Artificial Intelligence and Urban Air Quality: The Role of Government and Public Environmental Attention

Abstract

Artificial intelligence (AI) technology not only promotes rapid economic development but also plays an irreplaceable role in improving environmental quality. Based on the quasi-natural experiment of the National Artificial Intelligence Innovation Comprehensive Experimental Zone, this paper empirically studies the effect and mechanism of AI on urban air quality (AQ) using the multi-time difference-in-difference model. The research results showed that AI improved the AQ of cities. The mechanism analysis results indicated that there was a positive mediating effect of government environmental attention on the relationship between AI and AQ improvement. Public environmental attention can further enhance the role of AI in improving urban AQ. Further analysis revealed that the improvement effect of AI on urban AQ was mainly reflected in eastern cities and non-resource-based cities. The research conclusion of this study provides reliable empirical evidence for leveraging AI to empower urban green development and assist in air pollution prevention practices.

1. Introduction

Since the reform and opening up, China’s economic growth has been remarkable, but it has also paid an environmental price. In 2018, 64% of China’s 337 cities exceeded the air pollution standards, and in 2020, the air quality surpassed the standards in 40% of cities. Currently, urban air pollution in China is seriously affecting residents’ health and social life. Moreover, China’s economic and social development has accelerated into a high-quality development stage characterized by low-carbon transformation [1]. However, the structural, root, and trending pressures faced by air pollution prevention and control have not been fundamentally alleviated, and the overall situation of air pollution prevention and control remains severe [2]. China’s high-energy consumption industries, which have long been a deep-seated stock problem that has plagued the improvement of air quality, have not been fundamentally resolved, and new incremental problems, such as motor vehicle exhaust emissions, have gradually accumulated [3]. The need to deepen air pollution prevention and control comprehensively has become more prominent and urgent. The Chinese government work report in 2024 proposed strengthening the construction of ecological civilization. In the current situation, where industrial emissions remain the largest source of carbon emissions in China, accelerating the green transformation mode is the fundamental strategy to cope with the pressure of air pollution prevention and control.

AI, supported by key technologies such as large-scale models, deep learning, and neural networks, is becoming an important engine for promoting digital industrialization in China [4]. Cultivating AI technologies that can serve as new growth engines at the technical and application levels is a realistic choice for China to catch up with the new wave of technological revolution. The Chinese government issued the National New Generation Artificial Intelligence Innovation and Development Pilot Zone (AIPZ) policy in 2019. This policy aims to build a favorable ecosystem for the development of AI, to solve major problems in AI technology and industrialization, to explore new paths for building an intelligent society, and to achieve sustainable development [5]. In 2023, the adoption rate of generative AI by enterprises increased to 15%, and the market size of the AI industry was as high as about 14.4 trillion yuan [6]. These data fully demonstrate the widespread application of AI and indicate that its integration in different industries will be deeper, and these data also provide strong technical support for promoting green transformation.

The widespread application of AI technology not only brings economic growth effects through factor substitution, efficiency improvement, and knowledge creation, but also generates employment “creation effects” and “destruction effects” through expanding labor demand and heterogeneous impacts on different positions. While AI has a series of effects on economic development, to what extent does it affect the ecological environment? Is the influence of economic growth brought about by the development of AI on the natural environment positive or negative? Is the thriving growth of AI environmentally friendly, and what influence mechanisms does it have? These issues have gradually become the focus of great attention for scholars and policymakers. Recently, the driving role of AI development in carbon reduction and green growth has been examined. These studies have indicated that the development of AI has a positive effect on promoting the green transformation of the local economy. However, the relationship between AI and urban AQ has not received much attention from scholars. Can the development of AI in China reduce the current situation of air pollution? How does AI reduce air pollution? Against the backdrop of national economic transformation and upgrade, increasingly severe air pollution is hindering the green transformation of the economy. AI is an important way to improve air quality, promote economic transformation, and actively explore solutions to regional differences in air pollution. Figure 1 shows the research approach. This paper uses 274 cities in China from 2011 to 2021 as samples and combines the staggered difference-in-difference (DID) method to empirically study the effect of AI on urban AQ and to explore possible mechanisms of influence and urban heterogeneity effects. Improving AQ is of great significance for building livable cities and enhancing people’s quality of life. Good AQ is an important indicator for measuring the quality of urban life, which can attract investment and promote economic development. Therefore, improving AQ is a key aspect of achieving sustainable development goals. This study can provide evidence and reference for society to use AI to achieve sustainable development.

