Artificial Intelligence and Social Well-Being in the Yellow River Basin: A Cultural Lag Theory Perspective

Abstract

Amid comprehensive reforms, artificial intelligence (AI) has emerged as a vital force in solving people’s problems and enhancing quality of life. Yet, theoretical inquiries into the mechanisms by which AI influences social well-being remain limited. Drawing upon cultural lag theory, this study constructs a social well-being index system based on the Gini coefficient objective weighting method. By integrating a moderated mediation model with a spatial econometric model, it examines the mechanisms and impacts of artificial intelligence on social well-being. The findings reveal that AI induces multiple cultural lags and exerts a U-shaped impact on social well-being. AI enhances well-being through the channels of employment opportunities, human capital, and green innovation, while digital inclusion and foreign direct investment (FDI) further reinforce this relationship. Additionally, AI generates spatial spillover effects on social well-being, and the region’s well-being landscape exhibits convergence. However, both digital inclusion and FDI negatively moderate the convergence process, slowing its overall pace. These insights provide substantial practical guidance for crafting informed policies aimed at elevating public well-being.

1. Introduction

The pursuit of happiness for the people has been the unwavering mission of the Communist Party of China (CPC) [1]. Upholding a governance philosophy centered on the people, the Chinese government has consistently fulfilled its promise to “help the populace live happier and more fulfilling lives”. This commitment has led to comprehensive improvements in living standards and an ongoing enhancement of subjective well-being. Social well-being is influenced by an array of macro- and micro-level factors. As a crucial driver of emerging productive capacities, artificial intelligence (AI) has simultaneously triggered revolutionary technological transformations and injected new vitality into societal welfare. However, the increasing application of AI technologies has also elicited concerns among the public. Phenomena such as “Apollo Go” (driverless taxis) and other labor-replacing innovations have exacerbated anxieties about imminent job displacement and insufficient compensation, highlighting a significant structural imbalance. This has spurred scholarly debates on balancing emerging productive forces with traditional industries as well as on AI’s potential implications for social well-being [2].

Since ancient times, the Yellow River Basin has served as a crucial economic center and ecological security barrier in northern China, which is often referred to as the nation’s “energy basin”. Given its strategic importance in national economic, ecological, and energy security, the ecological protection and high-quality development of the Yellow River Basin was officially elevated to a national strategy in October 2021, becoming a key agenda for promoting regional coordination and sustainable development. Spanning multiple provinces across eastern, central, and western China, the Yellow River Basin encompasses developed, developing, and less-developed regions, which are characterized by significant disparities in economic structure, industrial composition, and social development levels. These disparities reflect the multi-layered nature of China’s regional economic development. Compared to more developed regions such as the Yangtze River Delta and Pearl River Delta, AI development in the Yellow River Basin is primarily focused on applied research with technology diffusion and industrial deployment exhibiting distinct gradient patterns across different areas. For instance, in the more developed eastern provinces, AI primarily drives industrial upgrading and smart city development, whereas in the central and western regions, AI plays a greater role in agricultural modernization, the transformation of resource-based cities, and the optimization of public services. These regional differences make the Yellow River Basin an ideal case for analyzing the impact of AI on social well-being while also enhancing the representativeness and generalizability of the study’s findings. Against the backdrop of comprehensive reforms, investigating the impact of AI on social well-being in the Yellow River Basin holds vital significance for advancing common prosperity and steadily enhancing societal happiness.

Extensive academic discourse has explored the interplay between AI and social well-being, with existing studies analyzing AI’s impact from various dimensions, including technological enablement, emotional interaction, economic growth, industrial applications, and smart services. Early research predominantly focused on AI’s efficiency enhancement effects. Munoz (2002) [3] observed that AI could effectively shorten design cycles and alleviate designers’ workloads, thereby improving convenience in human life. Komninou (2003) [4] analyzed AI’s role in emotional interaction and advocated for embedding empathetic care into its development. In recent years, research has increasingly shifted toward the economic empowerment effects of artificial intelligence. Makridis and Saurabh (2022) [5] argued that technological transformation could enhance well-being by stimulating economic activity. Dogan et al. (2023) [6] found that AI could unlock its potential to deliver higher-quality services to tourists, thereby enhancing their happiness. Similarly, Liang (2022) [7] demonstrated that AI-powered smart community services significantly improved residents’ happiness. Furthermore, studies have examined AI’s impact on organizations and industries. Gong (2025) [8] reported that collaborative service robots positively impacted employee well-being. Furthermore, some scholars have highlighted that artificial intelligence can have a positive impact on supply chain resilience and green innovation (Wang et al., 2024; Cui, 2024) [9,10]. Nevertheless, despite the breadth of research on AI and social well-being, several limitations persist. The existing literature exhibits three primary gaps, which necessitate further investigation. Firstly, existing research tends to adopt fragmented perspectives, focusing on specific industries, organizations, or individual experiences. However, these studies lack a unified analytical framework to consolidate findings across different domains and comprehensively assess AI’s overall impact on social well-being [11]. Secondly, while most studies highlight AI’s positive effects, they often overlook the challenges associated with AI adaptation. AI-driven automation can disrupt traditional labor markets, leading to employment displacement and social inequality. Finally, much of the existing literature relies on micro-level econometric models, which primarily examine AI’s direct effects on firms, industries, or individuals. However, few studies explore AI’s broader regional impact, particularly its spatial spillover effects. AI-driven innovations in leading regions may create positive externalities for neighboring areas or, conversely, widen regional disparities if technological diffusion is uneven.

Considering that research on how AI affects social well-being in the context of developing countries or regions with uneven economic development remains relatively limited, this study examines the case of China’s Yellow River Basin. Utilizing panel data from 71 prefecture-level cities in the Yellow River Basin spanning 2012 to 2022, the study explores the impact of AI on social well-being and its underlying mechanisms, guided by cultural lag theory, providing a new perspective on AI applications in developing countries. It examines the short-term lags introduced by AI, its spatial spillover effects, and the convergence dynamics of social well-being. This study makes three significant contributions to the existing literature. Firstly, it develops a novel framework for measuring social well-being, offering a fresh perspective for well-being evaluation. Secondly, it applies cultural lag theory to analyze AI’s impact on social well-being in the Yellow River Basin, enriching the related research discourse. Lastly, it identifies three pathways—human capital, employment opportunities, and green innovation—through which AI influences social well-being, advancing the understanding of the intricate logical linkages between AI and well-being.

This study seeks to offer theoretical insights and practical guidance for policymakers aiming to enhance societal well-being in the context of AI-driven transformation

2. Literature Review and Research Hypothesis

The notion of multiple lags stems from the cultural lag theory proposed by the American sociologist W.F. Ogburn. He posited that significant social transformations or transitions often result in certain aspects of culture lagging behind others, creating disparities (Keith et al., 2021) [12]. Zhang Bo, one of China’s leading pioneers in artificial intelligence (AI), emphasized that the ultimate purpose of AI development is to enhance human happiness. However, the substitution effects of AI have outpaced its creation effects, leading to frequent occurrences of “machine replacement” phenomena, which diverge from the expectation of improving social well-being (Daron et al., 2022) [13].

Drawing on cognitive evaluation theory, the reduction in employment opportunities caused by AI induces a range of societal issues, fostering pessimistic expectations and a sense of loss in happiness compared to anticipated gains (Qasim et al., 2021) [14]. Additionally, unemployment can lead to “scar effects” (Jonas et al., 2017) [15], exacerbating adverse impacts on mental health and subjective well-being, thereby delaying improvements in happiness. The rapid development of AI often results in workers’ skill sets falling short of practical needs, fostering a perception of falling behind the times, which further undermines well-being. In this sense, the multiple lags induced by AI contribute to a decline in happiness.

As technology continues to advance, initiatives such as “AI+” and “+AI” promote the adoption of AI across various scenarios, enhancing social environments for production and daily life. Specifically, in the eastern provinces of the Yellow River Basin (e.g., Shandong and Henan), the artificial intelligence industry has begun to take shape, with rapid advancements in intelligent manufacturing and smart city development, providing sustained momentum for high-quality regional economic growth. In contrast, AI technology adoption in the central and western regions (e.g., Ningxia and Qinghai) remains in its early stages. However, with strong government support, AI-driven applications focused on social well-being—such as smart agriculture and telemedicine—are gradually being implemented, offering new drivers for regional economic development and continuously improving the accessibility and quality of public services. This encourages individuals to form positive perceptions, thereby increasing social well-being (Charles and Aleksandr, 2020) [16]. At that stage, the negative effects of AI on well-being are expected to reach a nadir and undergo a U-shaped reversal, ultimately driving significant improvements in overall societal happiness.

