Next Article in Journal
Can Green Finance Be a Regulator of “Water–Energy–Food” Synergy? Evidence from the Yangtze River Delta
Previous Article in Journal
Custom-Trained Large Language Models as Open Educational Resources: An Exploratory Research of a Business Management Educational Chatbot in Croatia and Bosnia and Herzegovina
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Greening Automation: Policy Recommendations for Sustainable Development in AI-Driven Industries

by
Nicoleta Mihaela Doran
1,*,
Gabriela Badareu
2,
Marius Dalian Doran
3,
Maria Enescu
4,
Anamaria Liliana Staicu
1,5 and
Mariana Niculescu
6
1
Department of Finance, Banking and Economic Analysis, Faculty of Economics and Business Administration, University of Craiova, 13 A.I. Cuza Street, 200585 Craiova, Romania
2
Doctoral School of Economic Sciences, Faculty of Economics and Business Administration, University of Craiova, 13 A.I. Cuza Street, 200585 Craiova, Romania
3
Doctoral School of Economics and Business Administration, West University of Timisoara, 300223 Timisoara, Romania
4
Department of Management, Marketing and Business Administration, Faculty of Economics and Business Administration, University of Craiova, 200585 Craiova, Romania
5
Filantropia Craiova Municipal Clinical Hospital, 1 Filantropiei Street, 200143 Craiova, Romania
6
Department of Agricultural and Forestry Technologies, Faculty of Agriculture, University of Craiova, 13 A.I. Cuza Street, 200585 Craiova, Romania
*
Author to whom correspondence should be addressed.
Sustainability 2024, 16(12), 4930; https://doi.org/10.3390/su16124930
Submission received: 30 April 2024 / Revised: 7 June 2024 / Accepted: 7 June 2024 / Published: 8 June 2024

Abstract

:
This study delves into the dynamic relationship between artificial intelligence (AI) and environmental performance, with a specific focus on greenhouse gas (GHG) emissions across European countries from 2012 to 2022. Utilizing data on industrial robots, AI companies, and AI investments, we examine how AI adoption influences GHG emissions. Preliminary analyses, including ordinary least squares (OLS) regression and diagnostic assessments, were conducted to ensure data adequacy and model readiness. Subsequently, the Elastic Net (ENET) regression model was employed to mitigate overfitting issues and enhance model robustness. Our findings reveal intriguing trends, such as a downward trajectory in GHG emissions correlating with increased AI investment levels and industrial robot deployment. Graphical representations further elucidate the evolution of coefficients and cross-validation errors, providing valuable insights into the relationship between AI and environmental sustainability. These findings offer policymakers actionable insights for leveraging AI technologies to foster sustainable development strategies.

1. Introduction

Climate change represents one of humanity’s foremost challenges [1]. The escalation of gas emissions has led to a rise in global temperatures, resulting in an increase in the frequency and severity of natural disasters [2]. The primary substances emitted into the atmosphere include carbon dioxide (CO2), nitrogen oxide (NO2), sulfur dioxide (SO2), and ozone (O3), each exerting distinct environmental effects [3]. Carbon dioxide stands out as the primary greenhouse gas driving global warming, with the combustion of fossil fuels such as coal and oil yielding significant CO2 emissions [4]. Given the global nature of climate issues, it is essential that decisionmakers implement a complex and diversified strategy [5,6,7]. In this regard, global leaders are compelled to cooperate and collaborate to mitigate the adverse impact of pollution on global warming. Consequently, an agreement was reached on an action plan to limit global warming, known as the Paris Agreement, which encompassed various measures, including targets for net greenhouse gas emissions [8]. Key measures undertaken by EU states to combat climate change include the adoption of the European Climate Law, setting a new target for reducing net greenhouse gas emissions by at least 55% by 2030 compared to 1990 levels, and mandating the achievement of climate neutrality by 2050 as a legal obligation. This ambitious package of measures is known as “Fit for 55 in 2030” [9].
Artificial intelligence (AI) is a branch of computer science focused on creating systems and machines that perform tasks requiring human-like intelligence [10]. These tasks include learning, reasoning, problem solving, perception, language understanding, and decision-making. AI encompasses several key aspects, such as machine learning (ML), natural language processing (NLP), computer vision, robotics, expert systems, cognitive computing, and autonomous systems [11]. Additionally, AI involves the consideration of ethics and governance to ensure responsible development and usage, addressing fairness, transparency, and accountability [12]. Machine learning, a subset of AI, involves training algorithms on large datasets to make predictions or decisions without explicit programming for each task [13]. Techniques within ML include supervised learning, unsupervised learning, and reinforcement learning. Natural language processing enables AI systems to understand, interpret, and respond to human language, covering tasks like speech recognition, language translation, sentiment analysis, and text generation [14]. Computer vision allows AI to interpret and process visual information, such as recognizing objects, faces, and scenes in images and videos. Robotics integrates AI with mechanical devices to perform tasks autonomously or semi-autonomously, exemplified by industrial robots, autonomous vehicles, and service robots [15]. Expert systems emulate human expert decision-making using rule-based logic to analyze information and provide recommendations [16]. Cognitive computing refers to AI technologies that simulate human thought processes in complex situations where answers may be ambiguous or uncertain. Autonomous systems operate independently of human intervention, using AI to make real-time decisions, as exemplified by self-driving cars and drones [17].
The key cognitive capabilities of “intelligent” machine systems enable the application of artificial intelligence across a variety of domains [18], with climate change being just one of them [19]. The ability of artificial intelligence to process vast amounts of unstructured, multidimensional data using sophisticated optimization techniques already facilitates the understanding of large-scale climate datasets and the forecasting of future trends [20]. The primary concerns related to climate change include mitigating the existing effects of climate change and efforts to reduce emissions, aiming to limit planetary warming. It is widely accepted that the continued increase in greenhouse gas concentrations, if not reversed, will result in significant changes to the global climate, with consequent effects on society and the global economy [21].
According to Akhshik et al. [22], machine learning, considered a component of a broader set of technological systems encompassed under the title of artificial intelligence [23], can be extremely useful in many domains, although there are some domains that may not fully utilize the advantages of automated learning. A concrete example is the field of greenhouse gas emission prediction, which was analyzed in their study. The primary impediment to the development of automated learning in this field is the lack of available data.
However, in recent years, various artificial intelligence (AI) methods have been used to establish the relationship between energy input, energy production, and greenhouse gas (GHG) emissions for various products [24]. To this end, various known artificial intelligence methods for modeling and forecasting energy production and GHG emissions include fuzzy inference systems, adaptive neuro-fuzzy systems, genetic algorithms (GAs), and artificial neural networks (ANNs) [25,26]. Fuzzy logic is used to handle the fundamental concept of partial truth, where truth values can vary between completely true and completely false [27]. A genetic algorithm (GA) is an evolutionary heuristic search algorithm that takes into account natural selection and genetic science based on Charles Darwin’s theory of natural evolution [28]. An artificial neural network (ANN) is one of the AI approaches that solves a variety of complex problems compared to old computing methods [29]. A recent artificial intelligence method is the adaptive neuro-fuzzy inference system (ANFIS), which allows learning and adaptation by integrating fuzzy logic with artificial neural networks (ANNs) [30,31].
Although artificial intelligence offers numerous advantages and is of crucial importance in addressing climate change, there has been no exhaustive research on its impact on emissions in Europe to date. Therefore, the objective of this study is to analyze and interpret the role played by artificial intelligence in addressing the issue of emissions, which have a significant impact on climate change.
The study’s novelty arises primarily from its innovative integration of OLS and ENET regression techniques to assess the influence of artificial intelligence on greenhouse gas (GHG) emissions. This advanced analytical approach provides a more nuanced understanding of AI’s impact on GHG reduction, surpassing the limitations of traditional regression methods. Moreover, the research’s originality is underscored by its exploration of a previously unexamined sample concerning the study’s objectives. By addressing this gap in the literature, the study not only offers valuable insights, but also sets a benchmark for comparative analysis. This not only enhances our understanding of AI’s impact on GHG emissions within the studied region, but also provides broader insights that are applicable to various geographical contexts.
The paper’s structure commences with an introduction that highlights the timeliness, significance, and necessity of studying the proposed topic, followed by a brief overview of the current state of knowledge. Section 3 presents the data used in the analysis and the proposed model for assessing the impact of artificial intelligence on GHG emissions. This is followed by the results and discussion section. The paper concludes with a section on the conclusions, emphasizing aspects related to the political and practical implications, research limitations, and future research directions.