The potential marginal contribution lies in the following: (1) The development of AI, government environmental attention (GEA), public environmental attention (PEA), and urban AQ are placed in the same analytical framework, and the air pollution control effects of AI development are systematically evaluated. The synergistic effects of government environmental governance and public participation are considered to provide a new perspective for a profound understanding of how AI can contribute to green development. (2) Existing scholars use industrial robots or AI patents as proxy indicators to measure the development of AI when studying its environmental effects, which may lead to endogeneity issues due to variable measurement errors. This paper takes the establishment of AIPZ policy as a quasi-natural experiment and uses a staggered DID method to assess the atmospheric governance effect of AI empirically, which can effectively alleviate endogeneity problems and obtain more reliable conclusions. (3) The heterogeneous governance effect of AI on AQ is examined based on the geographical location and resource types of cities, which can provide empirical evidence for different types of cities to promote differentiated AI practices according to local conditions and improve urban air quality.

2. Literature Review and Research Hypotheses

2.1. Literature Review

2.1.1. Related Research on Air Pollution

The research on air pollution has always been a popular topic for scholars and mostly focuses on the influencing factors of air pollution. In the early days, the emphasis was mainly on the relationship between air pollution and economic growth. With the opening up to the outside world and economic globalization, many scholars have shifted their perspective to the effect of FDI on air pollution. In recent years, some scholars have performed the analysis from environmental regulation, Internet development, digital economy, and other perspectives. Numerous scholars have demonstrated the applicability of the Environmental Kuznets Curve from different perspectives and methodological models and believe in an inverted “U”-shaped relationship between air pollution and economic growth [7,8,9,10]. However, some scholars also believe in a more complex and turbulent relationship between air pollution and economic growth, such as the “N” shape [11]. Considerable controversy has always been noted in the academic community regarding the effect of FDI on AQ, and the conclusions vary greatly due to different research perspectives and methods. In summary, two viewpoints exist: One viewpoint believes that FDI can worsen the domestic environment. In the early stages of development, developing countries often relaxed their environmental control standards to attract FDI. Multinational corporations, to avoid strict environmental regulations in their own countries, shifted to developing countries and engaged in large-scale industrial intensive production, and caused serious air pollution [12,13,14]. However, some scholars believe that multinational corporations have advanced production technologies that can drive innovation levels in their surrounding areas through spillover effects. Environmental protection technologies are more advanced and thereby improve local air pollution [15,16]. Environmental regulation is an important means for the government to control air pollution. Many studies have shown that different environmental regulation policies can have a certain degree of governance effect on air pollution, that is, environmental regulation can guide enterprises to increase environmental governance investment, thereby improving the local air quality level [17,18].

In the new round of the global technology wave, the integration of the Internet and many fields, such as finance and green technology, has an important effect on the development of the world economy. Some scholars have begun to pay attention to the role of the Internet in air pollution. The emergence of big data and Internet intelligent technology has played an important role in the regulation of air pollution, water pollution, and other environmental issues in the economic development [19,20,21]. Nyhan et al. (2016) [22] used population activity patterns to evaluate the population-weighted PM2.5 concentration levels within cities to monitor the pollution exposure of the population to the air. Wonohardjo and Kusuma (2019) [23] immediately detected the concentration of carbon monoxide in the air through the environmental supervision and mapping system integrated with Internet technology and provided an effective monitoring means for the treatment of air pollution. With the advent of the digital age, the air pollution control effect of the digital economy has also received widespread attention. Scholars have found through empirical testing that the digital economy can improve AQ by optimizing resource allocation and promoting green technological progress [24,25].