In summary, the trajectory of social well-being in the Yellow River Basin demonstrates an initial decline followed by a subsequent rise as AI development and application progress. Specifically, in the short term, the economic and social disruptions caused by AI provoke negative perceptions, suppressing social well-being in the region. However, with advancements in AI technology and its widespread penetration across industries, the resulting technological dividends gradually foster positive perceptions, thereby enhancing well-being. Based on this analysis, Hypothesis 1 is proposed:

Hypothesis 1:

The impact of AI on social well-being in the Yellow River Basin follows a U-shaped trajectory.

The rapid evolution of AI has accelerated the transition from “thousand-person factories” to “unmanned factories”, breaking traditional constraints on labor quantity and human capital while significantly enhancing productivity. However, this has also triggered substantial structural unemployment and job polarization, creating a series of social challenges (Ross et al., 2020) [17]. In the short term, these disruptions prevent the social environment from achieving an effective equilibrium, resulting in a gap between existing conditions and societal needs. For example, the Yellow River Basin contains a significant number of resource-based cities, where AI technology is primarily applied in smart mining and intelligent energy dispatching. While these advancements enhance safety and production efficiency, they also reduce labor demand, introducing uncertainties in social well-being.

The efficiency of AI amplifies the “Matthew effect” in employment: low-skilled positions are continuously replaced in large numbers, while high-skilled roles remain underfilled due to high entry barriers, exacerbating their scarcity. Moreover, as reservoirs of talent, higher education institutions often struggle to align with market demands (Konstantin and Svetlana, 2020) [18]. Consequently, the resulting mismatch manifests as a lag in human capital development caused by AI.

In the domain of green innovation, as the primary drivers of innovation in China (Li and Xin, 2020) [19], enterprises’ increased adoption of AI technologies and investment in new product development inevitably crowd out resources allocated to green innovation, creating substitution effects that hinder progress (Yu et al., 2021) [20]. However, as AI matures, it facilitates more efficient resource allocation (Femi et al., 2022) [21], reduces the risk of failure in green innovation, and increases the output of environmentally friendly advancements (Lee et al., 2022) [22].

Thus, the lags in human capital, employment opportunities, and green innovation illustrate the multifaceted lags induced by AI development. From this analysis, Hypothesis 2 is proposed:

Hypothesis 2:

In its initial stages, AI exerts suppressive effects on human capital, employment opportunities, and green innovation, leading to lags that fail to meet societal needs.

As artificial intelligence advances beyond the technological singularity (William, 2021) [23], its technological dividends are expected to generate substantial job opportunities in new and emerging fields, effectively addressing prior employment deficits. In this scenario, abundant and high-quality employment will bolster confidence in economic recovery and positive growth trajectories, alleviating the stress previously associated with economic pessimism and labor market shortages. Furthermore, AI facilitates an increase in industrial value addition, improves remuneration for job positions, and provides individuals with more optimistic employment prospects, fostering positive perceptions.

In parallel, the development and application of AI compel an enhancement in human capital, driving a shift toward a higher-quality, skill-intensive workforce. Improvements in human capital elevate the societal income structure (Silvia et al., 2021) [24], thereby further enhancing social well-being and quality of life.

Green innovation, as a critical technological activity for advancing green technology and improving the ecological environment, plays an essential role in harmonizing ecological sustainability with economic growth (Junaid et al., 2022) [25]. AI not only provides robust technical support for green innovation but also injects renewed vitality into this domain (Tao, 2024) [26]. The environmentally friendly nature of green innovation projects often attracts government prioritization and support. Governments are generally inclined to provide financial and policy backing to promote progress in green technologies. Green innovation serves as a pivotal engine for the development of advanced productivity and is foundational to meeting the populace’s growing demand for a pristine ecological environment (Abdullah et al., 2024) [27].

Drawing on cognitive evaluation theory, as environmental quality improves, individuals’ increasing needs for an enhanced ecological environment will be better fulfilled. People will cognitively perceive the positive changes in their surroundings, which will, in turn, stimulate an intrinsic sense of social well-being.

Building on this analysis, the following hypotheses are formulated:

Hypothesis 3a:

AI influences social well-being through its impact on employment opportunities.

Hypothesis 3b:

AI influences social well-being through its impact on human capital.

Hypothesis 3c:

AI influences social well-being through its impact on green innovation.

Digital inclusion and foreign investment play critical roles in the relationship between AI and social well-being (Lin and Ju, 2024; Ye et al., 2022) [28,29]. Supported by cloud computing and big data, digital inclusion integrates deeply with advanced AI technologies, fostering intelligent financial services, health insurance systems, and other service ecosystems that provide more personalized and diversified services to the public (Alberto et al., 2023) [30]. Moreover, while AI technology reduces the failure risks associated with individual investments and entrepreneurship, digital inclusion alleviates financial resource constraints, thereby enhancing individuals’ resilience to external risks and stimulating economic and social dynamism (Sun et al., 2022) [31].

According to resource conservation theory, the availability of abundant resources effectively reduces individual stress and enhances autonomous decision-making capabilities, leading to positive psychological perceptions. From the perspective of emotional contagion theory, emotions within a social environment are influenced by the emotions of others. In this context, positive emotions can encourage similar emotional responses in others, thereby amplifying the positive impact of AI on social well-being. Consequently, digital inclusion positively moderates the U-shaped relationship between AI and social well-being.

The advancement of AI demands substantial investments in capital, talent, and infrastructure. Foreign investment not only provides critical funding but also introduces advanced technologies, thereby accelerating AI development and application, unlocking greater technological dividends, and strengthening its impact on social well-being (Zhang, 2024) [32]. Furthermore, foreign investment, through labor demand and technological spillover effects, enhances human capital, improves worker income, and fosters a favorable socioeconomic environment, ultimately contributing to increased social well-being. Building on this analysis, the following hypotheses are proposed:

Hypothesis 4a:

Digital inclusion positively moderates the association between AI and social well-being.

Hypothesis 4b:

Foreign investment positively moderates the association between AI and social well-being.

AI technology, powered by high-speed information networks, transcends geographical boundaries, fostering interregional complementarity and technological exchange. This has generated strong momentum for high-quality development in the Yellow River Basin and has significantly improved individual well-being. On the one hand, AI accelerates the transition of enterprises toward intelligent production. Under the influence of competitive effects, it drives industrial upgrading. When product quality improves, new consumer demand emerges and spreads across regions, leading to an overall improvement in residents’ living standards. On the other hand, according to the theory of social presence, AI technologies enrich market strategies, mental health interventions, and social interactions by incorporating elements like personalized services and anthropomorphic communication styles. These innovations enhance the sense of social presence, making individuals feel understood and valued, thereby increasing their subjective well-being (Gabriele et al., 2021) [33].

Given the non-exclusive nature of AI technologies, new advancements and business models are continuously transferred to surrounding cities. Supported by digital technologies, the spatial spillover effects of AI are fully realized, thereby enhancing social well-being throughout the Yellow River Basin. Accordingly, the following hypothesis is proposed:

Hypothesis 5:

AI exerts a spatial spillover effect on social well-being in the Yellow River Basin.

However, considering the regional characteristics of the Yellow River Basin, digital inclusion and foreign investment may not always positively influence the convergence process of social well-being. Challenges such as low levels of human capital, imbalances in digital infrastructure development, and insufficient technological readiness exacerbate inequalities, particularly in underdeveloped areas. Digital inclusion may fail to fully realize the benefits of technological dividends due to issues such as access gaps, usage gaps, and skill gaps, slowing the convergence of social well-being in lagging regions. Although digital inclusion theoretically breaks geographical barriers and mitigates information asymmetry, the adaptability and acceptance of digital technologies among aging populations remain limited. This may result in restricted consumer autonomy and negative user experiences, further compounding these challenges. The emotional distress associated with exclusion could transmit through familial ties, intensifying the negative impacts on social well-being improvement in less developed areas (Liu and Zhang, 2022) [34].

