Industry 5.0, which connects environmentally friendly practices and sustainability. According to the authors, collaboration between many economic sectors should be improved for the greater good [6].

Ref. [7] stated that green trade and investment are essential to supporting successful energy transitions and the implementation of nationally determined contributions (NDCs) in developing countries. The increasing need for energy has created a problematic tradeoff between environmental security and economic development. The pursuit of carbon neutrality exacerbates environmental corrosion and climate change. Greenhouse gas emissions (GHGs) and energy preservation are only two of the difficulties posed by this exceptional situation. Therefore, most recent energy- and environment-related studies aim to examine the connection between CO₂ emissions, environmental quality and advanced technology.

Since the Industrial Revolution, human activity has increased the quantity of greenhouse gases in the atmosphere and has caused significant global warming. Computer technology has been steadily improving since the 1990s, and numerous new economic models, including the digital economy, have been made possible by advances in artificial intelligence, blockchain technology, and 5G technology. In a digital economy, massive amounts of data are created, selected, filtered, stored, and used in a way that quickly and optimally allocates and regenerates resources, leading to high-quality economic development [6].

The current study adds fresh information that bridges separate streams of thought in the existing literature. In particular, CO₂ emissions, AI, EPU, and RENE are all investigated. As noted by [8], enhancing domestic energy-saving and emissions-reducing technologies depends on highly trained human resources. Developed countries with high levels of human capital are more likely to create cutting-edge technology, as detailed by [8].

A weak economy caused by EPU encourages companies to use more conventional, polluting, and less-expensive energy sources for production, such as coal and oil, which increases CO₂ emissions. Ref. [9] used U.S. sector data to conduct a new parametric test involving Granger causality, which was used to investigate the effect of EPU on CO₂ emissions; they determined the Granger causality between the two variables. In their study, ref. Ref. [10] used a bootstrap panel Granger causality test to examine the causative connection between EPU and both energy consumption and CO₂ emissions in the G7 countries. They stated that EPU had negative impacts on reducing emissions and conserving energy. Furthermore, ref. [11] reported strong correlations among geopolitical risk, economic policy uncertainty, energy consumption, economic growth, and CO₂ emissions in the long term, based on data from nations that are wealthy in resources yet prone to crises. These results show that higher EPU harms carbon abatement. This observation aligns with the outcome reported by [12]. Meanwhile, ref. [13] concluded that EPU reduces China’s CO₂ emissions economically. Ref. [14] proposed that the degree of economic policy uncertainty in China’s provinces substantially affects the carbon emission intensity of manufacturing enterprises. The second research stream indicates that EPU has a mitigating effect on CO₂ emissions.

According to the economic growth model proposed in Solow’s foundational 1956 book, technological advancement comes from outside the economy. Ref. [15] created a growth model to supplement natural technical progress. Romer’s model states that creating new goods through research and development by profit-maximizing corporate firms drives technological evolution. Various ideas and metrics have been used to assess the effects of globalization and technological advancements. Ref. [16] stated that technology is the repeatable use of scientific knowledge to achieve concrete goals. Finding knowledge outside of a company and incorporating it into the open innovation framework is one tactic that can lead to increased success. It may be possible to reduce barriers to the circular economy through open innovation. At the same time, we need to improve our understanding of how these fields may collaborate or how open innovation can contribute to developing a more sustainable economy. As noted previously, studies benefit from adopting multiple methodologies and ways of studying these issues. Nevertheless, more research is needed that investigates the connections among CO₂ emissions, cutting-edge (AI) technological adaptation, economic policy uncertainty, and renewable energy consumption in East Asian and Pacific countries. The literature indicates that there are relatively few studies on the effects of AI on the intensity of pollution emissions, suggesting that studies need to discuss the specific mechanisms and heterogeneity in AI’s impact on pollution emission intensity. With its capacity for deep learning, AI can be rapidly and broadly applied across various economic and social fields [17]. This capability can alter traditional production models, unlock economic growth potential, promote industrial structure upgrades, produce systemic effects on the economic system, and create new opportunities to overcome the bottleneck in emissions reduction.

This study investigates the extent of AI’s impact on CO₂ emission intensity and its mechanism of action by conducting a theoretical analysis and empirical tests. The significance and novelty of this article are as follows. First, this study uses East Asia and the Pacific as a case study to examine the impact of AI on carbon emissions intensity, based on the rapid growth of the intelligent market and the demand for green transformation. This serves as a model for developing green economies in other nations. Second, based on the fundamental properties of AI, this study provides an economic framework for analyzing the effects of artificial intelligence on CO₂ emissions. Third, this study improves the mechanism underlying the effect of economic policy uncertainty (EPU) on CO₂ emissions in the selected sample of countries. As noted in earlier studies, higher levels of EPU affect various macroeconomic indicators, including innovations, financial development, capital investment at the company level, the tourism sector, economic growth, and working capital and profits [3]. By analyzing the correlations between the two, this study concludes that renewable energy is the best method to combat environmental deterioration and increasing CO₂ emissions.

The rest of this paper is structured as follows. Section 2 provides an overview of the existing literature and a detailed study of the relevant theoretical concepts. Section 3 and Section 4 describes the data sources and the specific methodologies employed in this study. Section 5, represent the empirical results and discussion. Section 6 concludes the results of this study.

2. Literature Review and Hypothesis Development

2.1. AI Technology and CO₂ Emissions

The connection between new technologies and increases in carbon emissions has been the subject of many academic studies. Ref. [18] examined how patent technology affects pollution levels. To shed further light on this association, the authors applied the cluster method to panel data from many provinces in China; their study determined the importance of technical progress in reducing CO₂ emissions. It was also determined that Eastern China is more likely to embrace environmental innovations and technology than other parts of the country. Across OECD member countries, adopting RENE regulations has a positive effect on the development of environmentally sustainable technology, which is in line with the findings of [19]. The authors also point out that enabling competition that may favor poor green solutions is counterproductive and that passing RENE laws effectively improves environmental standards. Likewise, ref. [20] examined the relationship between R&D spending and carbon emissions in a panel of Mediterranean economies during the period 1990–2016, using the generalized method of moments (GMM) empirical technique. The data analysis showed a negative correlation between R&D spending and greenhouse gas output. The analysis indicated that research and development spending appeared to have a unidirectional causal relationship with CO₂. The study’s results provided strong evidence for the claim that promoting energy-efficient technology might significantly aid in reducing environmental damage.