2.1.2. Related Research on AI

The economic effects of AI have received widespread attention. The effect of AI on economic growth, employment demand, technological innovation, and corporate performance has been constantly discussed. Zhao et al. (2022) [26] used China as an example to construct an econometrics model and to analyze the effect of AI on economic growth. They found that a certain level of AI can promote economic growth. Based on Chinese samples, Li and Qi (2022) [27] found that AI has a promoting effect on employment, while the pandemic has exacerbated labor polarization and has remarkable growth potential for demand for nonprogrammatic interactive skills. Ma et al. (2022) [28] studied the effect of AI development on the employment skill structure of the workforce and found that AI can change the employment skill structure. Based on Taiwan’s electronics industry enterprise samples, Yang (2022) [29] found that AI can change the labor force composition of enterprises. Some literature also discussed the effect of AI on innovation. Rammer et al. (2022) [30] found that AI technology promotes enterprise innovation, and AI technology has saved approximately 6% of annual total costs for the German business sector. Based on industrial robot data from 14 manufacturing sectors in China from 2008 to 2017, Liu et al. (2020) [31] found that AI is conducive to the improvement of technological innovation.

Some literature also explored the effect of AI on environmental pollution. Based on the manufacturing industry in China, Liang et al. (2022) [32] found that AI can enhance carbon reduction performance through the green technology innovation effect. Liu et al. (2022) [33] used data from academic papers related to industrial robots and AI in China to find that AI remarkably reduced the carbon intensity of enterprises. To alleviate endogeneity caused by variable measurement issues, existing scholars conducted a quasi-natural experiment based on the AIPZ policy and found that AI can enhance the green development level of enterprises [34,35]. Currently, no literature studies the effect of AI on urban AQ. Therefore, this paper can supplement the research content in related fields and provide important references for future research on AI.

2.2. Research Hypothesis

2.2.1. Direct Effect of AI on AQ

AI can reduce air pollution levels by improving energy efficiency in the industrial sector and by advancing green, low-carbon technologies. AI can reduce energy consumption and avoid resource waste in daily life, transportation, and other processes by optimizing decision-making, and thereby helps improve energy efficiency [36]. AI can substantially improve the production efficiency [37,38], help enterprises win in market competition (especially in the manufacturing industry), expand their product market, and enhance the energy utilization efficiency of the entire society. Therefore, AI may reduce air pollution levels by improving energy efficiency. In addition, AI can improve AQ by promoting the advancement of green technologies. The expansion of AI application scenarios may directly promote the further development of green, low-carbon technologies. Previous studies showed that AI technology contributes to technological advances [39,40]. AI technology may also assist in technological innovation and indirectly promote the development of green, low-carbon technologies [41]. For example, the combination of AI and intelligent manufacturing can help reduce the mental labor of R&D personnel, improve their R&D efficiency, and thereby increase the speed of equipment upgrades and technological iterations, and indirectly promote the development of green technologies. The development of green technologies can improve AQ during the production process of enterprises, which is beneficial for improving air quality. In summary, hypothesis H1 is proposed: The development of AI can improve AQ.

2.2.2. Indirect Effect of AI on AQ

AI can enhance the government’s attention to environmental governance. AI enables governments to collect environmental data efficiently, and to evaluate government environmental governance performance scientifically. AI also improves the effectiveness of government environmental supervision and the environmental regulatory capabilities and environmental attention of the government. In addition, AI contributes to reduce environmental information asymmetry between different government departments, between government and enterprises, and between government and the public [42], and enhances GEA.

After the government pays more attention to environmental governance, it will increase investment in urban environmental pollution control and thus reduce air pollution [43]. Local governments can invest in environmental pollution control by operating urban pollution treatment facilities, updating emission reduction equipment, introducing green production technologies, and providing basic conditions for improving air quality. Local governments can strengthen their environmental supervision through information technology. The government can impose environmental administrative penalties on illegal enterprises [18], strengthen the investigation and punishment of environmental violations and administrative penalty cases, and restrain environmental violations by enterprises [44]. The deterrent effect generated by this will encourage enterprises to increase environmental investment and thereby reduce air pollution. Therefore, hypothesis H2 is proposed: AI affects AQ through GEA.