Regarding foreign investment, its profit-driven nature makes it highly sensitive to external environmental changes. In a complex international landscape, trade frictions increase the risk of foreign capital withdrawal, which adds uncertainty to economic stability (Canh and Gabriel, 2021) [35]. The “exit of foreign investment” could weaken the financial system’s stability in the Yellow River Basin, undermine confidence in economic and social development, and hinder improvements in social well-being. Additionally, underdeveloped regions often struggle to attract high-tech and high-value-added enterprises, relying instead on traditional industries like mining and metal processing. This exacerbates environmental degradation and constrains the public’s growing demand for an improved ecological environment (Xu et al., 2022) [36], dampening the convergence process of social well-being. Based on the above analysis, the following hypotheses are proposed:

Hypothesis 6a:

Digital inclusion negatively moderates the convergence process of social well-being in the Yellow River Basin.

Hypothesis 6b:

Foreign investment negatively moderates the convergence process of social well-being in the Yellow River Basin.

3. Research Design

3.1. Sample Selection

Drawing on the geographic boundaries defined by the Yellow River Conservancy Commission of the Ministry of Water Resources, and ensuring the availability and continuity of regional data, this study constructs a panel dataset of 71 prefecture-level cities in the Yellow River Basin for the period 2012–2022, building on the framework established by prior research (Zhao and Wang, 2024) [37]. The specific distribution of these cities is presented in

Table 1. Cities in the Yellow River Basin.

Provinces	Municipalities
Inner Mongolia	Hohhot, Baotou, Wuhai, Ordos, Bayannur, Ulanqab
Gansu	Lanzhou, Baiyin, Tianshui, Wuwei, Pingliang, Qingyang Dingxi, Longnan
Ningxia	Yinchuan, Shizuishan, Wuzhong, Guyuan, Zhongwei
Shaanxi	Xi’an, Tongchuan, Baoji, Xianyang, Weinan, Yan’an, Yulin, Shangluo
Shanxi	Taiyuan, Datong, Yangquan, Changzhi, Jincheng, Shuozhou, Jinzhong, Yuncheng, Xinzhou, Linfen, Luliang
He’nan	Zhengzhou, Kaifeng, Pingdingshan, Anyang Hebi, Xinxiang Puyang Xuchang, Luohe, Nanyang, Shangqiu, Xinyang, Zhoukou, Luoyang, Jiaozuo, Sanm enxia, Zhumadian
Shandong	Jinan, Qingdao, Zibo, Zaozhuang, Dongying, Yantai, Weifang, Jining, Tai’an, Weihai, Rizhao, Linyi, Dezhou, Liaocheng Binzhou and Heze.

.

3.2. Research Methodology

3.2.1. Gini Coefficient-Based Objective Weighting Method

The Gini coefficient, an economic indicator used to measure income inequality across regions or nations, provides an accurate reflection of regional income disparities. Drawing from the principles of the Gini coefficient, this study adopts an objective weighting method based on the Gini coefficient to evaluate the well-being levels of various prefecture-level cities in a more accurate and unbiased manner. By analyzing the Gini coefficients for various indicators, it becomes possible to capture their informational content and trends more precisely. The greater the variation among cities for a specific indicator, the higher its weight should be in assessing well-being. Guided by existing research (Zhao and Wang, 2024) [37], this study applies the Gini coefficient-based objective weighting method combined with a comprehensive evaluation model to measure the well-being levels across cities in the Yellow River Basin. The method of calculation is as follows:

• ① Gini coefficient calculation

$$G_j={\displaystyle \sum _{k=1}^n{\displaystyle \sum _{i=1}^n\left|U_{jk}-U_{ji}\right|}}/2n^2{\mu }_j$$

(1)

where G_j represents the Gini coefficient value for indicator j. U_jk and U_ji are the raw data, n is the total number of data points, and μ_j is the expected value of the j indicator.

• ② Weight calculation

$$w_j=G_j/({\displaystyle \sum _{j=1}^m{G}_j})$$

(2)

• where $w_j$ represents the weight of indicator j, and m is the number of indicators.

• ③ Standardization effectiveness indicators:
• Effectiveness indicators:

$$M_{ijt}={\displaystyle \frac{x_{ijt}-\min (x_j)}{\max (x_i)-\min (x_j)}}$$

(3)

• Cost-based indicators:

$$M_{ijt}={\displaystyle \frac{\max (x_j)-x_{ijt}}{\max (x_j)-\min (x_j)}}$$

(4)

• where M_ijt and x_ijt denote the standardized value of indicator j for city i in the year t. max(x_j) and min(x_j) are the maximum and minimum values of indicators j during the study period.

• ④ Comprehensive Evaluation

$$y_{it}={\displaystyle \sum _{j=1}^m{w}_j{m}_{ijt}}$$

(5)

3.2.2. Spatial Vector Angle Method

The concept of industrial structural upgrading refers to the transition of an economy’s industrial focus from lower to higher tiers as economic growth progresses, signifying the upgrading and transformation of the industrial structure. By constructing an industrial structural upgrading index using the angle between the vectors of the value-added contributions from the three main industries of each city and the corresponding spatial coordinate axes, this method effectively captures the dynamic characteristics of structural changes (Liu et al., 2024) [38].

3.2.3. Model Specification

To explore the impact of artificial intelligence (AI) on social well-being in the Yellow River Basin, the following model is constructed:

$$Happ{y}_{it}={\alpha }_0+{\alpha }_{1}A{i}_{it}+{\alpha }_{2}A{i}_{it}^2+\alpha Z_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(6)

In the equation, Happy_it represents the social well-being index of city i in year t; Ai denotes the level of artificial intelligence development in the city, serving as the core explanatory variable; Ai² is the squared term of artificial intelligence development, included to explore the U-shaped relationship between artificial intelligence development and social well-being; Z_it represents a series of control variables;  ${\mu }_i$ captures city-specific fixed effects; ${\delta }_t$ represents time-fixed effects; and ${\epsilon }_{it}$ is the stochastic error term.

To further explore the spatial spillover effects of artificial intelligence, we incorporate spatial factors into the model and construct analyses using the adjacency weight matrix, the geographic weight matrix, and the economic–geographic matrix. The specific models are as follows:

$$Happ{y}_{it}={\lambda }_0+\rho WHapp{y}_{it}+{\lambda }_{1}WA{i}_{it}+{\alpha }_{1}A{i}_{it}+\lambda W{Z}_{it}+\alpha Z_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(7)

$${\epsilon }_{it}=\eta W{{\epsilon }^{\prime }}_{it}+{\xi }_{it}$$

(8)

Here, $\rho$ represents the spatial autoregressive coefficient, and $W$ denotes the spatial weight matrix. ${\alpha }_1$ and $\alpha$ are the elasticity coefficients of the core explanatory variable and control variables, respectively, while ${\lambda }_1$ and $\lambda$ correspond to the elasticity coefficients of the spatial lag terms of the core explanatory variable and control variables. η represents the spatial error coefficient, which measures the spatial dependence of the stochastic error terms. When $\eta$ = 0, Equation (2) becomes the Spatial Durbin Model (SDM). When $\eta$ = 0, ${\lambda }_1$ = 0, and $\lambda$ = 0, Equation (2) simplifies to the Spatial Autoregressive Model (SAR). When $\rho$ = 0, ${\lambda }_1$ = 0, and $\lambda$ = 0, Equation (2) corresponds to the Spatial Error Model (SEM).

3.3. Variable Selection

3.3.1. Dependent Variable

The Third Plenary Session of the 20th Central Committee underscored the importance of strengthening inclusive, fundamental, and safety-net provisions for public welfare to continually fulfill people’s aspirations for a better life and enhance societal well-being. Societal well-being not only encompasses individuals’ internal psychological perceptions but also reflects their interaction and integration with broader societal systems. It serves as a critical metric for assessing the health and progress of society in the Yellow River Basin. Drawing on the World Happiness Report published by the United Nations, the Happiness Index Research Report 2021 by Peking University, and existing studies (Li et al., 2011) [39], this study constructs a Social Happiness Index (Happy) for prefecture-level cities in the Yellow River Basin. The index is derived from five dimensions: medical resources, ecological environment, social conditions, material living standards, and technological support, ensuring both data accessibility and scientific rigor. Details are shown in

Table 2. Social happiness index indicator system.