2. Literature Review

The relationship between artificial intelligence and climate change has been attributed with various degrees of importance, with scientists examining the implications of artificial intelligence from different perspectives. For example, Cowls et al. [19] explore the role that artificial intelligence (AI) plays as a technology for combating global climate change and the factors driving greenhouse gas (GHG) emissions. The study results suggest that harnessing the benefits offered by AI for addressing climate change, while minimizing the associated risks, requires responsive, evidence-based, and efficient government leadership to become an effective strategy.
Akhshik et al. [22] have shown that despite limitations and limited input data, machine learning can be beneficial in predicting gas emissions, can make predictions with an acceptable level of accuracy, and will help shape future emissions research. It has also been argued by Mardani et al. [32] and Wei et al. [33] that AI can aid in predicting carbon emissions based on current trends.
A study conducted by Rolnick et al. [34] identified 37 use cases across 13 domains where AI can be applied with high impact in the fight against climate change. Yavari et al. [35] also identified a significant role of artificial intelligence in measuring the actual state of emissions in regard to complex logistic activities. Thus, authors advocate for the use of the Internet of Things (IoT) and AI for real-time reporting of GHG emissions, while accurately anticipating CO2 levels.
For example, a 2018 Microsoft/PwC report estimated that the use of artificial intelligence could have environmental implications, reducing greenhouse gas emissions by 1.5 to 4% by 2030 (Microsoft 2018). There are studies that have investigated CO2 emissions in specific sectors (agriculture, telecommunications, architecture), using artificial intelligence methods. According to recent study by Olawade et al. [36], artificial intelligence plays a crucial role in optimizing energy systems, improving climate modeling and forecasting, and enhancing sustainability in fields as diverse as transportation, agriculture, and waste management. Artificial intelligence also facilitates effective monitoring of emissions, thus highlighting its ability to significantly contribute to building a sustainable, zero-emissions future.
On the other hand, Akhshik et al. [37] points out that AI research has a significant carbon footprint. They call attention to the need to gather more evidence that clarifies the balance between greenhouse gas emissions generated by AI research activities and the potential benefits related to energy efficiency and resource use that artificial intelligence can bring. Thus, according to these findings, it is essential to investigate this relationship, because the current context is ambiguous, and the results obtained are often contradictory.
Kamyab et al. [38] examined greenhouse gas emissions in the agriculture industry, revealing the complex range of obstacles facing this sector in regard to its emission reduction pursuits and also investigated new approaches to address them. The authors concluded that the adoption of sophisticated technologies, including carbon capture techniques and AI-based monitoring, is a cornerstone of numerous emission reduction strategies. Progress in terms of emission reduction strategies and the impact of technology, such as artificial intelligence (AI), on mitigating these issues in agriculture have also been identified by [39].
Due to agriculture being a vast domain with significant gas emissions, researchers have focused on studying emissions released by crop categories, applying a type of AI technique to see which type of crop emits the most gases. For example, Nabavi-Pelesaraei et al. [40] used artificial neural networks (ANNs) to assess energy consumption and greenhouse gas emissions in watermelon production in the Guilan province of Iran. The main sources of GHG emissions identified were nitrogen (54.23%), diesel (16.73%), and electricity (15.45%). The ANN model with a 11–10–2 structure yielded the best results, with correlation coefficients of 0.969 and 0.995 for yield and GHG emissions. A year later, Hosseinzadeh-Bandbafha et al. [23] performed another comparative study to evaluate the effectiveness of different artificial intelligence methods, including adaptive neuro-fuzzy inference systems and artificial neural networks (ANNs), in modeling and predicting energy production and greenhouse gas emissions at calf fattening farms in the cities of Abyek and Alborz in Iran. The results of the comparison indicate that neuro-fuzzy inference systems, by applying fuzzy rules, were more accurate in modeling energy production and evening gas emissions than the ANN model. Additionally, Khoshnevisan et al. [41] and Nabavi-Pelesaraei et al. [42] utilized ANN technology to predict gas emissions from wheat and kiwi cultivation. In a study conducted at the level of EU member states, it was found that the use of broadband digital techniques necessary for advancing AI has led, on the one hand, to an increase in renewable energy consumption and, on the other hand, to a reduction in greenhouse gas emissions [43].
Bonire et al. [44] conducted a study to evaluate and manage pollutant gas emissions at telecommunication base stations, using a combination of artificial neural networks (ANNs) and an Internet of Things (IoT) system. According to the results, the proposed ANN model accurately anticipated pollutant gas emissions, presenting a global correlation coefficient of 0.93719. The implementation of an IoT system was also suggested to reduce these emissions, thus providing practical solutions for greenhouse gas mitigation.
A recent study [45] monitors gas emissions from urban traffic, applying artificial intelligence and the IoT, aiming to reduce greenhouse gas emissions in urban environments. Buragohain and Mahanta [25] evaluated artificial intelligence methods (adaptive neuro-fuzzy inference systems and artificial neural networks, ANNs) for modeling and predicting energy production and greenhouse gas emissions from farms in Iran, and the results showed that both models have benefits, but due to the use of fuzzy rules, adaptive neuro-fuzzy inference systems could model energy production and greenhouse gas emissions more accurately than the ANN model.
Chang et al. [46] applied machine learning methods to identify factories producing harmful emissions at high levels and deviating from usual emission patterns. Detecting these environmental violations is crucial for promoting sustainability in long-term supply chain management.
By adopting a comprehensive AI-based method for measurement and verification (M&V) protocols in energy-efficient infrastructure, Moraliyage et al. [47] confirmed the effectiveness and contribution of the proposed model. This robust and explainable framework for energy efficient building infrastructure and achieving net zero carbon emissions underlines the positive impact of artificial intelligence in this area. The relevance of using artificial intelligence in estimating carbon dioxide emissions has also been supported by [48,49]. Qin and Gong [49], through the application of decision trees and random forest algorithms, identified key factors influencing carbon dioxide emissions, such as socioeconomic factors, humidity, average temperature, and precipitation. Using the same method, the research highlighted the main determinants of carbon dioxide emissions.
Thus, through an exhaustive analysis of the literature regarding the use of artificial intelligence in reducing greenhouse gas emissions, we can conclude that these technologies have significant potential in contributing to emission reduction. This can bring multiple benefits, including protecting the environment, improving air quality, and promoting a more sustainable future.
On one hand, the review of the literature suggests that this relationship has been explored in certain sectors or branches, such as agriculture, transportation, and telecommunications. However, there is not much research examining this relationship as a whole, nor at the level of country groups. Despite these gaps, as noted by Badareu et al. [50], these shortcomings in the current research in the field of new interest open many opportunities for exploring new knowledge and identifying solutions that can improve the quality of life.
Previous studies have focused more on individual sectors or specific countries, and their results are not always generalizable globally or to a larger number of countries. Thus, it is important to conduct a study that analyzes this relationship at the level of country groups to obtain a more comprehensive and generalizable understanding of the impact of artificial intelligence on greenhouse gas emissions.
Investigating this relationship on a broader scale could bring new perspectives and help identify more efficient strategies and solutions for reducing greenhouse gas emissions.