2.2.3. Regulatory Effect of PEA

Based on the complexity of environmental governance in China, public participation influences government environmental governance behavior [45]. Utilizing AI technology, the public can pay attention to environmental information through the Internet and public service platforms. AI technology also helps the public express their opinions and comments on environmental governance issues in regions and thereby reflect their demands for environmental quality improvement to the government. The government’s environmental governance behavior will be influenced by local public opinions, and the government will respond accordingly and make environmental policies meet public preferences. Therefore, to respond to the public’s demands and to assume social responsibility toward the public, the government will improve urban AQ by increasing environmental expenditure. In addition, the development of AI promotes public demand, participation, and supervision of the environment, and forces higher-level governments to incentivize and supervise lower-level governments’ environmental governance policies and behaviors, which helps improve local air quality [46]. A stronger PEA facilitates the use of AI technology to influence local governments’ environmental governance behavior and thereby improves AQ. Therefore, hypothesis H3 is proposed: PEA amplifies the improvement effect of AI on AQ.

3. Research Design

3.1. Model

Compared to traditional econometric models, the standard DID model can eliminate biases caused by unobservable factors using two differences and is commonly used for policy effectiveness evaluation. Staggered DID, on the basis of ordinary DID, takes into account the multiple occurrences of policies or events at different time points, allowing for more flexible analysis of the changes in policy effects over time. The staggered DID model requires sample data to be multiperiod panel data, with control group individuals not receiving policy intervention at all periods, and treatment group individuals not receiving policy intervention at exactly the same time and not allowing policy withdrawal. Given that AIPZ is implemented in batches in different cities, staggered DID is suitable for evaluating the policy implementation effectiveness of AIPZ. In the empirical process of this study, economic variable data and environmental variable data at the city level were mainly used, which came from the China Urban Statistical Yearbook and Wind database. This paper took Chinese prefecture-level cities as the research sample and then removed missing values from the sample. This paper ultimately selected 274 prefecture-level cities from 2011 to 2021 as the research sample. Given that the AIPZ policy was established in three batches from 2019 to 2021, the staggered DID was used to investigate the effect of AI on urban AQ:

$${AQ}_{it}={\alpha }_0+{\alpha }_1{AI}_{it}+{\alpha }_2{CV}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it},$$

(1)

In Equation (1), i represents the city and t represents the year. The variable AQ_it represents the AQ level. The variable AI_it represents whether city i established an AIPZ policy in year t. If city i established an AIPZ in year t, the value is 1, otherwise, it is 0. CV_it represents the control variables (CVs) at the city level, ${\alpha }_0$ is a constant term coefficient, ${\alpha }_1$ is the regression coefficient of the core explanatory variable, which can reflect the degree of impact of AI on urban AQ, and ${\alpha }_2$ is the regression coefficient of each control variable.

3.2. Variable

AQ: To obtain reliable results, this paper needs to determine the measurement indicator for AQ. The air pollutants in Chinese cities are complex, and most scholars currently use PM2.5 to characterize air pollution. Based on existing research [47,48,49], the PM_2.5 was used here to measure the AQ level.

AI: The AIPZ policy, a quasi-natural experiment, was used as a proxy variable for AI. This paper used whether city i established an AI experimental zone in year t to represent AI. If city i established AIPZ in year t, the value was 1, otherwise, it was 0.

CV: Based on the research objectives and drawing on existing literature [45,47], this study incorporated the following variables that may affect urban AQ into the DID model: economic development (EC), opening up (OU), city size (CS), financial development (FIN), and government intervention (GIN). Their descriptive statistics are shown in

Table 1. Descriptive statistics.

Variable	Obs	Definition	Mean	SD
AQ	3014	The logarithm of PM2.5 concentration	3.706	2.231
EC	3014	The logarithm of per capita GDP	10.705	0.552
OU	3014	The logarithm of the proportion of FDI to GDP	−6.411	1.226
CS	3014	The logarithm of the total population	8.112	1.587
FIN	3014	The logarithm of the balance of deposits in financial institutions	14.423	2.807
GIN	3014	The general public budget expenditure/GDP	0.198	1.159

.