Dimension	Indicator	Unit	Attributes	Weight
Medical Resources	Number of health technicians per 1000 people	Person	+	0.065
	Number of hospital beds per 10,000 people	Bed	+	0.062
Ecological Environment	Sewage treatment rate	%	+	0.005
	Non-hazardous waste disposal rate	%	+	0.013
	Green space area in parks	10,000 m2	+	0.091
Social Conditions	Per capita road area	m	+	0.013
	Number of criminal cases per 10,000 people	Case	−	0.077
Material Conditions	Urban residents’ Engel coefficient	%	−	0.017
	Rural residents’ Engel coefficient	%	−	0.025
	Per capita disposable income of urban residents	10,000 RMB	+	0.278
	Per capita disposable income of rural residents	10,000 RMB	+	0.031
Scientific Support	R&D investment	100 million RMB	+	0.092
	Number of invention patents granted	100 cases	+	0.231

.

3.3.2. Independent Variable

In the current academic discourse, there is no consensus on the measurement of artificial intelligence (AI). Scholars frequently adopt metrics such as the density of robot installations, fixed asset investments in the computer software industry, and the number of AI enterprises to measure AI development. Granted patents undergo rigorous examination to ensure novelty, inventiveness, and industrial applicability, making them a direct indicator of innovation and research achievements. Moreover, patent grants provide crucial support for the practical application of related technologies, thereby enhancing the overall development of artificial intelligence. Consequently, the number of AI patent grants effectively reflects a region’s technological progress and innovation capacity in this field. A higher number of granted patents indicates a stronger AI development level within the region. Considering that the number of AI patent grants can effectively reflect the technological progress and innovation capacity of a region in the AI field, this study, following previous research (Wang et al., 2023; Luo and Feng, 2024) [40,41], uses the logarithmic transformation of AI patent grants (Ai) as a proxy variable to represent AI development levels in a region. For robustness checks, methods from existing studies were employed, using Python3.13.2-based data-scraping techniques to collect data on the number of AI enterprises in cities (AiEnt) as an alternative measure of AI development (Cui, 2024) [9].

3.3.3. Mediating Variables

Human Capital Level (Hum): The rapid growth of AI drives enterprises to demand higher and broader levels of talent, thus reshaping the structure of human capital. Considering data availability and drawing on prior research (Li and Zhang, 2024) [42], this study uses the proportion of university students to the local population as a measure of human capital.

Green Innovation (Gre): Compared to green utility model patents, green invention patents more effectively reflect the attributes of technological progress, making them a superior measure of green innovation quality. Referring to established studies (Xu et al., 2023) [43], the number of green invention patent grants is used as a proxy for green innovation.

Employment Opportunities (Emp): The initial phases of AI adoption, marked by job displacement effects, can impact employment opportunities, creating short-term employment gaps. To measure employment opportunities more accurately, this study references existing literature and adopts two dimensions (Zhang and Qiao, 2023) [44]: employment density (measured as the ratio of employed individuals to city area) and unemployment rate (measured as the registered urban unemployment rate). Due to significant data limitations, the latter metric is primarily used. Consistent with the Social Happiness Index, employment opportunities (Emp) are calculated using the Gini coefficient-based objective weighting method.

3.3.4. Moderating Variables

Digital Financial Inclusion (Dig): Digital financial inclusion drives the digital transformation of traditional industries, providing momentum for industrial upgrading. Consistent with many academic studies, this variable is measured using the Digital Financial Inclusion Index published by the Digital Finance Research Center at Peking University.

Foreign Direct Investment (FDI): Foreign direct investment can alleviate capital shortages in the process of economic development and enhance total factor productivity through technological spillover effects. Adopting a common approach in academic research, this study measures FDI by calculating the ratio of actual utilized foreign capital (converted to local currency using annual average exchange rates) to GDP for each city.

3.3.5. Control Variables

Beyond the core independent variable, various factors such as economic development levels and urbanization rates across prefecture-level cities can influence social well-being. Drawing on prior studies, this research incorporates the following control variables:

Educational Investment (Edu): measured as the ratio of fiscal education expenditure to total public fiscal expenditure.

Economic Development Level (lnPgdp): represented by the natural logarithm of per capita GDP for each prefecture-level city.

Industrial Structure Upgrading (Indus): Based on the previously introduced spatial vector angle method, the index of industrial structure upgrading is constructed by calculating the angle between the vector of the value-added outputs of the three major industries in each city and the corresponding spatial coordinate axes. This metric captures the advancement and transformation of industrial structures.

Urbanization (Urban): measured as the ratio of urban population to total population.

Government Support (Gov): represented by the natural logarithm of general public fiscal expenditures in each prefecture-level city.

Information Infrastructure Development (Infor): Referring to previous research (Li, 2023) [45], this study adopts the natural logarithm of telecommunications business revenue as a proxy for the level of informatization.

3.4. Data Sources

In terms of data sources, the Digital Inclusive Finance Index is derived from the Digital Finance Research Center of Peking University, while data on green invention patent authorizations are obtained from the CNRDS database. Additional data are primarily sourced from prefecture-level cities’ statistical bulletins on national economic and social development, The China City Statistical Yearbook, the National Bureau of Statistics of China, the China National Intellectual Property Administration, the Wind database, and prefecture-level city statistical yearbooks. Missing data have been addressed using interpolation methods. Descriptive statistics are presented in

Table 3. Descriptive statistical results of variables.

Variable			Observations	Mean
Dependent Variable	Happy	781	0.186	0.186
Independent Variable	Ai	781	2.833	1.621
	AiEnt	781	0.455	1.363
Control Variables	Echu	781	0.065	0.044
	InPgdp	781	1.540	0.545
	Indus	781	6.585	0.315
	Urban	781	0.555	0.138
	Gov	781	5.740	0.592
	Infor	781	0.255	0.175
Mediating Variables	Hum	781	0.172	0.190
	Emp	781	0.327	0.088
	Gre	781	0.050	0.131
Moderating Variables	Dig	781	1.445	0.653
	Fdi	781	2.672	1.546

.

4. Empirical Results

4.1. Baseline Regression

A multicollinearity analysis of all variables was conducted initially. The results indicate that all variance inflation factors (VIFs) are below 6, significantly lower than the critical threshold of 10, suggesting no severe multicollinearity issues. Consequently, further analysis can proceed. The Hausman test and fixed effects test confirm the existence of both individual and time effects in the model. Accordingly, a two-way fixed effects model is utilized to estimate the impact of artificial intelligence (AI) on social well-being in the Yellow River Basin. The corresponding regression results are displayed in

Table 4. The impact of artificial intelligence on social well-being in the Yellow River Basin.

	(1)	(2)	(3)	(4)	(5)
Ai	0.009 ***	0.005 ***	−0.014 ***	−0.011 ***	−0.011 ***
	(0.002)	(0.001)	(0.003)	(0.002)	(0.002)
Ai2			0.004 ***	0.003 ***	0.003 ***
			(0.000)	(0.000)	(0.000)
Edu		0.901 ***		0.866 ***	0.807 ***
		(0.060)		(0.054)	(0.058)
lnPgdp		−0.025 ***		−0.020 ***	−0.028 ***
		(0.008)		(0.007)	(0.008)
Indus		0.019 **		0.018 **	0.015 *
		(0.009)		(0.009)	(0.009)
Urban		−0.024			−0.022
		(0.032)			(0.031)
Gov		0.021 *			0.028 ***
		(0.011)			(0.011)
Infor		0.004			0.002
		(0.008)			(0.007)
City/Time Fixed Effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	0.123 ***	0.169 ***	0.159 ***	0.029	−0.080
	(0.005)	(0.006)	(0.004)	(0.056)	(0.072)
Observations	781	781	781	781	781
R	0.298	0.429	0.488	0.633	0.637

Notes: Robust standard errors are reported in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. Other fixed effects are controlled. Subsequent tables follow the same format.

.

Columns (1) and (2) in the regression results indicate that when the squared term of AI (Ai²) is not included—whether or not other city-level factors influencing social well-being are controlled for—the regression coefficients of AI (Ai) are 0.009 and 0.005, respectively, both significantly positive at the 1% level. This finding suggests that AI exerts a significant positive effect on social well-being. This suggests that AI significantly promotes social well-being. Furthermore, after adding the quadratic term (Ai²) to the model, the results in Columns (3)–(5) reveal that the coefficient of (Ai²) is significantly positive at the 1% level, while the coefficient of (Ai) becomes significantly negative at the 1% level. This indicates that AI exerts a significant diminishing-then-enhancing effect on social well-being, providing preliminary evidence of a U-shaped relationship between artificial intelligence and social well-being. The turning point is calculated to be 1.833, indicating that when AI (Ai) is less than 1.833, AI hinders the enhancement of social well-being, while values above this threshold suggest that AI significantly boosts social well-being. Hence, Hypothesis 1 is validated.