According to [21], technological improvement is the primary means to decrease CO₂ emissions. Improvements in efficiency and scale expansion had a “double-edged sword effect”. Ref. [22] found that technical advancement had an unpredictable effect on pollutant emissions.

Technological advancements reduce environmental pollution by increasing the industrial sector’s efficiency in using multiple productive resources and lowering energy consumption per output unit [23]. The counterargument is that technological progress might cause production scales to rise, leading to more significant pollution and negating the benefits of higher efficiency in reducing emissions [24]. Emerging technologies, such as digitalization and AI, are thriving in China’s developing economy, which is presently approaching the era of Industry 4.0. An emerging area of study is the possibility that these technologies might lessen the amount of pollution released into the atmosphere.

According to [24], the development of the digital economy has an “industrial pollution reduction effect”, with the application of digital technology reducing industrial pollution emissions without causing yield loss. Ref. [25] concluded that the Internet reduced environmental pollution in the studied region and surrounding areas. Their model test of mediating effects demonstrated that encouraging industrial upgrading was the primary route through which the Internet affects environmental pollution. Using heavy metal enterprises as an example, ref. [26] proposed that the digital transformation of enterprises can achieve pollution reduction. However, under the agglomeration effect, there was a U-curve relationship between the digital transformation of enterprises and pollution reduction, with the final effect constrained by external scale. [23] examined the digital transformation of enterprises at the micro level and demonstrated that the use of digital equipment triggered economic scale expansion, leading to increased energy consumption. Simultaneously, the resulting technological and structural changes improved production efficiency and decreased energy consumption per output unit, significantly reducing pollution emissions. Ref. [27] showed that AI technologies had the potential to revolutionize several climate-friendly initiatives, such as the detection of greenhouse gas (GHG) leakage from pipelines, the monitoring of deforestation, and the invention of new materials with lower carbon footprints. However, statistics showing the environmental impact of AI are few or non-existent. AI businesses, such as OpenAI, should enhance their transparency regarding the expenses associated with system development, deep learning algorithm processing, and the training of their large language models (LLMs). It is critically important that complete transparency be afforded a higher priority as various nations tackle the task of AI regulation, especially in relation to the carbon emissions linked to the business.

By next year, the vast number of internet-connected devices might account for as much as 3.5% of worldwide carbon emissions. Computers and servers at data centers would quickly overheat if not for the constant, heavy usage of air conditioners, which contribute significantly to the overall energy consumption of these facilities. The AI industry significantly relies on data centers. If its usage and distribution continue to grow, it will inevitably result in increased carbon emissions from data centers in the coming years.

Hypothesis 1:

AI leads to an increase in CO₂ in East Asia and the Pacific.

2.2. Economic Policy Uncertainty and CO₂

Ref. [28] found both an overestimation and an underestimation of the implications of economic policy uncertainty for environmental policymaking. Ref. [29] assessed two strategies to provide a roadmap for Japan to achieve its challenging ecological and energy-related objectives. Their study’s conclusions indicated that, while air travel had a short-term effect, carbon dioxide emissions had a long-term relationship with GDP growth, renewable energy, and the economic complexity index.

It is assumed that uncertainty in economic policy significantly impacts the financial policies, investment plans, and consumer purchasing power of firms. According to [30], monetary policy uncertainty also has a nonlinear effect on inflation expectations and economic growth. These results suggest that it would be worthwhile to estimate the impact of EPU on environmental quality. As expected, relatively high EPU affects energy consumption, CO₂ emissions, and economic growth, all of which affect the sustainability and competitiveness of the environment [31].

As [32] explained, variations in production are the primary cause of wealth inequality among countries. It is impossible to overstate the significance of technological transfers in determining a country’s productivity. In most countries, foreign sources of technology transfer account for up to 90% of the improvement in domestic productivity [33]. Rapid efforts are required to decarbonize the energy sector because of global warming and environmental damage. According to [34], energy efficiency and technological improvement are the primary drivers of a seamless transition from fossil fuels to renewable sources. Although most technology is generated in wealthier countries, it is still possible for technical progress to affect climate change patterns in developing countries through the transfer of knowledge. Refs. [2,35] used the spillover and feedback effects model to examine the impacts of CO₂ emissions in seven BRI zones from 2000 to 2015. According to their research, CO₂ emissions ratios increased over time in North Africa, Northeast Asia, and Western Asia, but dropped in Central Asia. The impact of technological changes such as regional technology transfer, foreign technology imports, and local innovation on CO₂ emissions in China was evaluated using panel data from 2008 to 2017 [36]. Ref. [37] used the generalized Divisia index approach (GDIM) to examine the influence of RENE on CO₂ emissions for a panel of 25 BRI countries between 2005 and 2019. Their innovative study showed that the growth in RENE sources was a significant factor in CO₂ emissions in most BRI countries. Long-term financial development had an M-shaped influence on CO₂ emissions in the United States, Japan, and Canada; an inverted N-shaped effect in the United Kingdom, France, and Italy; and a W-shaped impact in Germany [38]. Similar variability was revealed in [36] empirical investigation of the influence of advances in green technology across 264 Chinese prefecture-level cities from 2006 to 2017.

Hypothesis 2:

Economic policy and CO₂ emissions have a positive relationship.

2.3. Renewable Energy Consumption and CO₂ Emissions

Ref. [39] thoroughly analyzed the relationship between RENE sources and CO₂ emissions in 128 countries between 1990 and 2014. Their study results indicated that switching to RENE might drastically reduce carbon emissions. On the other hand, CO₂ emissions in Europe were drastically different from those in the other five areas studied. According to the results of econometric research conducted by [40], carbon emissions decreased significantly across 16 EU countries when the pool mean group (PMG) approach was used for the data analysis. The use of alternative energy sources was credited for this decrease. Ref. [40] followed a methodology similar to that of [41]. Their study compared results from 24 African countries using data collected between 1985 and 2015. The data available in the African context supported the environmental Kuznets curve (EKC) hypothesis. According to the advocates, the availability of investments that emphasized ecological issues was crucial to the success of sustainable urban growth. The researchers were confident that using environmentally friendly forms of energy would allow them to meet their sustainable development goals.