4. Results Analysis

4.1. Benchmark Regression

Table 2. Benchmark results.

	(1)	(2)	(3)
	AQ	AQ	AQ
AI	−0.0834 *	−0.12 ***	−0.124 **
	(0.0468)	(0.0372)	(0.0524)
EC		0.191 ***	0.301 ***
		(0.0281)	(0.0681)
OPEN		−0.0187 *	0.0294 *
		(0.0108)	(0.0175)
CS		0.116 ***	0.426 ***
		(0.0156)	(0.152)
GIN		0.300 *	0.225
		(0.176)	(0.235)
FIN		2.795 ***	−0.591
		(0.968)	(1.225)
City	N	Y	Y
Year	N	N	Y
R2	0.555	0.182	0.561

Note: *, **, and *** represent significance levels at 10%, 5%, and 1%, respectively, with robust stand errors in parentheses.

reports the empirical regression results of the impact of AI on urban AQ. After continuously adding CV, the impact coefficients of AI on the urban AQ index were significantly negative, indicating that AI helped improve urban AQ. The reason behind this is that the development of AI has raised public awareness of environmental protection, forcing local governments to increase the intensity of environmental pollution control, constrain corporate environmental pollution behavior, and promote air pollution reduction. This indirectly reflects the dual role of AI development in driving economic growth and ecological environment construction.

4.2. Robustness Tests

During the execution of the quasi-natural experiment, there may be issues with self-selection effects, where the selection of pilot cities for AI is not random, but rather freely chosen by the state to become pilot cities. However, the selection of AI pilot cities is not directly related to the urban air pollution situation, so it basically meets the problem of policy exogeneity. Subsequently, this study employed some robustness testing methods to ensure the reliability of the core conclusions presented in this paper.

4.2.1. Parallel Trend

The parallel trend assumption is a prerequisite for the correct use of the DID model, which is mainly used to identify that the development trend of AQ in all cities is consistent before policy shocks occur, indicating that nonsystematic time trend differences between cities before being selected as AI pilot cities will not affect policy effectiveness. Based on the event study method, this paper conducted a parallel trend test. In Figure 2, the DID coefficients (Before3, Before2, and Before1) before the implementation of AIPZ were not significantly different from 0. The DID coefficients (Current, After1, and After2) after the implementation of AIPZ were significantly different from 0. This indicates the effectiveness of the DID model.

4.2.2. Placebo Test

To eliminate the influence of random factors, this paper drew on the approach of La Ferrara et al. (2012) [50] to make the selection of AI pilot cities random and then repeated this random process 500 times. Such random processing did not have an impact on the corresponding AQ. From Figure 3, it can be observed that the coefficients were concentrated around 0 in all 500 random processes, which proves that other unobserved random factors in the city had almost no impact on the estimation results.

4.2.3. Propensity Score Matching (PSM)-DID

In the previous analysis of the impact of AIPZ policy on AQ, it was assumed that the government’s selection of pilot cities was random. However, in practice, the government pays more attention to the economic development foundation of these cities when determining the list of AI pilot cities. Therefore, to further reduce estimation bias, the paper was subjected to robustness testing using the PSM-DID method. This method can minimize the differences in features between cities as much as possible, thereby making the evaluation results more robust. The significance of the AI coefficient in column (1) of

Table 3. Robustness test.

	(1)	(2)	(3)	(4)
	AQ	AQ	AQ	AQ
AI	−0.139 ***	−0.053 ***	−0.116 **	−0.123 *
	(0.051)	(0.017)	(0.052)	(0.067)
CV	Y	Y	Y	Y
City	Y	Y	Y	Y
Year	Y	Y	Y	Y
R2	0.555	0.513	0.562	0.56

Note: *, **, and *** represent significance levels at 10%, 5%, and 1%, respectively, with robust stand errors in parentheses.

verifies the robustness of the PSM-DID results.