Examining the control variables in Column (5), it is evident that educational investment (Edu), industrial structure upgrading (Indus), and government support (Gov) significantly enhance social well-being in the Yellow River Basin. However, the coefficients of urbanization (Urban) and informatization (Infor) are not significant, and economic development level (lnPgdp) exhibits a significant negative effect.

The positive impact of educational investment (Edu) on social well-being can be attributed to its role in alleviating educational financial burdens for residents particularly through policies aimed at supporting rural and remote populations. These initiatives not only lower costs but also optimize resource distribution, encouraging the flow of quality education resources to underdeveloped regions, thereby promoting educational equity and enhancing social well-being (Yin et al., 2019) [46]. Industrial structure upgrading (Indus), as a cornerstone of socioeconomic progress, reshapes the labor market by creating higher-quality and more diversified employment opportunities. The continuous development of high-value-added industries optimizes the job market, improves living standards, and reduces income disparities. This fosters social equity, enhancing both social well-being and a sense of belonging (Pan et al., 2023) [47].

Lastly, government support (Gov) serves as the foundation for improving livelihoods and social harmony. By precisely addressing public needs and expanding the scale and efficiency of public services, government support strengthens public trust and satisfaction, laying a solid foundation for increasing social well-being.

Conversely, the negative impact of economic development level (lnPgdp) might stem from the relatively low economic development levels in the upper and middle reaches of the Yellow River Basin. These regions are often constrained by a single industrial structure and significant transformation pressures. Heavy reliance on resource extraction has led to environmental degradation and hindered industrial upgrading, thereby suppressing economic diversification and the improvement of social well-being.

4.2. Endogeneity Discussion

Given that the selected control variables may not fully mitigate endogeneity issues arising from omitted variables, an instrumental variable (IV) approach was employed to re-estimate the relationship between artificial intelligence (AI) and social well-being. Following the mainstream practices in academia, a Bartik instrument (Ai_Bartik) was constructed based on the share-shift method, utilizing the product of lagged AI and the first-order difference in AI over time. The underlying rationale is that the simulated values derived from the initial share composition of AI and the overall growth rate are highly correlated with the actual values while remaining uncorrelated with the error term, thus satisfying the requirements for an effective instrument. Consequently, the Bartik instrument and its squared term Ai_Bartik² (Ai_Bartik²) are used as instruments in a two-stage least squares (2SLS) regression. The detailed results are presented in

Table 5. Endogeneity test analysis.

	(1)	(2)	(3)
Ai			−0.017 ***
			(0.004)
Ai2			0.004 ***
			(0.001)
Ai_Bartik	0.483 ***	−0.833 *
	(0.083)	(0.503)
Ai_Bartik2	0.023 *	0.784 ***
	(0.013)	(0.082)
Control Variables	Controlled	Controlled	Controlled
City/Time Fixed Effects	Controlled	Controlled	Controlled
Unidentifiability Test	224.413 ***
Weak Instrument F-Statistic	203.353	239.891
R2	0.785	0.815	0.556

.

Columns (1) and (2) report the first-stage results of the 2SLS regression. The Bartik instrument passes the underidentification test (Kleibergen–Paap rk), and the F-statistics for the weak instrument test are 203.353 and 239.891, far exceeding the commonly accepted threshold of 10 in academic practice. This indicates that the model does not suffer from weak instrument issues, and the estimates are reliable.

In Column (3), the second-stage 2SLS results demonstrate high consistency with the baseline regression findings. Although the coefficients of (Ai) and (Ai²) exhibit minor fluctuations, their directional signs remain unchanged, and both are significant at the 1% level. This further corroborates the preliminary conclusion of a U-shaped relationship between AI and social well-being.

4.3. Robustness Discussion

The results of the instrumental variable (IV) regressions further validate the hypothesis of a U-shaped relationship between artificial intelligence (AI) and social well-being in the Yellow River Basin. To ensure the robustness of these findings, five additional tests were conducted: ① Lind and Mehlum Test: Following Lind and Mehlum (2010), the U-test command in STATA was employed to examine the U-shaped relationship) [48], (Lind and Mehlum, 2010). ② Lagged Core Variable: Given that the reverse impact of social well-being on AI development is relatively weak and that a plausible transmission mechanism is difficult to establish, we account for potential endogeneity concerns by incorporating the two-period lag of the key explanatory variable. This approach also considers the inherent delay in patent approvals, ensuring a more cautious assessment. The regression is then re-estimated accordingly. ③ Alternative Explanatory Variable: The core explanatory variable was replaced with the number of AI enterprises to enhance the robustness of the conclusions. ④ Winsorization: All variables were Winsorized at the 1% level to mitigate the influence of outliers. ⑤ Visualization: To intuitively illustrate the fitting effect of the quadratic regression model, we have plotted a nonlinear quadratic fit to strengthen the argument.

Table 6. Endogeneity test analysis.

	Lower Bound	Upper Bound	Overall
Extreme Point			2.048
Interval	0.000	7.235
Slope	−0.114	0.029
T-value	−5.300	9.312
p-value	0.000	0.000
Overall T-Value			5.300
Overall p-Value			0.000

presents the results of the U-test. The observed extreme value is 2.048, which falls within the interval range [0.000, 7.235], and the null hypothesis is rejected at the 1% statistical significance level. The slope data indicate that the impact of artificial intelligence on social well-being in the Yellow River Basin follows a U-shaped trajectory, first declining and then rising. Thus, this U-shaped relationship demonstrates a certain degree of robustness.

As shown in column (1) of

Table 7. Robustness test results.

	(1)	(2)	(3)
L2.Ai	−0.012 ***
	(0.002)
L2.Ai2	0.003 ***
	(0.000)
AiEnt		−0.015 *
		(0.006)
AiEnt2		0.004 ***
		(0.000)
Ai			−0.011 ***
			(0.002)
Ai2			0.003 ***
			(0.000)
Constant	−0.097	−0.068	−0.070
	(0.086)	(0.069)	(0.066)
Control Variables	Controlled	Controlled	Controlled
City/Time Fixed Effects	Controlled	Controlled	Controlled
Observations	639	781	781
R2	0.622	0.610	0.608

, both the linear and quadratic terms of the explanatory variable, lagged by two periods, are significant at the 1% level, confirming the U-shaped relationship between artificial intelligence and social well-being. The regression results in column (2) indicate that after replacing the core explanatory variable, the linear term of AI enterprises (AiEnt) is significantly negative at the 10% level, while the quadratic term (AiEnt²) passes the 1% significance test, similarly demonstrating the existence of a significant U-shaped relationship. Column (3) reports the regression results after Winsorization, where the variable signs and significance levels remain consistent with the previous findings.

Figure 1 presents the regression model’s fitting results. Specifically, at low levels of AI development (AI < 3), social well-being either slightly declines or remains stable as AI levels increase, suggesting that AI may have a short-term negative impact on social welfare. However, as AI development progresses further (AI > 3), social well-being begins to rise, indicating that AI significantly enhances social well-being in the long run.

5. Mechanism Analysis

After confirming the existence of a U-shaped relationship between artificial intelligence (AI) and societal well-being, a critical question arises: Does the development of AI, as theorized earlier, trigger multiple lag effects? This section seeks to explore these issues, clarify the underlying mechanisms, and enrich the depth and breadth of existing research.

5.1. Examination of Multiple Lag Effects

Building on the prior theoretical analysis, the rapid advancement of AI, driven by substantial national support and a series of favorable policies, has led to a revolutionary leap in technological domains within a short period. However, this surge of AI adoption has temporarily constrained human capital, green innovation, and employment opportunities, failing to meet the desired equilibrium. Consequently, multiple lag effects—where these elements fall behind societal needs—have emerged. To examine this phenomenon, the econometric model in Equation (9) is proposed:

$$Me{d}_{it}=\beta +{\beta }_{1}A{i}_{it}+{\beta }_{2}A{i}_{it}^2+\gamma Z_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(9)

In the equation, Med_it represents the mediating variables, including human capital level (Hum), employment opportunities (Emp), and green innovation (Gre). Other variables are consistent with those described earlier. The specific estimation results of the above model are reported in

Table 8. Estimation results of multiple lag test.