Similar work was carried out by [42] in order to assess the role of renewable and non-renewable sources in mitigating greenhouse gas emissions. Ref. [43] used a panel dataset that included the years 1996–2012 for their analysis. Furthermore, the research conducted by [44] examined how imports and exports influenced carbon emissions in seven countries using a panel quantile regression approach. Their results demonstrated a strong connection among imports, exports, and carbon emissions. Recent research by [45] indicated that there is a strong correlation between rising CO₂ emissions in Asian countries and the unpredictability of their economic strategies. In addition, the pollution halo theory, based on a large body of prior academic research, offers an alternative explanation.

Policymakers and environmental economists worldwide are actively seeking strategies and solutions to address these pressing ecological difficulties due to the recent growth of global environmental concerns [46]. These studies consider factors such as international trade, knowledge transfer, and RENE use when examining the increase in CO₂ [47,48,49] (Researchers have examined how using renewable vs. non-RENE sources affects carbon emissions. As previously indicated, such research has been conducted in various countries using various econometric methods, techniques, and outcomes. Recent research by showed that reducing political risk and using ICT hold promise for effectively addressing CO₂ emissions in Morocco. Ref. [47] examined 42 countries in Sub-Saharan Africa to determine whether there was a connection between their utilization of RENE and their CO₂ emissions. The researchers included healthcare spending as a separate variable in their analysis, and the research covered a wide range of years, from 1995 through 2011. According to the available statistics, RENE use was linked to lower carbon CO₂ emissions.

Hypothesis 3:

Renewable energy consumption decreases CO₂ emissions in East Asia and Pacific countries.

3. Methodology and Data Sample

This section explains the econometric techniques employed in our study, including unit root tests, cross-sectional dependency tests, panel co-integration estimates, the Granger causality test, two-stage least squares, and the two-step generalized method of moments (GMM). Figure 1 shows the overall conceptual framework if the study. This study compiled its findings using data for 14 East Asian and Pacific countries from the World Development Indicators (WDI) and the Our World by Oxford University database.

Model of this study

CO₂ = β₀ + β₁ AI_ij + β₂ EPU_ij + β₃ RENE_ij + CV_ij + µ

(1)

where:

• CO₂: carbon dioxide emissions.
• AI: artificial intelligence.
• EPU: economic policy uncertainty.
• RENE: renewable energy.
• β: coefficient.
• CV: control variables.
• µ: error term.

Description of Variables

CO₂: Emissions of carbon dioxide, abbreviated as CO₂, are byproducts of many industrial processes, including but not limited to the combustion of fossil fuels, the production of cement, and the use of gas as a fuel source. To measure CO₂ emissions, we used CO₂ emissions in metric tons per year. The data were collected from the World Bank World Development Indicators [50].

Artificial intelligence (AI): There are several academic and commercial uses of AI. AI is a multipurpose, all-purpose technology, similar to electricity or computers. Even though AI and the cloud operate virtually, they have many real-world impacts. In addition to increasing energy consumption and resource demands, they amplify emissions of greenhouse gases. One manifestation of this issue is increased energy use. This study’s main econometric analytical problem involves finding data in the format of a cross-country panel dataset that can quantify the degree of AI. A variety of proxies, including high-tech specialists, patent filings, and AI investments in AI research, have been used in previous studies to quantify AI. This study used AI research publications as a proxy for evaluating AI, as there are variable amounts of data for each country. Similarly, this proxy shows the intentions and processes of each country moving toward the adoption of AI in our panel [51].

Economic policy uncertainty (EPU): Uncertainty concerning government policies and regulatory frameworks for the near future is known as economic policy uncertainty. A rise in EPU may lower CO₂ emissions by causing decreases in investment, consumption, and output. However, it may also impact innovation, R&D methods, and the usage of renewable energy sources, which might eventually result in increased CO₂ emissions. EPU can, therefore, either lessen or increase the impacts of environmental deterioration [52].

Renewable energy consumption: Renewable energy comes from sources that are naturally renewing yet limited in terms of flow. Renewable resources are almost endless in terms of length, but they are restricted in terms of the quantity of energy that is accessible per unit of time. We used per capita data for this variable (renewable energy consumption), as demonstrated by [53].

Exports: Concerns regarding the relationship between commerce and environmental degradation have arisen in response to the growing number of international trade agreements and the tightening of global value chains. This raises the question of the environmental consequences of trading. The liberalization of trade and investment might encourage businesses to embrace stricter environmental regulations. An increasing degree of international economic integration exposes a nation’s export industry to ecological regulations enforced by major importers [54].

Labor force participation: Although it is impossible to overstate human capital’s role in promoting sustainable development, there needs to be more discussion in the literature about whether or not labor force participation supports environmental sustainability. However, ecological quality has declined over time because of the ongoing increases in greenhouse gas (GHG) emissions worldwide. Climate change and other socioeconomic issues related to the dynamics of the labor market are caused by rising GHG emissions. Designing strategies to ensure social fairness through the creation of good jobs and to improve environmental quality is a growing priority for development organizations and governments.

Estimation

To analyze the data collected, this study used the following estimation tests.

Cross-sectional dependence

As a crucial component of panel data models, cross-sectional dependence (CD) may be influenced by the cultural, economic, and geographical links among the sampled nations. The cross-sectional dependency of East Asia and Pacific economies is a natural consequence of their close economic relationship. Ref. [55] noted that it is imperative to assess the likelihood of CD; failing to do so would lead to inaccurate and inconsistent estimates of stationarity and co-integrating traits.

Therefore, following [56], the CD test was employed in our investigation, considering its capacity to handle data with more constrained time frames and smaller cross-sectional units. The generalized method of moments (GMM) estimator accounts for the possibility of cross-sectional dependency in the data, eliminating endogeneity concerns in the regressors [57]. Unlike other estimation methods, such as least-squares regressions, GMM accounts for country-specific heterogeneities, eliminating dynamic panel bias. It is essential to conduct the GMM analysis after the cross-sectional dependence, unit root, and cointegration investigations. First, we use the Pesaran CD test, as described by [58], to determine whether there is a cross-sectional dependency problem. To reject the null hypothesis of cross-sectional independence, this technique estimates a test statistic that forecasts CD difficulties for each variable (or series). We used the CD estimate method proposed by Breusch and Pagan (1980), which takes into account a null hypothesis of cross-sectional independence, similar to that proposed by [59], for the robustness check.