4.2.4. Replace Core Variable

Considering that there are many ways to measure AI, existing literature mostly uses industrial robot usage data from different regions when measuring the AI level in cities. Following the approach of existing literature [51], I used the urban penetration of industrial robots to measure AI. The regression result is shown in the second column of Table 3, and the AI coefficient was negative, indicating that the main conclusion remained unchanged after variable replacement.

4.2.5. Eliminate Interference from Environmental Policy

During the analysis period of this paper, the Chinese government implemented a low-carbon city pilot (LCCP) in 2011. This LCCP promotes enterprises to increase green technology R&D investment and thus contribute to improving AQ through administrative orders and intense market competition. The policy effects of the LCCP may affect the estimation results of this paper and need to be excluded. I added a dummy variable of the LCCP to the benchmark regression model to eliminate the interference of LCCP. Specifically, if city i implemented a LCCP in year t, the corresponding dummy variable was assigned a value of 1, otherwise, it was 0. The regression result is shown in the third column of Table 3. The coefficient of AI was still significant, indicating that the core result was still robust after excluding the influence of relevant policy.

4.2.6. Using Clustering Robust Standard Error

Considering that cities within the same province may exhibit similarities in AQ, as well as differences in AQ between cities in different provinces, this paper used robust provincial clustering standard errors to regress. The result is shown in the fourth column of Table 3, indicating that even with the use of province-level clustering robust standard errors, the regression coefficient of the core explanatory variable remained significantly negative.

4.3. Mechanism Analysis

By examining the impact of AI on AQ, it was concluded that AI had a positive effect on improving the AQ. So, what is the reason for this positive effect? In other words, what transmission mechanism does AI use to reduce AQ? Based on the theoretical analysis, this paper examined the mechanism of AI from the perspective of local GEA. I constructed the following mediation effect model for testing:

$${GEA}_{it}={\beta }_0+{\beta }_1{AI}_{it}+{\beta }_2{CV}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it},$$

(2)

$${AQ}_{it}={\delta }_0+{\delta }_1{AI}_{it}+{\delta }_2{GEA}_{it}+{\delta }_3{CV}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it}.$$

(3)

The GEA refers to the government environmental attention. Drawing on the method of Chen et al. (2018) [52], I used the frequency of vocabulary related to the environment and haze in the work reports of prefecture-level municipal governments to measure the level of GEA. The estimated result in column 1 of

Table 4. Mediation effects.

	(1)	(2)
	GEA	AQ
AI	0.12 **	−0.116 **
	(0.046)	(0.052)
GEA		−0.064 ***
		(0.018)
CV	Y	Y
City	Y	Y
Year	Y	Y
R2	0.817	0.567

Note: ** and *** represent significance levels at 5%, and 1%, respectively, with robust stand errors in parentheses.

shows a significant positive correlation between the development of AI and GEA, indicating that the AI can significantly enhance GEA. The result in the second column shows that AI can reduce urban air pollution levels by enhancing GEA. When the development of AI enhances the government’s attention to environmental protection, it can improve AQ by forcing local governments to increase environmental investment [53,54,55]. Specifically, local governments can strengthen the environmental awareness of enterprise management by providing corresponding environmental resource support, enabling enterprises to have resources and be willing to continue carrying out green innovation activities. The enterprise has taken measures such as upgrading production processes, improving emission equipment, and replacing new energy to reduce pollutant emissions, which is in line with the local government’s goal of promoting environmental governance. In addition, AI technology can help governments predict the reduction effects of policy implementation, thereby assisting them in formulating more appropriate reduction policies. AI can also help governments monitor emissions, effectively enhancing their regulatory capabilities and better leveraging their regulatory role.