	(1)	(2)	(3)
	Hum	Emp	Gre
Ai	−0.018 ***	−0.004 ***	−0.023 ***
	(0.003)	(0.002)	(0.005)
Ai2	0.003 ***	0.001 ***	0.005 ***
	(0.001)	(0.000)	(0.001)
Constant	0.260 **	0.506 ***	0.153
	(0.105)	(0.053)	(0.178)
Control Variables	Controlled	Controlled	Controlled
City/Time Fixed Effects	Controlled	Controlled	Controlled
Observations	781	781	781
R2	0.371	0.811	0.566

.

Observing column (1), it can be noted that the regression coefficient of artificial intelligence (Ai) on human capital level (Hum) is significantly negative at the 1% level, while the coefficient of the squared term of artificial intelligence (Ai²) is 0.003, which is also significant at the 1% level. This indicates that artificial intelligence exerts a U-shaped effect on human capital, where its initial suppressive influence creates a gap between the actual level of human capital and societal needs. In column (2), the signs and significance levels of the regression coefficients for artificial intelligence (Ai) and its squared term (Ai²) remain consistent with those in column (1), demonstrating that artificial intelligence reduces employment opportunities (Emp) initially but eventually contributes to their increase. Column (3) reveals that both the coefficients of artificial intelligence (Ai) and its squared term (Ai²) are significant at the 1% level and have opposite signs, indicating that AI exerts a diminishing-then-enhancing effect on green innovation. This confirms the existence of a U-shaped relationship between artificial intelligence and green innovation (Gre), suggesting the presence of a short-term lag in the impact of AI on green innovation.

To ensure robustness, additional U-test validations were conducted with the results consistently supporting the findings (not presented here due to space constraints). In summary, the rapid development of artificial intelligence exhibits a U-shaped influence on human capital levels, employment opportunities, and green innovation, providing empirical evidence for the existence of multiple lags. Thus, Hypothesis 2 is validated.

5.2. Mediation Effect Analysis

After confirming the existence of multiple lags, another question arises: how do these lags induced by artificial intelligence (AI) influence social well-being in the Yellow River Basin? Drawing on existing research, the following econometric model was established for testing:

$$Happ{y}_{it}=\omega +{\omega }_{1}A{i}_{it}\times Me{d}_{it}+{\omega }_{2}A{i}_{it}^2\times Me{d}_{it}+{\omega }_{3}Me{d}_{it}+\gamma Z_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(10)

The variables in the equation are consistent with those described earlier and will not be elaborated upon. The interaction term is the core variable of interest in this section. Based on the baseline regression results, a significantly positive estimated coefficient for the interaction term indicates that the mediating variable reinforces the U-shaped relationship between AI and social well-being in the Yellow River Basin. This suggests that AI enhances its impact on social well-being through the mediating variable.

Table 9. Mechanism analysis.

	(1)	(2)	(3)
$Ai\times Hum$	−0.020 **
	(0.009)
$A{i}^2\times Hum$	0.006 ***
	(0.001)
Hum	−0.040
	(0.026)
$Ai\times Emp$		−0.021 ***
		(0.006)
$A{i}^2\times Emp$		0.006 ***
		(0.001)
Emp		0.376 ***
		(0.049)
$Ai\times Gre$			−0.026 **
			(0.013)
$A{i}^2\times Gre$			0.005 ***
			(0.002)
Gre			0.247 ***
			(0.025)
Constant	−0.048	−0.261 ***	−0.120 **
	(0.070)	(0.069)	(0.054)
Control Variables	Controlled	Controlled	Controlled
City/Time Fixed Effects	Controlled	Controlled	Controlled
Observations	781	781	781
R2	0.654	0.705	0.795

reports the regression results of the specific mechanism analysis. As seen in column (1), the coefficient of the interaction term is 0.006, which is significant at the 1% level. However, the regression coefficient of Hum is not statistically significant. Combined with the results of column (1) in Table 8, it can be inferred that human capital positively mediates the impact of AI on social well-being, serving as a channel between the two, thereby validating Hypothesis 3a. The results in column (2) show that the coefficient of the interaction term for employment opportunities is also 0.006, which is significant at the 1% level. Meanwhile, the regression coefficient of Emp is significantly positive at the 1% level with a specific value of 0.376. Together with the results in column (2) of Table 8, this confirms that the employment opportunity mechanism is significant, validating Hypothesis 3b. Column (3) shows that the interaction term’s coefficient for green innovation is 0.005, which is also significant at the 1% level. The impact coefficient of Gre on social well-being is significantly positive at the 1% level with a corresponding value of 0.247. Along with the results from column (3) in Table 8, it can be concluded that AI promotes the enhancement of social well-being through green innovation, thus validating Hypothesis 3c.

Additionally, the coefficients of the interaction terms between each mediating variable and the quadratic term of AI in columns (1)–(3) are larger than those in the baseline regression. This indicates that the influence of AI on social well-being in the Yellow River Basin has been strengthened. It is also worth noting that in terms of the strength of the mediating effects, employment opportunities exhibit the strongest mediating effect, followed by green innovation, while human capital plays the weakest mediating role. This suggests that employment opportunities serve as the primary channel through which artificial intelligence influences social well-being. Therefore, appropriately enhancing the development of these mediating variables can help amplify AI’s contribution to improving social well-being.

5.3. Moderating Effect Analysis

To examine whether digital financial inclusion (Dig) and foreign direct investment (FDI) play moderating roles in the relationship between artificial intelligence (AI) and social well-being in the Yellow River Basin, a moderating effect model was constructed as follows:

$$Happ{y}_{it}=\lambda +{\lambda }_{1}A{i}_{it}+{\lambda }_{2}A{i}_{it}^2+{\lambda }_{3}A{i}_{it}\times Mo{d}_{it}+{\lambda }_{4}A{i}_{it}^2\times Mo{d}_{it}+{\lambda }_{5}Mo{d}_{it}+\tau Z_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(11)

In this equation, Mod represents the moderating variable, while the other variables are consistent with previous specifications. The coefficient λ₄, corresponding to the interaction term between the quadratic term of AI and the moderating variable, is of particular interest as it captures the moderating effect. A significantly positive λ₄ would indicate that the moderating variable positively influences the relationship between AI and social well-being in the Yellow River Basin.

The corresponding test results are reported in

Table 10. Analysis of moderating effects.

	(1)	(2)
Ai	−0.010 ***	−0.008 ***
	(0.002)	(0.002)
Ai2	0.002 ***	0.002 ***
	(0.000)	(0.000)
$Ai\times Dig$	−0.006 **
	(0.002)
$A{i}^2\times Dig$	0.002 ***
	(0.000)
Dig	0.000
	(0.004)
$Ai\times Fdi$		−0.006 ***
		(0.001)
$A{i}^2\times Fdi$		0.001 ***
		(0.000)
Fdi		0.001
		(0.001)
Constant	−0.108	−0.071
	(0.070)	(0.069)
Control Variables	Controlled	Controlled
City/Time Fixed Effects	Controlled	Controlled
Observations	781	781
R2	0.659	0.670

. First, observing columns (1) and (2), both the linear term of AI (Ai) and its quadratic term (Ai²) are significant at the 1% level with coefficients slightly smaller than those in the baseline regression. This suggests that the U-shaped relationship between AI and social well-being remains robust. Specifically, in column (1), the coefficients of Ai × Dig and Ai² × Dig are both significant at least at the 5% level, indicating the presence of a moderating effect. Among them, the estimated coefficient for the interaction term λ₄ is 0.002 and is significant at the 1% level, indicating that digital financial inclusion positively moderates the effect of AI on social well-being, thereby validating Hypothesis 4a. Column (2) shows that the interaction term Ai² × FDI has a coefficient of 0.001, which is also significant at the 1% level. This demonstrates that foreign direct investment similarly exerts a positive moderating effect, supporting Hypothesis 4b.

In conclusion, these findings provide robust evidence that AI positively impacts social well-being in the Yellow River Basin through the moderating effects of digital financial inclusion and foreign direct investment. These results have critical implications for policy formulation aimed at enhancing public welfare and improving citizens’ sense of happiness and well-being.

6. Further Analysis

6.1. Analysis of Spatial Spillover Effects

Prior to performing spatial econometric analysis, Moran’s I test was employed to assess the spatial correlation of the variables.