3. Materials and Methods

3.1. Data and Variables

To investigate the effect of artificial intelligence on environmental performance, we utilized a series of selected and processed indicators from various sources, focusing on European countries. During the analysis, we utilized yearly frequency data from 2012 to 2022 for all 44 European states. This comprehensive dataset allowed us to conduct a thorough examination of trends and patterns over the specified timeframe, providing valuable insights into the dynamics of the variables under investigation across the entire European region. Thus, in order to characterize environmental performance, we used greenhouse gas (GHG) emissions as the dependent variable, expressed as per capita greenhouse gas emissions in CO2 equivalent, with the data being collected from the Eurostat database [51]. As explanatory variables from the AI domain, we used the number of annual installations of industrial robots (ROBOTS) reported by the International Federation of Robotics [52], the number of companies in AI (COMPANIES), the total investment in artificial intelligence (TOTAL_INV) expressed in billions of US dollars, and the annual private investment in artificial intelligence (PRIVATE_INV), which includes companies that received more than USD 1.5 million in investment, also expressed in US dollars, adjusted for inflation, taken from the Artificial Intelligence Index Report [53]. Industrial robots can be considered a representative indicator of the broader field of AI. While industrial robots embody many principles and applications of AI, they represent a subset of the entire AI domain, which is multifaceted and extends beyond robotics. Many modern robots use AI to improve their functionality. For example, AI algorithms can enable robots to recognize objects, navigate in environments, understand and respond to human speech, and learn from experiences. Autonomous robots leverage AI to make decisions without human intervention. Examples include self-driving cars, drones, and industrial robots on manufacturing lines. Robots, like personal assistants (e.g., robots used in homes or healthcare), use AI to interact with humans in a more natural and effective manner. The number of AI companies can indicate the size and growth of the AI industry. A larger industry typically has a more significant environmental footprint, both in terms of energy consumption and the resources required for operations and manufacturing. A higher number of AI companies can spur innovation, leading to the development of more efficient AI algorithms and hardware, which could reduce the overall energy consumption and GHG emissions of AI operations. AI companies often rely on large data centers for training and running AI models, which consume significant amounts of energy. More companies mean more data centers and computational resources, impacting GHG emissions.
Using AI components in analysis can provide a more detailed and granular understanding of how artificial intelligence technologies influence the outcomes or variables of interest. An AI index can be useful for synthesizing the complexity of different AI components into a single measure, but this can lead to the loss of specific details and subtleties behind the observed effects. By analyzing AI components, researchers can better identify which aspects or functionalities of artificial intelligence have the greatest impact on the studied outcomes and can provide a better understanding of the underlying mechanisms. Additionally, analyzing AI components can allow the identification of possible differences or variations in the effects of different AI components, which may be important for developing appropriate policies or strategies.
The plot on the left axis combines the kernel density and a regression line (red line) to illustrate the connection between the dependent and independent variables. The kernel density (blue area) portrays the probability distribution of the explanatory variables, while the regression line indicates data trends. This visual representation helps to assess the variable distribution (blue circles) and potential correlations, as indicated by the regression line. The downward trend illustrated in Figure 1 reveals a significant correlation between greenhouse gas emissions and several key factors related to AI investment levels, operational industrial robots, and the presence of the AI industry. As depicted, there is a noticeable decrease in greenhouse gas emissions concurrent with fluctuations in AI investment levels, operational industrial robots, and the presence of the AI industry. This suggests a potential relationship between advancements in AI technology and reductions in greenhouse gas emissions, indicating the potential for AI-driven industries to contribute to environmental sustainability efforts. Further analysis of these trends could provide valuable insights into the impact of AI adoption on environmental outcomes and inform strategic decision-making in policy and industry initiatives aimed at promoting sustainability.
Analyzing descriptive statistics plays a crucial role in comprehending the dataset utilized in the model. It offers insights into the distribution of data, detects outliers, and elucidates connections among variables. This analysis aids in pinpointing pertinent information and validating hypotheses. Through thorough data summarization and exploration, descriptive statistics analysis forms the foundation for decision-making and subsequent statistical inquiries.
The Jarque–Bera test assesses the normality of the data distribution for each variable in Table 1. This statistical test is utilized to determine if the skewness and kurtosis of the dataset match those of a normal distribution. A low p-value (below a chosen significance level, often 0.05) indicates that the data significantly deviate from a normal distribution. These results suggest that, except for TOTAL_INV and PRIVATE_INV, the other variables do not significantly deviate from a normal distribution at the conventional significance level of 0.05.

3.2. Elastic Net Regression Model

The methodology employed in this study is meticulously structured to ensure comprehensive and reliable data analysis, as depicted in Figure 2. The process begins with assessing the stationarity of the dataset, a crucial step in the analysis. To this end, augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) unit root tests are applied. The ADF test is conducted by estimating the following regression:
y t = α + β t + γ y t 1 + δ 1 y t 1 + + δ p y t p + ε t
where Δyt is the first difference of yt, t is a time trend, and εt is the error term. The null hypothesis H0 of the test is that γ = 0 (indicating a unit root), and if rejected, it indicates stationarity.
The PP test uses the following model:
y t = α + β t + γ y t 1 + ε t
The PP test corrects for any serial correlation and heteroscedasticity in the errors εt. These tests help determine whether the time-series data are non-stationary, which is essential for validating its appropriateness for further statistical procedures. Stationarity indicates that the statistical properties of the data, such as the mean and variance, remain constant over time, which is a prerequisite for many analytical models.
Following the confirmation of data stationarity, the next stage involves examining autocorrelation within the dataset. This is achieved through correlation matrix analysis, which reveals the degree of correlation between different time points. Understanding the autocorrelation structure helps in identifying any underlying patterns or dependencies in the data, which can significantly influence the model’s accuracy and predictive power.
To ensure the robustness of subsequent regression analyses, the methodology includes rigorous checks for multicollinearity, heteroscedasticity, and the distribution of residuals. Multicollinearity is examined to identify and mitigate any high correlations between independent variables, which could distort the regression coefficients and weaken the model. Heteroscedasticity checks are conducted to verify that the variance of residuals remains constant across all levels of the independent variables, ensuring that the model assumptions hold true. Additionally, assessing the residual distribution helps confirm that the residuals are normally distributed, which is crucial for the validity of the regression results.
The initial model parameters are estimated using ordinary least squares (OLS) regression. OLS provides a straightforward and reliable method for estimating the relationships between variables, serving as a foundation for more advanced modeling techniques. Building on the OLS results, the methodology then employs Elastic Net (ENET) regression to refine the model. ENET combines the penalties of Lasso and Ridge regressions, offering a balanced approach to handling both variable selection and regularization, which helps in improving the model’s generalizability and performance.
Within the ENET framework, optimal model selection is a critical step. This is achieved through a thorough evaluation of how the model coefficients evolve across various regularization parameters. Graphical representations of these changes provide valuable insights into the behavior and stability of the model, facilitating informed decision-making about the best parameter settings. Additionally, cross-validation errors are graphically represented to assess the predictive performance of the ENET model. Cross-validation ensures that the model is not overfitting and can generalize well to unseen data, thereby enhancing the reliability and robustness of the final analytical outcomes.
Since its inception, ENET regression has gained significant popularity in statistics, machine learning, and various other disciplines, due to its capacity to handle high-dimensional data and mitigate overfitting. It incorporates a penalty term into the ordinary least squares (OLS) objective function, encouraging less important variable coefficients to be precisely zero. Consequently, this results in sparse models, where only a subset of predictors is selected, while others are effectively disregarded.
Mathematically, the objective function of the ENET estimator [54] can be expressed as:
J = 1 2 m i = 1 m y i β 0 j = 1 p x i j β j 2 + λ ( 1 α ) 2 j = 1 p β j 2 + α j = 1 p β j
where m represents the number of samples, p signifies the number of predictors, yi denotes the target variable for the ith sample, xij indicates the value of the jth predictor for the ith sample, βj stands for the coefficient associated with the jth predictor, β0 represents the intercept term, and λ is the regularization parameter regulating the strength of the penalty term. As λ increases, more coefficients tend towards zero, resulting in a more straightforward model with fewer predictors.