4.4. Moderating Effect of PEA

As a supplement to government environmental regulations, how does public concern and participation in the environment affect local government environmental governance behavior and air pollution reduction? I constructed a moderation effect model to test the moderation effect of PEA:

$${AQ}_{it}={\rho }_0+{\rho }_1{AI}_{it}+{\rho }_2{PEA}_{it}++{\rho }_2{PEA\ast AI}_{it}+{\rho }_4{CV}_{it}+{\mu }_i+{\gamma }_t+{\epsilon }_{it},$$

(4)

PEA refers to public environmental attention. PEA is an important manifestation of the public’s preference and participation in environmental governance, reflecting their environmental behavior. Drawing on the research ideas of Zheng et al. (2012) [56], this paper used the Baidu Haze Search Index to characterize PEA. The larger the index, the higher the public’s attention to haze control.

Table 5. Moderating effect.

	(1)	(2)
	AQ	AQ
AI*PEA	−0.0338 ***	−0.0276 **
	(0.013)	(0.0129)
CV	N	Y
City	Y	Y
Year	Y	Y
R2	0.559	0.562

Note: ** and *** represent significance levels at 5% and 1%, respectively, with robust stand errors in parentheses.

reports the moderating effect of PEA, and regardless of whether CVs were included, the coefficients of PEA were significantly negative. This indicates that PEA had a regulatory effect, that is, it amplified the role of AI in improving AQ. PEA reflects the demands of the general public for environmental protection, which can be transformed into public participation in environmental pollution control and help improve the effectiveness of government environmental governance. Therefore, PEA can amplify the improvement effect of AI on urban AQ.

4.5. Heterogeneity

4.5.1. Geographic Location

Geographical differences have profound, complex effects on economic culture, natural resources, and industrial layout. This paper explored the effect of AI development on regional heterogeneity of AQ from geographical location. The sample was divided into the eastern and the mid-west region, and regression analysis was conducted separately, as shown in

Table 6. Heterogeneity analysis.

	(1)	(2)	(3)	(4)
	Mid-West Region	Eastern Region	Resource-Based	Non-Resource-Based
	AQ	AQ	AQ	AQ
AI	−0.12 *	−0.152 ***	−0.095 *	−0.176 ***
	(0.0721)	(0.0575)	(0.0527)	(0.0586)
CV	Y	Y	Y	Y
City	Y	Y	Y	Y
Year	Y	Y	Y	Y
R2	0.609	0.516	0.559	0.563

Note: * and *** represent significance levels at 10% and 1%, respectively, with robust stand errors in parentheses.

. The absolute value of the AI coefficient in the sample of the eastern region was considerably higher than that in the mid-west region and indicated that the role of AI in reducing air pollution was reflected in the eastern region. The development of AI not only relies on policy support in the later stage but also on the economic development foundation of the city itself in the early stage. The AI pilot cities emphasize the application of information technology, which requires the updating of a large amount of infrastructure. This goal cannot be achieved without the city’s economic strength and advanced industrial structure. Cities in the eastern region developed earlier and had strong economic strength, more technological talents, and advanced, complete infrastructure, which is more conducive to the development of AI. In the mid-west region, abundant natural resources and low labor costs have attracted industrial transfer, but most of these industries that migrate to the mid-west region are relatively low-end, high-energy-consuming labor-intensive industries. The arrival of these industries has increased the proportion of the secondary industry in local cities, which has brought economic growth but also generated certain pollution. Due to the late start of economic development and weak economic foundation in the mid-west region, there is a severe shortage of public service facilities, such as education and healthcare. In addition, the relatively low-end industry types make it difficult to attract and retain talent, resulting in the main low-skilled labor force in the mid-west region. The application of intelligent technology in production and daily life requires sufficient funding and highly skilled personnel to support it. Therefore, in the economically weak and talent-scarce mid-west region, the effect of AI on urban air pollution control is not particularly significant. Therefore, the improvement effect of AI on AQ in the eastern cities is better than that in the mid-west cities.