Long-run estimation test

Popular panel data estimation methods, such as FE, RE, DOLS, FMOLS, and GMM, all rely on slope homogeneity across cross-sections, which could significantly impact the results.

$${\beta }_{NT}^{*}=N^{-1}{{\sum }_{i=1}^N\left[{\sum }_{i=1}^T{\left(X_{it}-{\overline{X}}_i\right)}^2\right]}^{-1}{\sum }_{i=1}^T\left(X_{it}-{\overline{X}}_i\right){\gamma }_{it-T_{\tau l}}$$

$${\gamma }_{it}^{*}=\left({\gamma }_{it}-{\overline{\gamma }}_i\right)-{\displaystyle \frac{L_{21l}}{L_{22l}}}\Delta X_{it},{\widehat{\tau }}_l={\widehat{\Gamma }}_{21l}+{\widehat{\Omega }}_{21l}^0-{\displaystyle \frac{L_{21l}}{L_{22l}}}\left({\widehat{\Gamma }}_{22l}+{\widehat{\Omega }}_{22l}^0\right)$$

The DOLS is written as follows:

$${\gamma }_{it}={\alpha }_i+{\beta }_i{X}_{it}+\sum _{j=-ji}^{jl}{\theta }_{ij}\Delta X_{it-j}+{{\epsilon }_{it}}^{*}$$

$$\beta {\ast }_{DOLS}=N^{-1}\sum _{i=1}^N{\left(\sum _{t=1}^T{Z}_{it}{Z}_{it}^i\right)}^{-1}\left(\sum _{t=1}^T{Z}_{it}{\gamma }_{it}^{*}\right)$$

$$Z_{it}=\left(X_{it}-{\overline{X}}_l,\Delta X_{it-j},\dots ,\Delta X_{it+k}\right)2\left(K+1\right)$$

$$Y_t=\sum _{j=0}^k{\varphi }_i{y}_{t-i}+{\epsilon }_t$$

$${\varphi }_i=\left\{\begin{array}{l}{I}_k,i=0\\ {\displaystyle \sum _{j=1}^i}{\varphi }_{t-j}{A}_j,i=1,2,\dots \end{array}\right.$$

$$y_{it+h}-E\left[y_{it+h}\right]=\sum _{i=0}^{h-1}{\epsilon }_i\left(t+h-1\right){\varphi }_i$$

$$\sum _{i=0}^{hl}{\theta }_{nm}^2={\sum _{i=0}^{hl}\left(i_m^{\prime }K{\varphi }_{in}\right)}^2$$

4. Data Analysis

This study investigated the impact of AI, economic policy uncertainty, and renewable energy use on environmental quality in a panel of 14 East Asian and Pacific economies from 2000 to 2023. The summary statistics and correlation matrix are presented in

Table 1. Summary statistics.

	CO2	AI	EPU	RENC	EXP	BFP
Mean	1.144614	−6.752220	4.808009	2.219551	3.912695	3.825545
Median	1.360290	−6.354429	4.794594	3.114838	3.933426	3.841317
Maximum	3.077580	−1.551640	5.787578	4.401584	5.433695	4.223207
Minimum	−1.820287	−12.92531	4.164067	−4.605170	2.852507	3.135320
Std. Dev.	1.156430	2.502786	0.430687	2.013087	0.589109	0.256398
Observations	254	254	254	254	254	254

and

Table 2. Pairwise correlations.

Variables	CO2	AI	EPU	RENC	EXP	BFP
CO2	1.000
AI	0.492	1.000
EPU	0.155	0.151	1.000
RENC	−0.475	−0.232	−0.105	1.000
EXP	0.249	−0.129	−0.064	−0.447	1.000
BFP	−0.568	−0.193	−0.223	0.629	−0.233	1.000

Source: authors’ calculations.

.

This study incorporated various variables. As the summary statistics indicate, all variables exhibit significant variability in their minimum and maximum values. Similarly, the matrix reveals a negative association between renewable energy and economic policy uncertainty and a positive correlation between the dimensions of AI and CO₂ emissions.

Before examining the presence of unit root and cointegration among the variables, we assessed the cross-sectional dependence among the nations included in the sample with the rise of liberalization and globalization. During this period, there has been growing economic and social interconnectedness across nations. Consequently, the actions implemented in one country can have an impact on another nation as well. Following [56], the cross-sectional dependence test was utilized to ascertain the presence of CD within the chosen East Asia and Pacific countries. The findings displayed in

Table 3. Cross-sectional dependence tests.

Variables	Breusch–Pagan LM	Pesaran Scaled LM	Pesaran CD
CO2	1076.1676	73.0254	14.8976
	(0.0000)	(0.0000)	(0.0000)
EPU	467.6181	31.19442	2.93362
	(0.0000)	(0.0000)	(0.0034)
AI	468.8277	31.29126	−0.611140
	(0.0000)	(0.0000)	(0.5411)
RENE	494.2888	33.32978	3.515709
	(0.0000)	(0.0000)	(0.0000)
EXP	394.2699	25.3218	14.8976
	(0.0000)	(0.0000)	(0.0000)
LBFP	838.0242	55.3731	22.9312
	(0.0000)	(0.0000)	(0.0000)

Source: authors’ calculations.

validate the presence of a correlation among CO₂ emissions, AI, renewable energy consumption, time, and economic policy uncertainty in the sample nations. This suggests that any alteration in these factors in East Asian–Pacific countries can also impact the other Asian andPacific countries. Table 3 presents the findings of the slope homogeneity test introduced by [56] for all three regression models in this study.

Both the constant term only and constant term and trend term versions of the three-unit root test techniques used in this study are shown in

Table 4. Panel unit root tests.