4. Results and Discussion

Prior to conducting an Elastic Net regression analysis, it is beneficial and essential to conduct several preliminary tests and analyses to ascertain the adequacy of the data preparation and model readiness. Initial tests will involve assessing the stationarity of the time-series data using the augmented Dickey–Fuller (ADF) [55] and Phillips–Perron (PP) [56] tests. These tests evaluate the null hypothesis that the time-series data possesses as a unit root.
The results of the ADF and PP tests, shown in Table 2, provide valuable insights into the stationarity of the time-series data for the variables GHG, ROBOTS, COMPANIES, TOTAL_INV, and PRIVATE_INV. Overall, these tests indicate varying levels of evidence against the null hypothesis of a unit root, suggesting differing degrees of stationarity among the variables. For GHG, while both tests suggest some evidence against stationarity, the significance level is borderline according to the ADF test. ROBOTS exhibits evidence against stationarity, with both tests providing significant results, particularly the PP test. COMPANIES and PRIVATE_INV display strong evidence against stationarity, with both tests yielding highly significant results, indicating a clear rejection of the null hypothesis. TOTAL_INV shows evidence against stationarity, with the PP test providing a more significant result compared to the ADF test.
Additionally, it is important to examine the data for autocorrelation. Table 3 displays the outcomes of the correlation matrix for the variables within the model. These outcomes depict the covariance analysis as correlation coefficients among the various variables. According to the results provided, there is no evidence of autocorrelation among the variables.
Conducting ordinary least squares (OLS) regression prior to employing the Elastic Net regression enables diagnostic assessments of the data and the evaluation of linear regression assumptions. These checks may involve examining the multicollinearity, heteroscedasticity, and scrutinizing residual distributions. OLS regression aids in identifying pertinent predictors for incorporation into the Elastic Net model. Techniques such as stepwise selection or feature importance ranking based on OLS coefficients are utilized to pinpoint significant variables before regularization is applied.
The results from the ordinary least squares (OLS) estimation reveal valuable insights into the relationship between predictor variables and the dependent variable, GHG emissions (Table 4). Each coefficient estimate signifies the strength and direction of this relationship, while the associated t-statistics and probabilities assess the statistical significance of these estimates. The high R-squared value of 0.917 indicates that the model explains approximately 91.7% of the variance in the GHG emissions, suggesting strong explanatory power.
The coefficient of 5.62 × 10−05 with a standard error of 1.17 × 10−05 suggests that for every unit increase in the number of robots, there is a statistically significant increase of approximately 5.62 × 10−05 units in GHG emissions, indicating that automation processes involving robots might contribute to higher energy consumption or emissions. On the other hand, with a coefficient of −0.011093 and a standard error of 0.002528, the negative relationship for COMPANIES implies that as the number of companies increases by one unit, GHG emissions decrease by approximately 0.011093 units, suggesting that larger corporations might have more efficient processes or environmental policies to mitigate emissions compared to smaller companies. The coefficient of 1.416516 with a standard error of 0.519320 for TOTAL_INV suggests that a one unit increase in the total investment is associated with a statistically significant increase of approximately 1.416516 units in GHG emissions, indicating that economic activities spurred by increased investment might lead to higher emissions, possibly due to increased production or consumption. However, although the coefficient for PRIVATE_INV is negative (−1.38 × 10−09), the relationship with GHG emissions is not statistically significant at the conventional level (p = 0.0616), suggesting that higher levels of private investment may be associated with slightly lower GHG emissions, but further investigation is needed to confirm this relationship.
The Elastic Net (ENET) analysis presents a robust framework that effectively tackles several drawbacks encountered in ordinary least squares (OLS) regression, such as multicollinearity and overfitting. It achieves a harmonious equilibrium between bias and variance, thereby enhancing the predictive accuracy and interpretability across various real-world contexts. The outcomes derived from the ENET estimation are outlined in Table 5.
The Elastic Net regularization method was utilized with an alpha value set to 0.5, and the resulting lambda value at the minimum error is 6.839 × 10−05. This lambda value serves as a determinant of the degree of regularization applied to the model, where higher values signify stronger regularization. Additionally, the regressors underwent transformation using the standard deviation of the population. Cross-validation was conducted employing the K-fold method with five folds, incorporating a random number generator with a seed of 930,002,577, facilitating the evaluation of the model’s performance and its generalization ability. The mean squared error served as the selection measure during the cross-validation process, aiming to minimize prediction errors and enhance model accuracy. Regarding the coefficients, for the variable ROBOTS, the coefficient is 4.98 × 10−05, indicating that an increase in the number of robots corresponds to a rise in GHG emissions. Conversely, the coefficient for COMPANIES is −0.011453, suggesting that an increase in the number of companies is associated with a decrease in GHG emissions. TOTAL_INV exhibits a coefficient of 0.331522, indicating a positive relationship with GHG emissions, implying that a higher total investment leads to increased GHG emissions. PRIVATE_INV, however, displays a notably small coefficient (−2.61 × 10−10), suggesting a negligible impact on GHG emissions. Furthermore, the model explains 82.63% of the variance in GHG emissions, signifying a satisfactory fit to the data.
A comparison between the results obtained from the OLS regression and Elastic Net regularization highlights significant disparities in the coefficient estimates for predictors related to the GHG variable. In the OLS model, the coefficient estimates for predictors such as ROBOTS, COMPANIES, and TOTAL_INV demonstrate a level of consistency that aligns with expectations, suggesting their respective impacts on GHG emissions. However, upon closer examination of the Elastic Net regularization model, while the coefficients for ROBOTS, COMPANIES, and TOTAL_INV exhibit similar magnitudes, there are noticeable discrepancies in the coefficients attributed to PRIVATE_INV. Although the coefficients for PRIVATE_INV in both models hover around zero, the Elastic Net regularization model produces considerably smaller standard errors, indicative of enhanced precision in the estimation. This discrepancy underscores the importance of utilizing robust regularization techniques like Elastic Net to refine the coefficient estimates and improve the overall accuracy of the model, particularly in cases where the predictor variables exhibit multicollinearity, or when dealing with high-dimensional datasets. Additionally, the coefficient for the intercept term (C) in the Elastic Net regularization model differs from that in the OLS model, showcasing the regularization effect on the model’s intercept. Overall, these comparisons underscore the impact of regularization techniques like Elastic Net on coefficient estimates and model robustness, particularly in scenarios where predictors have varying levels of influence on the dependent variable.
The results obtained through ENET can provide a more stable and precise estimation of the model coefficients, and the embedded regularization can help reduce the variation of the estimators and enable the automatic selection of relevant variables. Therefore, while OLS may be simpler and easier to interpret in some cases, ENET can provide more robust estimates and better handle certain specific data problems, making it more relevant in certain analysis contexts.
Table 6 exhibits the lambda progression in the primary column, alongside the model’s degree of freedom, delineated in the subsequent column. The ensuing column presents the L1 norm of the coefficients, while the ultimate column denotes the R-squared value, offering insights into the model’s explanatory power.
The trajectory of the ENET analysis summarily delineates the transformation of the model coefficients across varying regularization parameters. This depiction offers discernment into the coefficient adjustments concerning the intensity of regularization, elucidating the balance between model intricacy and efficacy. Such a visual exposition facilitates the identification of the optimal regularization parameters and the discernment of the pivotal variables that substantiate the model’s predictive prowess.
In Figure 3, we delve into the evolution of the coefficients in relation to the lambda penalty. As expected, heightened penalization leads to a reduction in the model’s complexity, resulting in the coefficients gradually converging towards zero. Through cross-validation, the model at +1 SE (lambda = 0.09694) is chosen, aligning with the juncture where the majority of coefficients have been eliminated from the model.
Following this, we introduce a graphical depiction of the cross-validation errors. Figure 4 portrays the lambda path on the x-axis, juxtaposed with the mean error metrics of both the training and test sets on the y-axis. As expected, the training error persistently maintains a lower level than the test error, implying superior model performance on the training data compared to unseen data. This observation aligns with the customary behavior anticipated during model evaluation and validation procedures.
The results presented in our study provide a different perspective from the findings of other researchers, who have highlighted the significant benefits of using robots in reducing carbon dioxide emissions [48,49]. In contrast, research by Akhshik et al. [22], Mardani et al. [32], and Wei et al. [33] do not identify a direct impact of AI on emissions reduction, but rather a contribution to carbon emissions prediction, thus facilitating future modeling and research in this area. Also, Yavari et al. [35] highlighted an important role of AI in accurately measuring emissions in regard to complex logistics activities, arguing in favor of using the Internet of Things (IoT) and AI for real-time monitoring of GHG emissions and for an accurate estimate of CO2 levels. The results identified in this study contradict the findings of other researchers in the field who have identified a significant impact of robot usage in reducing carbon dioxide emissions [57,58]. These divergences in findings are also corroborated by the research conducted by Li et al. [59], who explored the impact of robot use in various sectors, noting that the effects on greenhouse gas emissions are variable. In some cases, the introduction of robots can even increase emissions, by increasing the energy consumption required for their production and operation. Thus, while some studies highlight the potential benefits of advanced technologies, including artificial intelligence and robots, in reducing greenhouse gas emissions, our research and that of other authors [59] suggests that the impact may vary significantly, depending on the specific context and way these technologies are implemented. This variety in the results underscores the need for more in-depth and tailored analysis to optimize the use of advanced technologies in carbon reduction efforts. In addition, our findings add a new dimension to this debate by identifying the impact of larger companies and significant investments that could contribute to lower emissions. This correlation has not yet been extensively explored in the literature, with studies specifically investigating the relationship between corporate size and the effectiveness of technology investment in reducing emissions lacking. Thus, our research proposes a new avenue of investigation, suggesting that larger resources and long-term strategic commitments could play a decisive role in optimizing the impact of green technologies. Therefore, it is essential to continue research in this area, to clarify these dynamics and to optimize the implementation strategies for emerging technologies in order to maximize their ecological benefits. This integrated approach could provide more effective solutions for reducing global greenhouse gas emissions and promoting a sustainable future.