4.5.2. Resource Type

Drawing on existing literature [57], the sample cities were divided into resource-based and non-resource-based cities. Table 6 shows the effectiveness of AI in addressing air pollution in the two types of cities. The regression results revealed that the absolute value of the AI coefficient in non-resource-based cities was substantially higher than that in resource-based cities, and AI had a more considerable effect on improving air quality in non-resource-based cities. A possible reason for this outcome is that resource-based cities have a single type of industry, mainly focused on high-energy-consuming mineral resource industries. To a certain extent, they sacrifice the ecological environment for economic growth, and path dependence leads to greater resistance in promoting the green transformation of development mode. The green technology level and various supporting environments of resource-based cities are relatively backward, and the development of AI is easily hindered in this area, making it difficult to fully exert its positive impact on AQ. Furthermore, the environmental regulatory requirements in resource-based cities may be relatively relaxed, leading to weaker awareness, driving force, and environmental protection pressure for enterprises to self-innovate. Therefore, the development of AI may be more difficult to drive the green upgrading of local industries. In this case, the development of AI may not have a strong effect on improving urban AQ in these areas. Therefore, non-resource-based cities are more effective in using AI to improve AQ than resource-based cities.

5. Conclusions

AQ is a considerable global challenge, and its discussion has a long history with rich achievements, which provides a theoretical basis for later research. Most discussions on AQ focus on its sources and transmission pathways, which are influenced by natural and social factors. However, few papers have conducted in-depth research on the development of AI. This paper took the urban air pollution status as the research object, used the DID method to explore the effect of AI on AQ, and then conducted a heterogeneity discussion and mechanism testing. (1) A remarkable negative correlation existed between AI and urban air pollution; thus, AI can improve AQ. (2) AI can reduce air pollution by strengthening government environmental attention, while public environmental attention can enhance the improvement effect of AI on AQ. (3) The reduction effect of AI on urban air pollution was mainly reflected in eastern cities and non-resource-based cities.

The following policy recommendations are proposed.

First, the government should encourage various regions across the country to develop AI based on local conditions. AI can substantially optimize air quality, alleviate urban disease problems, promote sustainable development of cities, and narrow the gap in air pollution between cities. Moreover, AI is one of the important ways to promote China’s economic development in the dual context of the global fourth industrial revolution and domestic economic transformation. In addition, when the country encourages the construction of AI cities, local governments should actively respond to the national call, draw on advanced development experience according to the specific situation of each region, formulate suitable AI city construction plans for local development, and apply them into practice.

Second, the government should incorporate environmental protection into the performance evaluation system of local officials, effectively collect, integrate, and share environmental pollution monitoring data through AI technology, accurately evaluate government environmental governance performance, and enhance government environmental supervision capabilities. In addition, the government can guide residents to form environmental awareness and encourage residents to participate in pollution control. The government also needs to use AI to guide public attention and participation in environmental governance.

Third, to promote the implementation of AI cities nationwide, the government needs to break down regional barriers and strengthen regional communication. Given that the pilot cities for AI emphasize the application of information technology, the development of information technology in the eastern region is also ahead of that in the mid-west region. The government needs to encourage the transfer of advanced technology from the east to the mid-west region through a series of supportive policies. For resource-based cities, local governments need to break free from resource path dependence, actively introduce and develop AI industries, form new economic growth points, and thus mitigate air pollution.

Fourth, the government needs to support the development of advanced AI enterprises in China. The government can focus on supporting a group of advanced AI enterprises through post-tax incentives and government subsidies, improving their research and innovation capabilities, and enabling them to lead the development of the entire industry as leading enterprises. The government also needs to actively build high-tech AI industrial parks, improve the level of supporting facilities, fully leverage industrial agglomeration effects and industrial linkage effects, promote the research and application of high-tech cutting-edge AI technologies, achieve green transformation, and reduce pollution emissions.

Due to limited data conditions, this study has some limitations. Existing research has not only analyzed the impact of AI on PM_2.5 but also explored its effects on harmful gases, such as SO₂ and NO₂, as well as its impact on water pollution. Whether AI can assist enterprises in green transformation and reduce air pollution is also a matter of concern, and this issue should receive attention in future research. In addition, limited by data conditions, the evaluation of AI in this study may not be scientifically comprehensive enough. Future research can further optimize this aspect to obtain more scientific and effective AI evaluation indicators, providing important assistance for further research in the future.