Table 11. Spatial correlation test.

Happy			Ai
Year	Moran’ I	Z Value	Year	Moran’ I	Z Value
2012	0.273 ***	3.693	2012	0.435 ***	5.674
2013	0.285 ***	3.863	2013	0.438 ***	5.706
2014	0.288 ***	3.906	2014	0.436 ***	5.679
2015	0.219 ***	3.045	2015	0.439 ***	5.728
2016	0.222 ***	3.115	2016	0.433 ***	5.652
2017	0.183 ***	2.613	2017	0.438 ***	5.708
2018	0.180 ***	2.585	2018	0.251 ***	3.358
2019	0.148 **	2.147	2019	0.357 ***	4.691
2020	0.163 ***	2.375	2020	0.266 ***	3.545
2021	0.149 **	2.179	2021	0.405 ***	5.294
2022	0.136 **	2.012	2022	0.405 ***	5.301

presents the spatial correlation test results based on the adjacency weight matrix. The research findings indicate that during the sample period, the Moran’s I values for both social well-being and artificial intelligence are significantly positive, suggesting a strong positive spatial correlation. This confirms the feasibility of conducting further spatial analysis. In addition, spatial correlation tests were also performed using the geographic distance matrix and the economic–geographic matrix, along with a series of model selection tests, including Lagrange Multiplier (LM) tests. For brevity, these results are not reported here. The diagnostic tests indicate that the spatial error model (SEM) is the most appropriate for this analysis. Unlike other models, the spatial error model assumes that spatial effects are embedded in the stochastic error terms, capturing the extent to which shocks to the social well-being of adjacent cities influence the social well-being of a given city. Consequently, this study employs a dual fixed-effects SEM to conduct an in-depth examination of the spatial spillover effects.

Table 12. Spatial regression results.

	Adjacency Weight Matrix	Geographical Weight Matrix	Economic—Geographical Weight Matrix
	(1)	(2)	(3)
Ai	−0.009 ***	−0.010 ***	−0.010 ***
	(0.002)	(0.002)	(0.002)
Ai2	0.002 ***	0.002 ***	0.002 ***
	(0.000)	(0.000)	(0.000)
λ	0.229 ***	0.329 **	0.270 ***
	(0.055)	(0.160)	(0.090)
σ2	0.000 ***	0.000 ***	0.000 ***
	(0.000)	(0.000)	(0.000)
Control Variables	Controlled	Controlled	Controlled
City/Tim e Fixed Effects	Controlled	Controlled	Controlled
Observations	781	781	781
R2	0.777	0.776	0.778

presents the SEM estimation results under three different spatial weight matrices: adjacency, geographic distance, and economic–geographic.

Firstly, as shown in columns (1)–(3), the coefficients of the linear and quadratic terms of artificial intelligence (AI) on social well-being are highly significant at the 1% level, confirming the robustness of the baseline regression results. Specifically, in column (1), under the adjacency weight matrix, the coefficient of the spatial error term λ is 0.229, which is significant at the 1% level, confirming the objective existence of spatial effects in the random error terms of the spatial error model. Meanwhile, the R² value is 0.777, indicating a good overall fit of the model. These results indicate that in the long term, AI exerts positive spatial externalities on social well-being, and the spatial effects of social well-being are objectively present and exhibit spillover trends. The results reported in columns (2) and (3), based on the geographic distance and economic–geographic weight matrices, further validate the presence of spatial spillover effects. It can be observed that the regression coefficients of artificial intelligence remain significant at the 1% level with no substantial changes in their values. Meanwhile, the coefficients of the spatial error term are all significant at least at the 5% level, once again confirming the objective existence of spatial aggregation in social well-being. This outcome can be attributed to the integration of AI, as a new productive force, with the real economy, which not only promotes industrial upgrading within the Yellow River Basin but also injects substantial momentum into the region’s high-quality development. Moreover, the sustainable development of both the economy and society improves the quality of life while enhancing individuals’ sense of belonging, happiness, and fulfillment. The continuous improvement in social well-being gradually spreads to surrounding neighboring regions, creating a positive spillover effect that further amplifies the overall level of social well-being in the Yellow River Basin.

6.2. Conditional β-Convergence with Moderation Mechanism: Examination and Analysis

Previous sections confirmed the positive moderating effects of digital financial inclusion and foreign direct investment (FDI) on the relationship between artificial intelligence (AI) and social well-being in the Yellow River Basin. This section explores whether these moderating mechanisms persist in the convergence process of social well-being. Specifically, we extend the traditional conditional β-convergence framework by incorporating interaction terms between moderating variables and the initial level of social well-being as specified in the following model. Following existing studies, all interaction terms have been mean-centered.

$${\displaystyle \frac{\ln (Happ{y}_{it+T}/Happ{y}_{it})}{T}}=\alpha +{\beta }_1\ln Happ{y}_{it}+{\beta }_2\ln Happ{y}_{it}\times Mo{d}_{it}+\eta Mo{d}_{it}+\vartheta Z_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(12)

In the equation, ${\displaystyle \frac{\ln (Happ{y}_{it+T}/Happ{y}_{it})}{T}}$ represents the growth rate of social well-being in various prefecture-level cities during the (t + T) period with T set to 1 based on existing studies; ln Happy denotes the initial level of social well-being in the city; Mod represents the moderating variable (Mod > 0), while other variables remain consistent with the previous analysis. The sign of the convergence coefficient (β₁) determines whether social well-being exhibits a convergence trend. If β₁ < 0, it indicates an inverse relationship between the growth rate of social well-being and its initial level among cities in the Yellow River Basin, implying that regions with a lower initial level experience faster growth, leading to an overall convergence in social well-being. Furthermore, to better elucidate the role of the moderating variable in the convergence of social well-being, the partial derivative of the equation is taken, resulting in Equation (13).

$${\displaystyle \frac{\mathsf{\partial }\left[\frac{\ln (Happ{y}_{it+T}/Happ{y}_{it})}{T}\right]}{\mathsf{\partial }\ln Happ{y}_{it}}}={\beta }_1+{\beta }_{2}Mo{d}_{it}$$

(13)

It is evident that the growth rate of social well-being in the Yellow River Basin is jointly influenced by the convergence coefficient β₁ and the coefficient β₂ of $\ln Happy\times Mod$. Under the condition of convergence (β₁ < 0), if β₂ < 0, it indicates that the moderating variable accelerates the convergence process of regional social well-being. Conversely, if β₂ > 0, it suggests that the moderating variable slows down the convergence process. Therefore, the coefficient β₂ of $\ln Happy\times Mod$ is the focal point of analysis in this section.

Table 13. Conditional β convergence extends the regression results of the model.

	(1)	(2)	(3)	(4)
$\ln Happy$	−0.441 ***	−0.622 ***	−0.795 ***	−0.997 ***
	(0.034)	(0.037)	(0.061)	(0.062)
$\ln Happy\times Dig$			0.090 ***
			(0.026)
Dig			0.147 ***
			(0.043)
$\ln Happy\times Fdi$				0.085 ***
				(0.012)
Fdi				0.152 ***
				(0.021)
Edu		0.947 ***	0.730 ***	0.567 ***
		(0.174)	(0.183)	(0.175)
lnPgdp		−0.060 ***	−0.055 **	−0.054 **
		(0.023)	(0.023)	(0.022)
Indus		0.109 ***	0.111 ***	0.095 ***
		(0.027)	(0.027)	(0.026)
Urban		−0.156	−0.103	−0.141
		(0.124)	(0.124)	(0.119)
Gov		0.069 **	0.065 **	0.076 **
		(0.031)	(0.031)	(0.030)
Infor		0.868 ***	0.819 ***	0.813 ***
		(0.156)	(0.156)	(0.150)
Constant	−0.790 ***	−2.344 ***	−2.654 ***	−2.967 ***
	(0.064)	(0.261)	(0.274)	(0.265)
City/Time Fixed Effects	Controlled	Controlled	Controlled	Controlled
Observations	710	710	710	710
R2	0.250	0.355	0.368	0.408

reports the detailed regression results. It can be observed that in column (1), the regression coefficient of lnHappy is −0.441 and is significant at the 1% level. After incorporating a series of control variables, the regression coefficient of lnHappy in column (2) remains significantly negative. This indicates that cities with higher initial levels of social well-being exhibit slower growth rates, while those with lower initial levels experience faster growth, suggesting a convergence trend in social well-being across regions. This demonstrates a significant catch-up effect among cities and suggests a conditional β-convergence trend in social well-being across the Yellow River Basin. Furthermore, columns (3) and (4) present the regression results of the extended conditional β-convergence model incorporating moderating mechanisms. The findings indicate that the convergence coefficients remain significantly negative, while the coefficients of $\ln Happy\times Dig$ and $\ln Happy\times Fdi$ are significantly positive at the 1% level. This suggests that both digital inclusiveness and foreign investment exert an opposing moderating effect on the convergence of social well-being, thereby slowing the convergence process. This may imply that despite the catch-up effect in improving social well-being across cities, intertwined factors such as the “digital divide” and “foreign capital withdrawal” generate a damping effect on the convergence process. Such findings warrant attention, emphasizing the need for governments to prioritize these issues in future economic development strategies and policy formulations.