5. Conclusions and Policy Implications

These findings suggest that the deployment of industrial robots and investment in AI technologies may have a significant impact on GHG emissions, with larger companies and higher investment levels potentially contributing to reduced emissions. However, the relationship between private investment in AI and GHG emissions requires further scrutiny. These conclusions provide valuable insights for policymakers and stakeholders aiming to develop strategies for mitigating GHG emissions, while harnessing the potential of AI technologies for sustainable development.
This study’s findings suggest various policy recommendations to mitigate the impact of predictors on greenhouse gas (GHG) emissions and to foster sustainable development. Governments can promote the adoption of green technologies and sustainable practices in industries heavily reliant on automation and AI through incentives, such as subsidies or tax breaks for companies investing in energy-efficient machinery and processes. Additionally, funding research and development endeavors geared towards environmentally friendly technologies can further accelerate progress in this domain. Stricter environmental regulations and standards for AI and automation industries are also crucial, encompassing measures like setting GHG emission limits, enforcing energy efficiency standards, and encouraging the use of renewable energy sources where feasible.
Moreover, investments in research and development initiatives aimed at enhancing the environmental sustainability of AI technologies should be prioritized, focusing on projects dedicated to developing cleaner energy sources, improving energy efficiency in AI systems, and exploring innovative solutions for emission reduction. Education and training programs for workers in affected industries can facilitate the adoption of sustainable practices, including training on energy-efficient technologies and raising awareness about the environmental impact of industrial processes. Collaboration through public–private partnerships is essential, leveraging the expertise and resources of diverse stakeholders to effectively address environmental challenges, drive innovation, and promote sustainable development. Implementing carbon pricing mechanisms and fostering international cooperation are also imperative steps towards incentivizing emissions reduction and achieving common goals in combating climate change on a global scale.
By implementing these policy recommendations, governments, businesses, and other stakeholders can effectively address the impact of AI on GHG emissions, while harnessing the potential of AI and automation for sustainable development. These measures can contribute to the transition towards a low-carbon economy and help mitigate the adverse effects of climate change.

Author Contributions

Conceptualization, N.M.D. and G.B.; methodology, N.M.D.; software, M.N.; validation, A.L.S., M.E. and M.D.D.; formal analysis, M.D.D.; investigation, M.N.; resources, N.M.D. and G.B.; data curation, M.E.; writing—original draft preparation, A.L.S.; writing—review and editing, G.B.; visualization, M.D.D.; supervision, M.E.; project administration, M.N. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Informed Consent Statement

Not applicable.

Data Availability Statement

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.