Variable	Level		First Difference
	With Constant	Constant and Trend	With Constant	Constant and Trend
Levin, Lin, and Chu
CO2	1.14929	0.40833 *	−2.74790 **	−2.24790 ***
AI	−2.4150	−3.6099 **	−7.5668 ***	−5.4556 ***
EPU	1.9604	−4.2088 ***	−9.1004 ***	−9.6434 ***
RENE	0.128158	1.271932	−0.5498	0.047541 *
Im, Pesaran, and Shin test
CO2	2.30494	1.68899	−4.4875 ***	−3.3289 ***
AI	−0.7038	−4.2729 ***	−10.880 ***	−8.0240 ***
EPU	4.3012	−3.4030 ***	−8.7916 ***	−7.0831 ***
RENE	2.9665	1.97422	−4.3652 ***	−3.84803 ***
ADF-Fisher Chi-square test
CO2	23.5113	22.2661	72.0801 ***	59.4662 ***
AI	49.89272	72.3376 ***	164.1347 ***	115.6323 ***
EPU	3.32278	52.1911 **	126.2833 **	97.14846 ***
RENE	15.49658	11.82574	65.02550 **	58.6137 ***

Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Source: authors’ calculations.

. Except for the LLC trial, every one of the discovered variables in the five trials rejected the null hypothesis at the 1% significance level. As a result, we examined the data using a first-order differential. We found that at the crucial 1% level, no hypotheses were rejected for each variable’s unit root. However, this indicates the possibility of spurious regression; thus, the KAO test for cointegration is required.

Table 5. Kao test for cointegration.

		Null Hypothesis	t-Statistics	Probability
1	ADF	No-cointegration	1.599509	0.0549
2	ADF	No-cointegration	−1.860175	0.0314
3	ADF	No-cointegration	−1.525220	0.0636

Source: authors’ calculations.

presents the findings of the cointegration test for CO₂ emissions, AI, economic policy uncertainty, and renewable energy. All three model groups rejected the initial hypothesis, suggesting that the panel data exhibit a cointegration relationship. The findings validate the existence of a long-term equilibrium cause-and-effect association among the variables, thus facilitating further investigation of this relationship.

The estimation results for the FMOLS and DOLS panel models are presented in

Table 6. Benchmark results for CO₂, AI, EPU, and RENE (FMOLS and DOLS).

Variables	CO2	CO2	CO2
FMOLS
AI	1.316716 *
	(0.699672)
EPU		0.867909 **
		(0.392765)
RENE			−2.328345 **
			(1.058345)
DOLS
AI	0.166554 ***
	(0.066719)
EPU		0.237011 **
		(0.099588)
RENE			−0.365873 **
			(0.117710)

Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Source: authors’ calculations.

. According to the parameters, DOLS provides a more accurate match. It may be inferred that a 1% increase in AI is associated with a corresponding 0.1665% increase in CO₂ emissions. Likewise, a 1% rise in economic policy uncertainty will result in a 0.237% increase in CO₂ emissions, leading to environmental damage. Additionally, CO₂ emissions will fall by −0.3658 if renewable energy usage increases. Policy ambiguity and adopting digitalization/AI will generally impact environmental degradation, but renewable energy consumption will exacerbate the ecological situation. Similarly, to check the robustness of the data, we incorporated the Hausman fixed effect and generalized method of moment to confirm the relationship.

Table 7. Hausman fixed effect.

Variables	CO2	CO2	CO2
AI	0.195 ***
	(0.0234)
EPU		0.00191 *
		(0.00103)
REN			−0.431 ***
			(0.0337)
LEXP	0.386 ***	0.323 ***	−0.249 ***
	(0.104)	(0.114)	(0.0947)
LBFP	−1.794 ***	−2.149 ***	−0.515 **
	(0.244)	(0.267)	(0.244)
Constant	7.908 ***	7.882 ***	5.018 ***
	(1.102)	(1.253)	(0.962)
Observations	265	269	258
Number of years	22	22	22
R-squared	0.48	0.63	0.59

Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1

and

Table 8. GMM.

Variables	CO2	CO2	CO2
AI	0.214 ***
	(0.0241)
EPU		0.595 **
		(0.230)
REN			−0.455 ***
			(0.0358)
LEXP	0.382 ***	0.352 ***	−0.276 ***
	(0.103)	(0.120)	(0.0979)
LBFP	−1.788 ***	−2.028 ***	−0.427 *
	(0.239)	(0.285)	(0.253)
Constant	8.021 ***	4.696 **	4.841 ***
	(1.068)	(1.955)	(0.992)
Observations	265	269	258
Number of years	22	22	22
Arellano–Bond test for AR(1) in first differences: z = −3.47 Pr > z = 0.001 Arellano–Bond test for AR(2) in first differences: z = 0.80 Pr > z = 0.421
Sargan test of overid. restrictions: chi2(14) = 238.35 Prob > chi2 = 0.000

Standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1.

show the results of the Hausman test and GMM. The relationship between CO₂ emissions and AI and economic policy uncertainty was positive, whereas the result for renewable energy consumption was the opposite. This means that a unit increase in AI adoption and monetary policy uncertainty will contribute 1.316 and 0.867% to environmental degradation, respectively. The same results were found in the GMM.

Impulse response and variance decomposition

Before analyzing the pulse effect and variance decomposition as endogenous variables in VAR systems, defining the best lag order for mechanization, rainfall, and agricultural carbon emissions is recommended. This study presents the following five approaches for comprehensive judgment: the LR test statistic (LR), Final Prediction Error (FPE), Akaike Information Criterion (AIC), Schwarz Information Criterion (SIC), and Hannan–Quinn Information Criterion (HQ). We found that lag order 2 is the best lag term, as indicated in

Table 9. Optimal lag period selection.

Lag	LogL	LR	FPE	AIC	SC	HQ
0	−1078.181	NA	4.996539	12.96025	13.03494	12.99057
1	136.9475	2357.495	0.0000029	−1.400569	−1.027156	−1.249009
2	159.4494	42.57845	0.00000268	−1.478436	−0.806294 *	−1.205628
3	175.6993	29.96996	0.00000267	−1.481429	−0.510557	−1.087374
4	181.8982	11.13563	0.00000301	−1.364050	−0.094448	−0.848746
5	206.4746	42.97200	0.00000273	−1.466762	0.101570	−0.830211
6	239.9958	57.00601	0.00000222	−1.676596	0.190466	−0.918797
7	318.4096	129.5942	0.00000106	−2.424067	−0.258275	−1.545020
8	353.1120	55.69001	0.00000085	−2.648048	−0.183526	−1.647753

Source: authors’ calculations. * Shows the lag period.