7. Conclusion and Policy Implications

7.1. Conclusion and Recommendations

Based on panel data from 71 prefecture-level cities in the Yellow River Basin from 2012 to 2022, this study conducts an in-depth investigation into the impact of artificial intelligence on social well-being in the region. Although this study is based on data from China’s Yellow River Basin, its findings may be applicable to a broader range of developing countries or regions with uneven economic development. For example, the promotion of AI in Brazil, South Africa, or India faces similar challenges related to infrastructure, skills training, and policy regulation. Therefore, this research holds significant implications for enhancing people’s sense of gain, satisfaction, and happiness, thereby advancing the goal of common prosperity. The key findings are as follows:

(1)

There is a U-shaped relationship between AI and social well-being in the Yellow River Basin, where AI initially decreases but subsequently increases social well-being.

(2)

In the short term, AI introduces multiple frictions, influencing social well-being through three transmission mechanisms: employment opportunities, human capital development, and green innovation.

(3)

Both digital inclusiveness and foreign investment positively moderate the relationship between AI and social well-being, amplifying AI’s impact on social well-being.

(4)

The spatial effects of AI on social well-being are evident and exhibit spillover trends.

(5)

The social well-being landscape in the Yellow River Basin demonstrates a convergence pattern, although both digital inclusiveness and foreign investment exert negative moderating effects, slowing the overall convergence process.

Based on these findings, policy recommendations are proposed to address these dynamics effectively.

(1)

Formulate Forward-Looking Legal Frameworks and Enhance Fiscal Support.

Drawing on the finding that AI exerts a U-shaped impact on social well-being—generating multiple short-term frictions—policymakers must establish forward-looking and inclusive legal frameworks that ensure technological development aligns with ethical, security, and social considerations. Such measures should embody the principle of “technology for good”. Given that existing legal systems do not fully regulate all facets of AI applications, rapid technological advancement may spawn new ethical dilemmas and social risks. Thus, the government should expedite the establishment of relevant laws and regulations that clearly define the boundaries of technology usage. These should safeguard privacy, security, and fairness, particularly in areas susceptible to labor substitution and data security breaches. By doing so, a multi-layered governance structure—encompassing laws, ethics, policies, and technical standards—can balance technological progress with labor rights, ensuring equitable access to employment opportunities.

Moreover, to fully leverage the spatial spillover effects of AI, the government should increase investment in “new infrastructure” projects related to AI and cloud computing. Such investment can foster the circulation of digital factors, diminish administrative fragmentation, and bridge the digital divide, thereby extending the spatial radius of social well-being spillovers.

Additionally, considering that AI is characterized by substantial investment requirements, capital intensity, and long project cycles, small and medium-sized enterprises often struggle to bear its high costs. Policymakers should thus not only strengthen fiscal support but also create diversified investment mechanisms that pool resources from various stakeholders. This holistic approach would support cutting-edge research and development (R&D), technology promotion, and market application of AI. At the same time, accelerating the establishment of regional collaborative innovation platforms and reinforcing industry-university-research integration will expedite the commercialization of AI achievements, enabling AI-driven social well-being to rebound more swiftly.

(2)

Strengthen Vocational Training and Promote Age-Friendly Products.

The advent of AI inevitably disrupts labor markets, accelerating the segmentation between traditional and emerging sectors. Against the backdrop of a slowing macroeconomic environment, it is imperative to implement enterprise support and employment stabilization measures along with effective unemployment relief programs. These initiatives should be complemented by practical economic stimulus policies to safeguard economic and social stability, boost confidence and expectations, and mitigate the frictional effects introduced by AI.

Enhancing vocational education and training is crucial to equipping the workforce with the necessary skills to adapt to technological transformation. Governments at various levels in the Yellow River Basin should tailor broad-based vocational training systems to their unique locational characteristics and resource endowments, guiding citizens to rapidly improve their AI literacy. For instance, the upper and middle reaches of the Yellow River, endowed with agricultural–pastoral advantages and abundant resources, can leverage industry–university research platforms to offer specialized training programs in “smart agriculture and animal husbandry”, “intelligent mining management”, and “smart inspection”. Such initiatives, including career counseling and re-employment services, help workers acquire emerging skills—such as intelligent production and data analytics—in alignment with evolving market demands.

The education sector must also design curricula and training initiatives that align with actual market needs, revamping educational methods and reforming academic disciplines to guide talent development toward an intelligent era. This approach optimizes human capital to meet the demands of AI-driven technological progress.

Simultaneously, promoting age-friendly AI products is vital given the “digital divide” faced by elderly populations. Currently, many smart devices and services do not fully accommodate the practical needs of older adults. Governments should encourage enterprises to develop intuitive and user-friendly AI solutions—such as smart health monitoring devices and voice assistants—enabling the elderly to enjoy the convenience of technological advancements. Considering the relatively low levels of digital infrastructure in the upper and middle reaches of the Yellow River and the linguistic diversity among elderly communities, firms must enhance the environmental adaptability and natural language processing capabilities of their products. For local governments, it is essential to provide basic digital skills training for middle-aged and elderly populations, women, and rural laborers through village broadcasting, mobile applications, and public training programs. Additionally, “AI + entrepreneurship” technology services should be offered to groups at risk of technological displacement, such as traditional artisans, farmers, and miners in resource-based cities, encouraging them to explore new business models and economic opportunities. This will better serve elderly users and ultimately elevate overall social well-being.

(3)

Enhance Collaboration and Promote Sustainable Development.

Robust AI development necessitates stronger interregional collaboration and knowledge exchange. However, “going it alone” remains prevalent in key western areas of the Yellow River Basin, such as the upper and middle reaches. To address this, western regions should foster partnerships with more developed eastern counterparts by establishing innovation funds and collaborative R&D platforms to facilitate resource complementarities. Simultaneously, governments at all levels can employ preferential policies to encourage and support cross-regional scientific and technological cooperation, attracting research institutes and enterprises from both domestic and international sources. This approach will bolster the growth of big data, AI, and other digital technology clusters, thereby expanding interregional AI application networks. As a result, it will catalyze industrial transformation and upgrading in the Yellow River Basin, realize green, low-carbon, and coordinated regional economic development, and leverage positive spatial spillover effects to enhance social well-being in surrounding areas. Over time, this virtuous cycle—“technology sharing—improvement in social well-being—enhanced spatial spillover effects”—will become self-reinforcing.

In this process, the careful regulation of foreign investment thresholds is necessary, coupled with a focus on attracting high-tech enterprises. Additionally, governments should steer policy incentives to promote the flourishing of emerging industries such as green energy and tourism. By designing appropriate environmental regulations that guide enterprises toward green innovation and reduce reliance on traditional resources, the Yellow River Basin can achieve sustainable high-quality development, ultimately underpinning improvements in social well-being.

7.2. Limitations

Although this study provides a relatively comprehensive discussion, certain limitations remain. First, our analysis primarily relies on empirical methods to examine the impact of artificial intelligence on social well-being in the Yellow River Basin. Incorporating considerations of AI-induced ethical concerns, policy challenges, and other societal impacts could contribute to a more holistic understanding of the topic. Second, due to data availability constraints, this study selects 71 cities in China’s Yellow River Basin as the research sample and employs multiple robustness checks in the empirical analysis. Future research could expand the dataset to include a broader range of cities, thereby mitigating potential regional selection bias. Additionally, incorporating subjective well-being indicators or survey data in robustness checks could provide a more comprehensive assessment of AI’s impact on social well-being, enhancing the study’s external validity and policy relevance.