Acknowledgments

This work was supported by a grant from the Romanian Ministry of Research, Innovation and Digitalization, the project had the title “Economics and Policy Options for Climate Change Risk and Global Environmental Governance” (CF 193/28.11.2022, Funding Contract no. 760078/23.05.2023), within Romania’s National Recovery and Resilience Plan (PNRR)–Pillar III, Component C9, Investment I8 (PNRR/2022/C9/MCID/I8)–Development of a program to attract highly specialized human resources from abroad in research, development, and innovation activities.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Huntingford, C.; Jeffers, E.S.; Bonsall, M.B.; Christensen, H.M.; Lees, T.; Yang, H. Machine learning and artificial intelligence to aid climate change research and preparedness. Environ. Res. Lett. 2019, 14, 124007. [Google Scholar] [CrossRef]
  2. Appiah, M.; Li, F.; Korankye, B. Modeling the linkages among CO2 emission, energy consumption, and industrialization in sub-Saharan African (SSA) countries. Environ. Sci. Pollut. Res. 2021, 29, 38506–38521. [Google Scholar] [CrossRef] [PubMed]
  3. Kosovac, A.; Medić, A.; Begović, M. Machine Learning Modeling for Reducing Greenhouse Gas Emissions in Urban Areas; University of Sarajevo, Faculty of Traffic and Communications: Sarajevo, Bosnia and Herzegovina, 2023; pp. 131–136. [Google Scholar] [CrossRef]
  4. Xu, Z.; Liu, L.; Wu, L. Forecasting the carbon dioxide emissions in 53 countries and regions using a non-equigap grey model. Environ. Sci. Pollut. Res. 2021, 28, 15659–15672. [Google Scholar] [CrossRef] [PubMed]
  5. Manta, A.G.; Doran, N.M.; Bădîrcea, R.M.; Badareu, G.; Țăran, A.-M. Does the implementation of a Pigouvian tax be considered an effective approach to address climate change mitigation? Econ. Anal. Policy 2023, 80, 1719–1731. [Google Scholar] [CrossRef]
  6. Dima, B.; Dima, S.M.; Lobont, O.R. New empirical evidence of the linkages between governance and economic output in the European Union. J. Econ. Policy Reform. 2013, 16, 68–89. [Google Scholar] [CrossRef]
  7. Dima, B.; Lobonţ, O.R.; Moldovan, N.C. Does the Quality of Public Policies and Institutions Matter for Entrepreneurial Activity? Evidences from the European Union’s Member States. Panoeconomicus 2016, 63, 425–439. [Google Scholar] [CrossRef]
  8. European Parliament. Renewable Energy: Setting Ambitious Targets for EUROPE. 2017. Available online: https://www.europarl.europa.eu/topics/en/article/20171124STO88813/renewable-energy-setting-ambitious-targets-for-europe (accessed on 15 April 2024).
  9. Council of the European Union. Paris Agreement on Climate Change. Available online: https://www.consilium.europa.eu/ro/policies/climate-change/paris-agreement/ (accessed on 15 April 2024).
  10. Hunt, E.B. Artificial Intelligence; Academic Press: Vienna, Austria, 2014. [Google Scholar]
  11. Fetzer, J.H.; Fetzer, J.H. What Is Artificial Intelligence? Springer: Dordrecht, The Netherlands, 1990; pp. 3–27. [Google Scholar]
  12. Oke, S.A. A literature review on artificial intelligence. Int. J. Inf. Manag. Sci. 2008, 19, 535–570. [Google Scholar]
  13. Jackson, P.C. Introduction to Artificial Intelligence; Courier Dover Publications: Mineola, NY, USA, 2019. [Google Scholar]
  14. Rao, A.S.S.; Rao, C.R.; Krantz, S. Artificial Intelligence; Elsevier: Amsterdam, The Netherlands, 2023. [Google Scholar]
  15. Zhang, B.; Zhu, J.; Su, H. Toward the third generation artificial intelligence. Sci. China Inf. Sci. 2023, 66, 121101. [Google Scholar] [CrossRef]
  16. Qin, M.; Su, C.W.; Qiu, L.; Lobonţ, O.R. Are there digital tech bubbles in China? Technol. Econ. Dev. Econ. 2023, 30, 1–24. [Google Scholar] [CrossRef]
  17. Wang, H.; Fu, T.; Du, Y.; Gao, W.; Huang, K.; Liu, Z.; Chandak, P.; Liu, S.; Van Katwyk, P.; Deac, A.; et al. Scientific discovery in the age of artificial intelligence. Nature 2023, 620, 47–60. [Google Scholar] [CrossRef] [PubMed]
  18. Manta, A.G.; Bădîrcea, R.M.; Doran, N.M.; Badareu, G.; Gherțescu, C.; Popescu, J. Industry 4.0 Transformation: Analysing the Impact of Artificial Intelligence on the Banking Sector through Bibliometric Trends. Electronics 2024, 13, 1693. [Google Scholar] [CrossRef]
  19. Cowls, J.; Tsamados, A.; Taddeo, M.; Floridi, L. The AI gambit: Leveraging artificial intelligence to combat climate change—Opportunities, challenges, and recommendations. AI Soc. 2023, 38, 283–307. [Google Scholar] [CrossRef] [PubMed]
  20. Buckland, C.E.; Bailey, R.M.; Thomas, D.S.G. Using artificial neural networks to predict future dryland responses to human and climate disturbances. Sci. Rep. 2019, 9, 3855. [Google Scholar] [CrossRef] [PubMed]
  21. Coyle, E.D.; Simmons, R.A. Understanding the Global Energy Crisis; Purdue University Press: West Lafayette, IN, USA, 2014. [Google Scholar]
  22. Akhshik, M.; Bilton, A.; Tjong, J.; Singh, C.V.; Faruk, O.; Sain, M. Prediction of greenhouse gas emissions reductions via machine learning algorithms: Toward an artificial intelligence-based life cycle assessment for automotive lightweighting. Sust Mat. Technol. 2022, 31, e00370. [Google Scholar] [CrossRef]
  23. Hosseinzadeh-Bandbafha, H.; Nabavi-Pelesaraei, A.; Shamshirband, S. Investigations of energy consumption and greenhouse gas emissions of fattening farms using artificial intelligence methods. JA Environ. Prog. Sustain. Energy 2017, 36, 1546–1559. [Google Scholar] [CrossRef]
  24. Buragohain, M.; Mahanta, C. A novel approach for ANFIS modelling based on full factorial design. Appl. Soft Comput. 2008, 8, 609–625. [Google Scholar] [CrossRef]
  25. Uthayakumar, H.; Radhakrishnan, P.; Shanmugam, K.; Kushwaha, O.S. Growth of MWCNTs from Azadirachta indica oil for optimization of chromium (VI) removal efficiency using machine learning approach. Environ. Sci. Pollut. Res. 2022, 29, 34841–34860. [Google Scholar] [CrossRef]
  26. Kushwaha, O.S.; Uthayakumar, H.; Kumaresan, K. Modeling of carbon dioxide fixation by microalgae using hybrid artificial intelligence (AI) and fuzzy logic (FL) methods and optimization by genetic algorithm (GA). Environ. Sci. Pollut. Res. 2023, 30, 24927–24948. [Google Scholar] [CrossRef]
  27. Shahlaei, M.; Madadkar-Sobhani, A.; Saghaie, L.; Fassihi, A. Application of an expert system based on genetic algorithm–adaptive neuro-fuzzy inference system (GA–ANFIS) in QSAR of cathepsin K inhibitors. Expert. Syst. Applic. 2012, 39, 6182–6191. [Google Scholar] [CrossRef]