. Figure 2 was created following this order, which clearly shows that all the roots fall inside the unit circle, meaning this VAR model meets the requirements for variance decomposition and impulse response analysis.

The VAR model of a standard deviation of the random disturbance impact on the trajectories of other variables and the influence of current and future values may be visually represented using the impulse response function. Thus, using an impulse response function diagram, we further examined how AI and EXP affect the other CO₂ emissions [29]. We set a reaction time of twenty years. In Figure 3, the range of the potential impulse response is indicated by dotted lines on either side of the solid lines; the abscissa shows the lag length of the effect, and the longitudinal coordinates show the degree of reaction. Table 9 shows the optimal lag period selection.

Table 10. Variance decomposition.

Variance Decomposition of CO2
Period	S.E.	CO2	AI	EXP
1	0.083	100.000	0.000	0.000
2	0.132	99.456	0.521	0.023
3	0.169	99.146	0.822	0.032
4	0.199	98.899	1.068	0.034
5	0.225	98.678	1.290	0.031
6	0.246	98.465	1.507	0.028
7	0.265	98.251	1.724	0.025
8	0.282	98.032	1.946	0.022
9	0.298	97.806	2.174	0.020
10	0.312	97.572	2.410	0.018
11	0.324	97.328	2.654	0.018
12	0.336	97.076	2.906	0.018
13	0.347	96.815	3.167	0.019
14	0.357	96.544	3.435	0.020
15	0.367	96.265	3.712	0.023
16	0.375	95.978	3.997	0.025
17	0.384	95.683	4.289	0.029
18	0.392	95.379	4.588	0.032
19	0.399	95.069	4.895	0.037
20	0.406	94.751	5.208	0.041
Variance Decomposition of AI:
Period	S.E.	CO2	AI	EXP
1	0.2694	1.8192	98.1808	0.0000
2	0.3544	1.9866	98.0029	0.0105
3	0.4239	2.0386	97.9330	0.0284
4	0.4821	2.0457	97.8986	0.0557
5	0.5331	2.0302	97.8775	0.0923
6	0.5788	2.0030	97.8587	0.1383
7	0.6204	1.9694	97.8368	0.1938
8	0.6586	1.9323	97.8089	0.2588
9	0.6942	1.8934	97.7733	0.3333
10	0.7275	1.8537	97.7290	0.4173
11	0.7588	1.8137	97.6755	0.5108
12	0.7884	1.7739	97.6124	0.6138
13	0.8165	1.7345	97.5393	0.7262
14	0.8433	1.6957	97.4561	0.8482
15	0.8690	1.6577	97.3627	0.9795
16	0.8935	1.6205	97.2592	1.1203
17	0.9171	1.5842	97.1453	1.2705
18	0.9398	1.5488	97.0213	1.4300
19	0.9617	1.5143	96.8870	1.5987
20	0.9830	1.4807	96.7426	1.7767
Variance Decomposition of EXP:
Period	S.E.	CO2	AI	EXP
1	0.083	0.230	3.641	96.128
2	0.121	0.422	2.108	97.470
3	0.149	0.540	1.618	97.843
4	0.173	0.596	1.351	98.053
5	0.194	0.620	1.179	98.201
6	0.213	0.625	1.055	98.319
7	0.230	0.621	0.959	98.420
8	0.246	0.611	0.881	98.508
9	0.261	0.598	0.815	98.588
10	0.275	0.582	0.757	98.660
11	0.289	0.566	0.707	98.727
12	0.302	0.549	0.662	98.789
13	0.314	0.531	0.622	98.847
14	0.326	0.514	0.586	98.900
15	0.338	0.497	0.552	98.951
16	0.349	0.480	0.522	98.998
17	0.360	0.463	0.494	99.042
18	0.370	0.448	0.468	99.084
19	0.381	0.432	0.445	99.123
20	0.391	0.417	0.423	99.160

Source: authors’ calculations.

represents the results of the variance decomposition analysis of CO₂ emissions, AI, and exports. The results show that the variation in CO₂ emissions is self-generated in the short term. Similarly, in the short term, AI adoption has minimal impact on variations in CO₂ emissions. Our study’s results align with the previous research conducted by [60], and we can see that the value for period 20 is 5.208, compared with 94.751. This minimal variation is because AI is an emerging cutting-edge technology, and most countries are in line to adopt it. It is also the case that AI has not yet been embraced fully. Figure 3 shows the impulse response between AI and CO₂ emission. In addition, the environmental concerns of AI are relatively understudied at present, compared to other phenomena.

Figure 4 and Figure 5 represent the impulse response of the relationship between CO₂ emissions and economic policy uncertainty (EPU). The graph shows that the reaction of CO₂ to CO₂ declines over a certain period, while, in the second graph, the relationship between CO₂ emissions and EPU is initially positive. In contrast, in the later stages, it becomes damaging.

Table 11. VDC of CO₂ and EPU.

Variance Decomposition of CO2
Period	S.E.	CO2	EPU	EXP
1	0.0830	100.0000	0.0000	0.0000
2	0.1330	99.4992	0.4993	0.0015
3	0.1705	99.0061	0.9855	0.0083
4	0.2001	98.5239	1.4605	0.0156
5	0.2247	97.9988	1.9785	0.0228
6	0.2457	97.3764	2.5932	0.0304
7	0.2642	96.6016	3.3595	0.0389
8	0.2808	95.6112	4.3398	0.0489
9	0.2960	94.3278	5.6112	0.0609
10	0.3103	92.6547	7.2698	0.0755
11	0.3243	90.4728	9.4338	0.0934
12	0.3384	87.6418	12.2430	0.1152
13	0.3534	84.0075	15.8509	0.1416
14	0.3701	79.4213	20.4055	0.1732
15	0.3895	73.7756	26.0143	0.2101
16	0.4131	67.0531	32.6951	0.2517
17	0.4424	59.3802	40.3230	0.2968
18	0.4795	51.0567	48.6004	0.3429
19	0.5267	42.5350	57.0777	0.3873
20	0.5868	34.3352	65.2376	0.4272
Variance Decomposition of EPU:
Period	S.E.	CO2	EPU	EXP
1	26.5331	0.3259	99.6741	0.0000
2	42.7444	1.6824	98.0691	0.2485
3	58.8412	2.5311	97.1289	0.3400
4	76.1302	3.0476	96.5675	0.3850
5	95.5157	3.3739	96.2154	0.4107
6	117.7649	3.5928	95.9799	0.4273
7	143.6526	3.7476	95.8135	0.4389
8	174.0314	3.8616	95.6911	0.4473
9	209.8775	3.9480	95.5983	0.4538
10	252.3306	4.0148	95.5264	0.4588
11	302.7338	4.0673	95.4700	0.4627
12	362.6776	4.1090	95.4251	0.4659
13	434.0518	4.1423	95.3893	0.4684
14	519.1052	4.1691	95.3604	0.4705
15	620.5170	4.1906	95.3371	0.4723
16	741.4812	4.2081	95.3182	0.4737
17	885.8076	4.2223	95.3029	0.4748
18	1058.0415	4.2337	95.2904	0.4758
19	1263.6072	4.2431	95.2803	0.4766
20	1508.9786	4.2507	95.2720	0.4773
Variance Decomposition of EXP:
Period	S.E.	CO2	EPU	EXP
1	0.0826	0.1527	0.0734	99.7739
2	0.1199	0.2353	3.5233	96.2414
3	0.1497	0.1782	5.4266	94.3952
4	0.1753	0.1321	6.9265	92.9414
5	0.1983	0.1043	8.3646	91.5311
6	0.2199	0.0943	9.9084	89.9973
7	0.2407	0.1032	11.6607	88.2362
8	0.2612	0.1335	13.7039	86.1625
9	0.2820	0.1890	16.1145	83.6965
10	0.3037	0.2738	18.9651	80.7611
11	0.3268	0.3921	22.3208	77.2870
12	0.3522	0.5479	26.2306	73.2216
13	0.3806	0.7435	30.7145	68.5420
14	0.4132	0.9791	35.7503	63.2706
15	0.4512	1.2514	41.2615	57.4871
16	0.4964	1.5535	47.1131	51.3334
17	0.5505	1.8749	53.1205	45.0045
18	0.6158	2.2029	59.0703	38.7268
19	0.6949	2.5237	64.7511	32.7252
20	0.7908	2.8254	69.9840	27.1906

shows the variance decomposition between CO₂ emissions and EPU. In the short term, the variation in CO₂ is self-generated while, in the long term, the variation in CO₂ arises from EPU. The results show that the value of EPU in year 20 was recorded at 65.2376, which is higher than 34.3352. Our study’s results support those of [61].

Table 12. VDC for CO₂ and RENE.

Variance Decomposition of CO2:
Period	S.E.	CO2	RENE	EXP
1	0.07788329	100	0	0
2	0.1312	99.9869	0.0036	0.0095
3	0.1736	99.9831	0.0048	0.0122
4	0.2078	99.9846	0.0040	0.0114
5	0.2360	99.9872	0.0031	0.0096
6	0.2597	99.9889	0.0031	0.0080
7	0.2800	99.9885	0.0045	0.0070
8	0.2976	99.9854	0.0077	0.0069
9	0.3131	99.9794	0.0128	0.0079
10	0.3267	99.9703	0.0198	0.0099
11	0.3389	99.9583	0.0287	0.0130
12	0.3498	99.9431	0.0397	0.0172
13	0.3596	99.9250	0.0525	0.0225
14	0.3685	99.9039	0.0673	0.0288
15	0.3766	99.8799	0.0839	0.0362
16	0.3839	99.8530	0.1024	0.0446
17	0.3906	99.8235	0.1226	0.0539
18	0.3967	99.7912	0.1446	0.0642
19	0.4023	99.7565	0.1682	0.0754
20	0.4074	99.7192	0.1934	0.0874
Variance Decomposition of RENC:
Period	S.E.	CO2	RENE	EXP
1	0.077906609	10.17802357	89.82197643	0
2	0.1205	10.4838	89.3727	0.1435
3	0.1527	10.5850	89.1621	0.2529
4	0.1790	10.5266	89.1335	0.3399
5	0.2016	10.3687	89.2130	0.4183
6	0.2214	10.1536	89.3519	0.4946
7	0.2392	9.9071	89.5212	0.5717
8	0.2555	9.6447	89.7041	0.6512
9	0.2705	9.3756	89.8904	0.7340
10	0.2845	9.1056	90.0739	0.8206
11	0.2976	8.8380	90.2507	0.9112
12	0.3099	8.5751	90.4186	1.0063
13	0.3215	8.3183	90.5759	1.1058
14	0.3325	8.0684	90.7217	1.2099
15	0.3430	7.8260	90.8552	1.3187
16	0.3530	7.5914	90.9763	1.4323
17	0.3626	7.3647	91.0846	1.5507
18	0.3718	7.1461	91.1800	1.6739
19	0.3807	6.9354	91.2627	1.8019
20	0.3892	6.7326	91.3326	1.9347
Variance Decomposition of EXP:
Period	S.E.	CO2	RENE	EXP
1	0.0858	0.0665	0.0012	99.9323
2	0.1237	0.3678	0.0085	99.6237
3	0.1523	0.5565	0.0108	99.4328
4	0.1762	0.6368	0.0096	99.3536
5	0.1970	0.6503	0.0078	99.3419
6	0.2156	0.6278	0.0067	99.3654
7	0.2325	0.5880	0.0069	99.4051
8	0.2481	0.5415	0.0088	99.4497
9	0.2627	0.4946	0.0123	99.4931
10	0.2764	0.4507	0.0176	99.5317
11	0.2893	0.4118	0.0246	99.5636
12	0.3016	0.3791	0.0333	99.5876
13	0.3133	0.3531	0.0436	99.6033
14	0.3245	0.3339	0.0554	99.6106
15	0.3353	0.3217	0.0688	99.6095
16	0.3456	0.3161	0.0836	99.6003
17	0.3557	0.3170	0.0998	99.5832
18	0.3654	0.3241	0.1172	99.5587
19	0.3748	0.3370	0.1359	99.5271
20	0.3839	0.3554	0.1557	99.4888

Source: authors’ calculations.

reports the variance decomposition results between CO₂ emissions and renewable energy. The variation in CO₂ emissions in the case of renewable energy consumption is slightly different from the relationship with AI. The variation in the short-term is totally self-generated while, in the long-term, the variance decomposition for CO₂ arises from renewable energy consumption. As noted in an earlier study by [62], adopting renewable energy production and consumption sources will decrease CO₂ emissions in East Asia and Pacific countries. Figure 6 shows the CO₂ emission trend in the selected countries.

5. Results and Discussion

Global warming is a serious environmental issue that affects every nation on the planet and is related to the long-term viability of human life. Furthermore, there is a strong link between agricultural carbon emissions and climate change. Thus, we sought to build an empirical framework to study the influences of AI, economic policy uncertainty, and renewable energy consumption on CO₂ emissions. Our approach produced empirical findings. First, the association among the variables was confirmed using a cross-sectional correlation test. We used the ADF, Im, Pesaran and Shin, and LLC tests to evaluate the stability of the unit root of panel data. According to the results, each variable is an integrated sequence of the same order and may be employed in the PVAR model. This also reveals that the variable after the first-order difference is stable. Additionally, we used the Kao test to confirm the long-term cointegration connection among the variables. The findings indicate that these three variables have a long-term integration connection. The link among the variables was then empirically studied using the Hausman test, the generalized method of moments, and VAR-based impulse response techniques. The findings demonstrate that the impulse response function more accurately captures the dynamic interaction among the examined factors. The FMOLS and DOLS test results confirm the robustness of the long-term findings. The causal link among the variables was also analyzed. We found that digitalization—for which we took AI as a proxy—showed a positive relationship with environmental degradation. Therefore, the more that AI is integrated into a country, the more vulnerable the environment is. It is worth noting that the impact of digitalization is twofold; on one hand, it can contribute to the economy by boosting production and the transparency of different projects, while, on the other hand, it can cause damage to the environment, leading to increases in CSR costs. Similarly, economic policy uncertainty also showed a positive relationship for the following reasons. First, the danger posed by EPU is uncertain because of its unexpected nature [6,63]. Second, since 1997, several financial crises have affected the world’s economies and financial markets [64]. Regrettably, the size, rate of spread, and complexity of EPU have all risen with each global economic crisis. Thus, the literature has demonstrated that EPU is significantly correlated with economic recessions [64], increased unemployment, and volatile exchange rates. On the other hand, it is unclear how EPU affects carbon emissions globally. Hence, an empirical study is necessary. Third, research indicates that a firm’s financial performance, investment choices, and business competitiveness are all impacted by EPU. Thus, we conclude that EPU impacts a firm’s carbon emissions. Real options and prospect theories provide the foundation of our argument. Fourth, earlier research indicated that the extraordinary global economic expansion over the last 25 years has come at the price of a clean and sustainable environment for future generations. The leading cause of environmental deterioration and the threat of climate change is global CO₂ emissions [35]. Similarly, the relationship between renewable energy consumption and CO₂ emissions is harmful, as many earlier studies have pointed out. In their foundational study, ref. [65] established what is now known as the Environmental Kuznets Curve (EKC) framework, which is the primary theory used to explain global CO₂ emissions trends over the long term. found a non-linear (inverted U-shaped) relationship between per capita GDP and environmental outcomes including CO₂ emissions. Multiple review studies have demonstrated the validity of the EKC hypothesis [66,67]. “Strong evidence in support of EKC” was found by [68], who completed a revised meta-analysis of 101 papers. The results of our study align with previous studies in the case of East Asia and Pacific countries.

6. Conclusions and Recommendations

Global warming and climate change are global issues that have gained tremendous momentum in spheres ranging from politics to the public domain and academia. At the same time, uncertainty in the economy, the emergence of AI, and the demand for renewable energy exacerbate these environmental concerns. This study focused on the relationships among these factors. Notably, earlier studies have examined similar factors for different countries. A significant contributor to climate change is the human-caused emission of gases into the atmosphere, including carbon dioxide. Energy consumption from renewable sources, EPU, AI, and CO₂ emissions are the subjects of this study’s dynamic interconnections. In this study, panel data for East Asian and Pacific nations from 2000 to 2023 were collected to facilitate an empirical analysis of the links among these factors. The variance decomposition test indicates that AI does not affect CO₂ emissions, whereas the benchmark regression indicates a positive link between AI and CO₂ emissions. To similar extents, the variance decomposition test and benchmark regression FE, RE, and GMM tests all demonstrate a robust positive correlation between economic policy uncertainty and CO₂ emissions. Carbon dioxide emissions are positively affected by an increase in EPU. Renewable energy significantly reduces CO₂ emissions in East Asian and Pacific nations. The findings show that a unit increase in the use of renewable energy results in a unit decrease in CO₂ emissions.

Policy recommendations

Based on the results of this study and by investigating the components of environmental degradation via an increase in CO₂ emissions, this study suggests the following policy recommendations, which will help to reduce CO₂ emissions. The first concerns the use of fossil fuels and inducement towards renewables overall. The most effective, efficient, and cost-effective tool for encouraging investments in clean technology is carbon pricing laws, which include emission trading systems and carbon taxes. Investments in environmentally friendly goods, regulations that promote a greener economy, and sustainable development projects are also important factors. Second, machine learning researchers should be incentivized to create more effective machine learning (ML) models to disclose their energy use and carbon footprints. An innovative model that incorporates these aspects from the outset has the potential to decrease emissions.

Third, to help their customers understand and lower their energy usage and carbon footprint, data center providers should be incentivized to share information regarding data center efficiency and the cleanliness of the energy supply by location. Cloud data centers use 30% less energy than the typical local data centers, and they have cooling and power delivery overheads of less than 10%. Finally, experts in machine learning (ML) deserve recognition for training models in the most environmentally friendly data centers, which are now frequently located in the cloud. They can produce 5 to 10 times fewer emissions for the same work, even in the same place.