  28. Ceylan, H. Analysis and Design of Concrete Pavement Systems Using Artificial Neural Networks; University of Illinois at Urbana-Champaign: Ann Arbor, MI, USA, 2002. [Google Scholar]
  29. Zhou, J.; Li, C.; Arslan, C.A.; Hasanipanah, M.; Amnieh, H.B. Performance evaluation of hybrid FFA-ANFIS and GA-ANFIS models to predict particle size distribution of a muck-pile after blasting. Eng. Comput. 2019, 37, 265–274. [Google Scholar] [CrossRef]
  30. Singh, R.; Kainthola, A.; Singh, T.N. Estimation of elastic constant of rocks using an ANFIS approach. Appl. Soft. Comput. 2012, 12, 40–45. [Google Scholar] [CrossRef]
  31. Meghişan-Toma, G.M.; Puiu, S.; Florea, N.; Meghişan, F.; Bădîrcea, R.; Manta, A. Sustainable transformation of Romanian companies through industry 4.0, green production and environment commitment. Amfiteatru Econ. 2022, 24, 46–60. [Google Scholar] [CrossRef]
  32. Mardani, A.; Liao, H.; Nilashi, M.; Alrasheedi, M.; Cavallaro, F. A multi-stage method to predict carbon dioxide emissions using dimensionality reduction, clustering, and machine learning techniques. J. Clean. Prod. 2020, 275. [Google Scholar] [CrossRef]
  33. Sun, W.; Wang, Y.; Zhang, C. Forecasting CO2 emissions in Hebei, China, through moth-flame optimization based on the random forest and extreme learning machine. Environ. Sci. Pollut. Res. 2018, 25, 28985–28997. [Google Scholar] [CrossRef] [PubMed]
  34. Rolnick, D.; Donti, P.L.; Kaack, L.H.; Kochanski, K.; Lacoste, A.; Sankaran, K.; Ross, A.S.; Milojevic-Dupont, N.; Jaques, N.; Waldman-Brown, A.; et al. Tackling Climate Change with Machine Learning. 2019. Available online: http://arxiv.org/abs/1906.05433 (accessed on 15 April 2024).
  35. Yavari, A.; Mirza, I.B.; Bagha, H.; Korala, H.; Dia, H.; Scifleet, P.; Sargent, J.; Tjung, C.; Shafiei, M. ArtEMon: Artificial Intelligence and Internet of Things Powered Greenhouse Gas Sensing for Real-Time Emissions Monitoring. Sensors 2023, 23, 7971. [Google Scholar] [CrossRef] [PubMed]
  36. Olawade, D.; Wada, O.; Olawade, D.; Fapohunda, O.; Ige, A.O.; Ling, J. Artificial intelligence potential for net zero sustainability: Current evidence and prospects. Next Energy 2024, 4, 100041. [Google Scholar] [CrossRef]
  37. Doran, M.D.; Poenaru, M.M.; Zaharia, A.L.; Vătavu, S.; Lobonț, O.R. Fiscal policy, growth, financial development and renewable energy in Romania: An autoregressive distributed lag model with evidence for growth hypothesis. Energies 2022, 16, 70. [Google Scholar] [CrossRef]
  38. Kamyab, H.; Saberi-Kamarposhti, M.; Hashim, H.; Yusuf, M. Carbon dynamics in agricultural greenhouse gas emissions and removals: A comprehensive review. Carbon Lett. 2024, 34, 265–289. [Google Scholar] [CrossRef]
  39. Zhang, K.; Cui, J.; Zhou, Y.; Chen, A.; Ouyang, C.; Palocz-Andresen, M.; Lou, Z. Ghg emissions reduction patterns from waste sectors after forced source separation. Process Saf. Environ. Prot. 2023, 180, 443–450. [Google Scholar] [CrossRef]
  40. Nabavi-Pelesaraei, A.; Abdi, R.; Rafiee, S. Neural network modeling of energy use and greenhouse gas emissions of watermelon production systems. J. Saudi Soc. Agric. Sci. 2016, 15, 38–47. [Google Scholar] [CrossRef]
  41. Khoshnevisan, B.; Rafiee, S.; Omid, M.; Yousefi, M.; Movahedi, M. Modeling of energy consumption and GHG (greenhouse gas) emissions in wheat production in Esfahan province of Iran using artificial neural networks. Energy 2013, 52, 333–338. [Google Scholar] [CrossRef]
  42. Nabavi-Pelesaraei, A.; Rafiee, S.; Hosseinzadeh-Bandbafha, H.; Shamshirband, S. Modeling energy consumption and greenhouse gas emissions for kiwifruit production using artificial neural networks. J. Cleaner Prod. 2016, 133, 924–931. [Google Scholar] [CrossRef]
  43. Doran, M.D.; Puiu, S.; Berceanu, D.; Țăran, A.M.; Para, I.; Popescu, J. Combining the broadband coverage and speed to improve fiscal system efficiency in the Eastern European Union countries. Electronics 2022, 11, 3321. [Google Scholar] [CrossRef]
  44. Bonire, G.; Gbenga-Ilori, A. Towards artificial intelligence-based reduction of greenhouse gas emissions in the telecommunications industry. Sci. Afr. 2021, 12, e00823. [Google Scholar] [CrossRef]
  45. Terrada, L.; El Kha, M.; Daaif, A.; Ouajji, H. Towards a Green Supply Chain Based on Smart Urban Traffic Using Deep Learning Approach. Stat. Optim. Inf. Comput. 2022, 10, 25–44. [Google Scholar] [CrossRef]
  46. Chang, X.; Huang, Y.; Li, M.; Bo, X.; Kumar, S. Efficient Detection of Environmental Violators: A Big Data Approach. Prod. Oper. Manag. 2021, 30, 1246–1270. [Google Scholar] [CrossRef]
  47. Moraliyage, H.; Dahanayake, S.; De Silva, D.; Mills, N.; Rathnayaka, P.; Nguyen, S.; Alahakoon, D.; Jennings, A. A Robust Artificial Intelligence Approach with Explainability for Measurement and Verification of Energy Efficient Infrastructure for Net Zero Carbon Emissions. Sensors 2022, 22, 9503. [Google Scholar] [CrossRef] [PubMed]
  48. Brauer, M.; Sarath, K.G.; Nishad, K.D.; Sagnik, D.; Sachchida, N.T.; Crystal, W.; Randall, V.M. Examination of monitoring approaches for ambient air pollution: A case study for India. Atmos. Environ. 2018, 216, 116940. [Google Scholar] [CrossRef]
  49. Qin, J.; Gong, N. The estimation of the carbon dioxide emission and driving factors in China based on machine learning methods. Sustain. Prod. Consum. 2022, 33, 218–229. [Google Scholar] [CrossRef]
  50. Badareu, G.; Doran, N.M.; Puiu, S.; Bădîrcea, R.M.; Manta, A.G. Is the Relationship between Corporate Social Responsibility, Environment and Energy Sufficiently Debated Nowadays? Energies 2023, 16, 3405. [Google Scholar] [CrossRef]
  51. Eurostat, Greenhouse Gas Emission Statistics—Emission Inventories. Available online: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Greenhouse_gas_emission_statistics_-_emission_inventories (accessed on 10 April 2024).
  52. International Federation of Robotics, World Robotics Report. 2023. Available online: https://ifr.org/img/worldrobotics/2023_WR_extended_version.pdf (accessed on 23 March 2024).
  53. Maslej, N.; Fattorini, L.; Perrault, R.; Parli, V.; Reuel, A.; Brynjolfsson, E.; Etchemendy, J.; Ligett, K.; Lyons, T.; Manyika, J.; et al. The AI Index 2024 Annual Report. AI Index Steering Committee, Institute for Human-Centered AI; Stanford University: Stanford, CA, USA, 2024. [Google Scholar]
  54. Friedman, J.; Hastie, T.; Tibshirani, R. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw. 2010, 33, 1. [Google Scholar] [CrossRef] [PubMed]
  55. Dickey, D.A.; Fuller, W.A. Distribution of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 1979, 74, 427–431. [Google Scholar]
  56. Phillips, P.C.B.; Perron, P. Testing for a unit root in time series regression. Biometrika 1988, 75, 335–346. [Google Scholar] [CrossRef]
  57. Wang, J.; Wang, W.; Liu, Y.; Wu, H. Can industrial robots reduce carbon emissions? Based on the perspective of energy rebound effect and labor factor flow in China. Technol. Soc. 2023, 72, 102208. [Google Scholar] [CrossRef]
  58. Yao, W.; Liu, L.; Fujii, H.; Li, L. Digitalization and net-zero carbon: The role of industrial robots towards carbon dioxide emission reduction. J. Clean. Prod. 2024, 450, 141820. [Google Scholar] [CrossRef]
  59. Li, Y.; Zhang, Y.; Pan, A.; Han, M.; Veglianti, E. Carbon emission reduction effects of industrial robot applications: Heterogeneity characteristics and influencing mechanisms. Technol. Soc. 2022, 70, 102034. [Google Scholar] [CrossRef]
Figure 1. Trend representation of the variables within the model.
Figure 1. Trend representation of the variables within the model.
Sustainability 16 04930 g001
Figure 2. Illustration of the analytical process.
Figure 2. Illustration of the analytical process.
Sustainability 16 04930 g002
Figure 3. Dynamics of the coefficients in the ENET modeling.
Figure 3. Dynamics of the coefficients in the ENET modeling.
Sustainability 16 04930 g003
Figure 4. Visualization of cross-validation errors.
Figure 4. Visualization of cross-validation errors.
Sustainability 16 04930 g004
Table 1. Descriptive statistics of the variables.
Table 1. Descriptive statistics of the variables.
GHGROBOTSCOMPANIESTOTAL_INVPRIVATE_INV
Mean9.336262,222.22172.00003.61613.79 × 10+09
Median9.403466,000.00170.00002.43132.69 × 10+09
Maximum9.749482,000.00343.000012.50071.25 × 10+10
Minimum8.678143,000.0053.00000.52686.13 × 10+08
Std. Dev.0.300514,042.5995.70783.84893.83 × 10+09
Skewness−1.0406−0.06170.37731.47491.3982
Kurtosis3.79291.58612.07244.16023.9493
Jarque–Bera1.86010.75530.53613.76793.2705
Probability0.39450.68540.76480.15190.1948
Sum84.0263560,000.01548.0032.545643.41 × 10+10
Sum Sq. Dev.0.72271.58 × 10+0973,280.00118.51661.17 × 10+20
Table 2. Standard unit root tests.
Table 2. Standard unit root tests.
ADFPP
t-Statisticp-Valuet-Statisticp-Value
GHG−3.67020.0719−3.76460.0271
ROBOTS−3.21490.0533−5.08310.0043
COMPANIES−4.11420.0216−6.45910.0013
TOTAL_INV−3.41060.0512−4.44640.0118
PRIVATE_INV−5.09670.0097−4.95500.0065
Table 3. Correlation matrix.
Table 3. Correlation matrix.
Covariance Analysis: Ordinary
Sample: 2012–2022
Balanced sample (listwise missing value deletion)
Correlation
ProbabilityGHGROBOTSCOMPANIESTOTAL_INVPRIVATE_INV
GHG1.000000
-----
ROBOTS−0.4283371.000000
0.2500-----
COMPANIES−0.6291580.9448571.000000
0.06950.0001-----
TOTAL_INV−0.5702100.7846600.9110241.000000
0.10890.01230.0006-----
PRIVATE_INV−0.5787890.7945220.9171530.9996401.000000
0.10250.01050.00050.0000-----
Table 4. Results from the OLS estimation.
Table 4. Results from the OLS estimation.
Dependent Variable: GHG
Method: Least Squares
VariableCoefficientStd. Errort-StatisticProb.
ROBOTS5.62 × 10−051.17 × 10−054.7884010.0087
COMPANIES−0.0110930.002528−4.3880850.0118
TOTAL_INV1.4165160.5193202.7276370.0526
PRIVATE_INV−1.38 × 10−095.35 × 10−10−2.5761370.0616
C7.8513890.40905619.193920.0000
R-squared0.916951  Mean dependent variable9.336266
Adjusted R-squared0.833903  S.D. dependent var0.300580
S.E. of regression0.122501  Akaike info criterion−1.061209
Sum squared residual0.060026  Schwarz criterion−0.951640
Log likelihood9.775442  Hannan–Quinn criterion−1.297659
F-statistic11.04115  Durbin–Watson statistic1.203396
Prob (F-statistic)0.019546
Table 5. Results from the ENET estimation.
Table 5. Results from the ENET estimation.
Dependent Variable: GHG
Method: Elastic Net Regularization
Sample (adjusted): 2013–2022
Penalty type: Elastic Net (alpha = 0.5)
Lambda at minimum error: 6.839 × 10−05
Regressor transformation: Std Dev (pop)
Cross-validation method: K-fold (number of folds = 5), rng = kn, seed = 930,002,577
Selection measure: Mean Squared Error
(minimum)(+1 SE)(+2 SE)
Lambda6.839 × 10−050.096940.1694
Variable Coefficients
ROBOTS4.98 × 10−050.0000000.000000
COMPANIES−0.011453−0.001294−0.000951
TOTAL_INV0.3315220.0000000.000000
PRIVATE_INV−2.61 × 10−10−2.23 × 10−12−1.64 × 10−13
C8.0009159.5673169.500402
d.f.422
L1 Norm8.3439409.5686109.501352
R-squared0.8262580.3593250.290519
Table 6. Summary path.
Table 6. Summary path.
Lambdad.f.L1 NormR-Squared
10.35659319.3362662.00 × 10−08
20.32491419.3623900.058190
927.51 × 10−0548.3368410.825623
936.84 × 10−0548.3439400.826258
946.23 × 10−0548.3388890.826274
993.91 × 10−0548.3352060.825954
1003.57 × 10−0548.3451620.826606
Note: The grey color indicates the optimal model selected according to lambda.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Doran, N.M.; Badareu, G.; Doran, M.D.; Enescu, M.; Staicu, A.L.; Niculescu, M. Greening Automation: Policy Recommendations for Sustainable Development in AI-Driven Industries. Sustainability 2024, 16, 4930. https://doi.org/10.3390/su16124930

AMA Style

Doran NM, Badareu G, Doran MD, Enescu M, Staicu AL, Niculescu M. Greening Automation: Policy Recommendations for Sustainable Development in AI-Driven Industries. Sustainability. 2024; 16(12):4930. https://doi.org/10.3390/su16124930

Chicago/Turabian Style

Doran, Nicoleta Mihaela, Gabriela Badareu, Marius Dalian Doran, Maria Enescu, Anamaria Liliana Staicu, and Mariana Niculescu. 2024. "Greening Automation: Policy Recommendations for Sustainable Development in AI-Driven Industries" Sustainability 16, no. 12: 4930. https://doi.org/10.3390/su16124930

APA Style

Doran, N. M., Badareu, G., Doran, M. D., Enescu, M., Staicu, A. L., & Niculescu, M. (2024). Greening Automation: Policy Recommendations for Sustainable Development in AI-Driven Industries. Sustainability, 16(12), 4930. https://doi.org/10.3390/su16124930

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop