The Innovative Construction of Provinces, Regional Artificial Intelligence Development, and the Resilience of Regional Innovation Ecosystems: Quasi-Natural Experiments Based on Spatial Difference-in-Differences Models and Double Machine Learning

Abstract

Based on the theory of regional innovation niches, this study calculates the resilience of regional innovation ecosystems and constructs a comprehensive evaluation index system for regional artificial intelligence development, resulting in a panel dataset for 30 provinces in China from 2009 to 2021 (excluding Tibet, Hong Kong, Macao, and Taiwan). Within the framework of the construction of innovative provinces, regional artificial intelligence, and the resilience of regional innovation ecosystems, spatial double-difference and double machine learning models are employed for a quasi-natural experiment. The main research conclusions are as follows: (1) Both the construction of innovative provinces and artificial intelligence have a significant positive impact on the resilience of regional innovation ecosystems. (2) However, regional artificial intelligence exhibits a negative spatial spillover effect on the resilience of regional innovation ecosystems. (3) The construction of innovative provinces can positively moderate the effect of artificial intelligence on the resilience of regional innovation ecosystems. (4) Through the promotion of regional artificial intelligence, the construction of innovative provinces can indirectly enhance the diversity, evolutionary potential, buffering capacity, fluidity, and coordination of regional innovation ecosystems, thereby driving a leap in resilience. (5) The mechanisms by which the construction of innovative provinces stimulates regional intelligent input, application, innovation, and market dynamics to further enhance the resilience of regional innovation ecosystems are effective not only in the treatment group but also in the control group.

1. Introduction

In the context of globalization, all countries are facing the challenges of rapid technological development and intensifying international competition and are adapting to new trends in global technological development through a series of innovative policies. For example, China is strongly implementing policies highlighted at the 19th National Congress of the Communist Party of China, which emphasized the need to adhere to the strategies of strengthening the country through science, education, and talent and driving development through innovation. These efforts aim to achieve a high level of self-reliance in science and technology and, in the medium-to-long term, build an innovative country. The European Union has put forward the “Horizon Europe” plan (The “Horizon Europe” plan is a significant funding initiative established by the European Union to maintain its leading position in the global science and innovation landscape while addressing global challenges such as climate change, biodiversity loss, and population aging. This program was officially approved by the European Parliament and the Council of the European Union on December 11, 2020, with an implementation period spanning from 2021 to 2027), aiming to promote Europe’s leading position in scientific research in response to global challenges through “three pillars” (The “Three pillars” refer to the three main components that make up the "Horizon Europe" program: Excellence in Science, Global Challenges and Industrial Resilience, and Innovation Europe).

From the perspective of regional innovation, a regional innovation ecosystem is a regional economic system that dynamically exchanges matter, energy, and information among various innovation entities (producers, consumers, and decomposers), elements, and environments [1]. The resilience of regional innovation ecosystems is a key variable in the construction of an innovative country. It examines the development level and evolutionary stage of the innovation entities and environments in regional innovation ecosystems, as well as the synergistic evolution trend of innovation entities and resource environments. Liu and Ji argue that the capacity, vitality, and driving force of regional innovation entities, the level of aggregation of innovation resource elements, the degree of construction of innovation support platforms, and the suitability of the innovation environment directly determine the strength of the resilience of regional innovation ecosystems [2].

Taking China as an example, in the current context of accelerating the establishment of a new national system to achieve high-level self-reliance in science and technology, the significant adjustment of the structure of the innovation ecosystem and a substantial increase in its resilience driven by the new system mechanism and policy adjustments are essential to the construction of an innovative country and the high-quality development of regions [3]. Driven by strategy, the construction of innovative provinces at the provincial level has become an important strategic support for regional innovation development and a key link in promoting the construction of an innovative country in a coordinated manner.

At the national level, the policy for constructing innovation provinces emphasizes the transformation of economic development models as the central theme, with a focus on deepening reforms in the scientific and technological system. The primary objectives include promoting the conversion of scientific and technological achievements, nurturing new economic growth points and hubs, and facilitating the rapid development of high-tech industries and strategic emerging industries, as well as the emergence of innovative enterprises and cities, alongside the establishment of regional innovation systems. The policy defines innovation provinces as those that prioritize innovation as the core driving force for socio-economic development. They achieve this by increasing investment in innovation, enhancing the contribution rate of technological advancement, and boosting innovative output, thereby establishing sustained innovation capabilities and competitive advantages. Typical innovation provinces, such as Zhejiang and Guangdong in China, typically exhibit high levels of innovation investment, substantial contributions from technological progress, abundant innovative outputs, well-developed innovation systems, and a concentration of innovative talent.

Since 2013, the National Government of China has approved pilot projects for the construction of innovative provinces in 11 provinces, including Jiangsu. At the same time, the Ministry of Science and Technology of China issued the “Guidelines for the Construction of Innovative Provinces” in 2016, clearly stating that the relevant pilot areas should be developed into regional innovation centers, important sources of scientific and technological innovation, and strategic sources of emerging industries. Since the implementation of the policy pilot for the construction of innovative provinces, the pilot areas have frequently made efforts in top-level design, institutional arrangements, the construction of innovation infrastructure, and the supply of innovation resources, demonstrating strong policy effects. Therefore, against this background, it is of great significance to discuss the impact mechanism of the construction of innovative provinces on the resilience of regional innovation ecosystems via a quasi-natural experiment.

In addition, under the guidance of the “Guidelines”, each innovative pilot province in China has issued provincial-level guidelines for the construction of innovative provinces and formulated a large number of supporting policies and institutional arrangements. From various local policy documents, it can be seen that most of the innovative pilot provinces will promote regional artificial intelligence as a key focus of the construction of innovative provinces. For example, based on the strategic inherent logic of the “Eight-Eight Strategy” and the construction of innovative provinces, Zhejiang proposes to lead the construction of innovative provinces through the promotion of an artificial intelligence economy and intelligent development, focusing on areas such as intelligent video processing, high-performance special chips, intelligent systems and supercomputing, big data and cloud computing, and new architectures for industrial Internet, thus fostering new formats and models and promoting the transformation and upgrading of innovation chains. In Jiangsu’s provincial “14th Five-Year Plan for Scientific and Technological Innovation”, it is also clear that accelerating breakthroughs and development in new-generation information technologies represented by integrated circuits, artificial intelligence, intelligent communication, high-end software, and blockchain is important for the construction of innovative provinces. These policy documents indicate that the construction of innovative provinces will promote regional artificial intelligence as an important anchor point for policy implementation.

As China’s economy and society are undergoing profound changes in the wave of the artificial intelligence economy, creating new advantages in the artificial intelligence economy, guiding the transformation of industries towards artificial intelligence and vigorously promoting the development of intelligent manufacturing and intelligent services have become key aspects of regional development. From a content perspective, regional artificial intelligence is based on the latest achievements of new-generation information technologies, and through new technologies such as big data, artificial intelligence, virtual reality, and large language models, it empowers the transformation of economic growth models, the transformation and upgrading of industrial organizations, and profound social changes [4]. Overall, the transition from artificial intelligence to artificial intelligence is not only reflected in the explosive upgrade of information technology but also in the all-around transformation of factor mechanisms and production organization models, which has a disruptive impact on the overall development of regions. In this context, it is particularly important to discuss whether the process of artificial intelligence driven by the construction of innovative provinces can produce economic effects that empower the upgrading of innovation entities, optimize the allocation of innovation resources, and improve the innovation environment in regional innovation ecosystems, thus making the impact mechanism of artificial intelligence regarding the resilience of regional innovation ecosystems another important research theme.

The construction of innovative provinces and the development of regional artificial intelligence not only have a high degree of synergy and complementarity but also have a profound impact on regional innovation ecosystems and their resilience. Therefore, it is necessary to incorporate the construction of innovative provinces, artificial intelligence, and the resilience of regional innovation ecosystems into a unified research framework and explore the role and mechanism of artificial intelligence in enabling an increase in the resilience of regional innovation ecosystems under the influence of the construction of innovative provinces.

In summary, this study aims to investigate the interrelationships among the innovation province construction policy implemented in China, the level of regional AI development, and the resilience of regional innovation ecosystems, using the provinces of mainland China as the focal point of analysis.

The remainder of this study is organized as follows: Section 2 provides a comprehensive review of the existing literature and research; Section 3 presents the research hypotheses, constructs a difference-in-differences and dual machine learning model, and explains the variables involved; Section 4 conducts an empirical analysis based on data from 30 provinces in mainland China; and Section 5 summarizes the key research conclusions and offers policy recommendations.

The anticipated contributions of this study are as follows: The research findings aim to integrate innovation province policies, regional AI development, and the resilience of regional innovation ecosystems into a cohesive analytical framework. Through empirical analysis, the study seeks to evaluate the effectiveness of China’s implementation of regional innovation policies while also aspiring to extend the identified mechanism transmission pathways to other emerging market nations.

2. Literature Review

2.1. Theoretical Evolution and Research Dynamics of the Regional Innovation Ecosystem

For a long time, the regional innovation ecosystem has been regarded as a crucial driving force for supporting the sustainable development of the national innovation system and the construction of an innovative country [5], and the development level, dynamic evolution, and connectivity density of its basic constituent elements (including the regional innovation subjects, innovation resources, and innovation environment) are closely linked to the region’s achievement of innovative development and high-level technological self-reliance [6,7,8]. From the perspective of the regional innovation niche, the evaluation of and empirical research on the suitability, resilience, symbiosis, health, and operational efficiency of regional innovation ecosystem niches have become a hot research topic in academic circles at home and abroad [9,10,11], and these research perspectives have provided solid theoretical support and specific quantitative standards for exploring the overall development process of regional innovation ecosystems.

Most existing studies have focused on a single constituent element or a single research perspective of the regional innovation ecosystem, with few studies focusing on the overall resilience of the regional innovation ecosystem and its driving factors. In this field, Liu and Ji designed a competitive evaluation system for the regional innovation ecosystem with three hierarchical dimensions, including urban innovation subjects, intra-city innovation ecosystem construction, and inter-city innovation ecosystem embedding [2]. Meanwhile, Ding and Yang introduced the “state” and “potential” of the innovation niche in the evaluation of resilience and considered the state and potential of the innovation niche as the core dimensions forming the system’s resilience [12].

2.2. Research Review of the Development and Evolution of Regional Innovation Ecosystems under a Policy-Driven Framework

In the existing research on regional innovation ecosystems, many scholars have paid attention to the dominant impact of innovation policies on the formation of regional innovation ecosystems and have explored the evolutionary characteristics and operational laws of the various dimensions of regional innovation ecosystems under different policy drivers. Liang Lin et al. drew upon a panel dataset containing 31 Chinese provinces from 2011 to 2020 to construct an econometric model for empirical analysis, providing a comprehensive evaluation of the formation and evolution of regional innovation subjects [13]. Cao et al. used the method of comparative analysis to discuss the characteristics of tropical rainforests as the most stable and efficient natural ecosystem and deeply explored the essential features of an efficient regional innovation ecosystem, including an innovative soft and hard environment, multi-level and differentiated producers, consumers and decomposers of innovative activities, and a virtuous cycle of innovation resources [14]. Meanwhile, Tao et al. used Heilongjiang Province as a case study and established a regional military–civilian technological innovation ecosystem with a focus on defense industry enterprises, private enterprises, and local governments as key innovation entities, incorporating the precise implementation and effectiveness of the “military–civilian integration” policy into the perspective of the innovation niche [15].

2.3. Research Review of the Development Paths of Regional Innovation Ecosystems in the Artificial Intelligence Context

With the rise of new information technology and the promotion of regional artificial intelligence, regional innovation ecosystems have demonstrated a clear adaptability to artificial intelligence contexts and a path of artificial intelligence development during the evolutionary process. This has led to the concentration of nurturing artificial intelligence new productive forces and the emergence of the “data elements x” effect as a characteristic of the era in which regional innovation ecosystems serve as a crucial driving force for the construction of an innovative country. Xu Dan et al. built a framework for analyzing regional innovation ecosystems in terms of five aspects: innovation actors, government, innovation carriers, innovation resources, and innovation environment. Through fsQCA analysis, it has been demonstrated that four configurations can drive highly regional innovation: the “innovation participants-government-operators-resources” type; the “innovation actors-government-carriers-personnel-environment” type; the “knowledge innovation actors-government-carriers-resources-environment” type; and the “technology innovation participants-government-operators-resources-environment” type [16]. Liu Fan et al. believe that artificial intelligence transformation can improve the ecological suitability of regional innovation ecosystems by driving industrial integration and improving institutional environments [17]. Additionally, based on a longitudinal case study of Zhongguancun, Wang et al. analyzed the organizational and evolutionary mechanisms of regional innovation ecosystems in the context of artificial intelligence [18].

2.4. Literature Summary and Evaluation

The aforementioned literature has formed a research system focusing on policy orientation, regional artificial intelligence as a context and background, and the regional innovation ecosystem as the core, encompassing the theoretical evolution and research dynamics of regional innovation ecosystems, the driving role of regional policies in regional innovation ecosystems, and the impact mechanism of artificial intelligence smart scenarios on regional innovation ecosystems.

However, from the existing research perspective, further exploration is needed for the following aspects:

(1)

Existing studies have provided useful insights for exploring and evaluating the resilience of regional innovation ecosystems based on different sub-dimensions, evaluation methods, and theoretical frameworks. However, empirical research on the overall resilience of regional innovation ecosystems and their driving factors is still not sufficient. There is a lack of literature that can introduce external factors such as innovation policies, regional development models, and development levels into the evaluation of regional innovation ecosystem resilience and conduct an in-depth analysis of their driving factors to explore the dynamic impact mechanisms of external variables on the resilience of regional innovation ecosystems.

(2)

While existing literature has considered policy factors in the construction and evolution of regional innovation ecosystems, there are still certain limitations. First, existing studies have not directly taken the resilience of regional innovation ecosystems as a core variable and explored the deep-level impact of policies on it. Second, pilot policies such as “comprehensive innovation reform” can provide a good quasi-natural experimental opportunity to explore China’s reform process, policy paths, and reform directions in the regional innovation field. However, there is currently a lack of empirical research in this area. Additionally, as regional innovation ecosystems are crucial regional fulcrums for the construction of an innovative country, and pilot innovative province construction is a direct exploration of the construction of an innovative country at the regional level, it is essential to explore the formation and strengthening of the resilience of regional innovation ecosystems through the construction of innovative province pilots.

(3)

The existing research has demonstrated the close connection between the development and evolution of the artificial intelligence economy and even the regional smart context with various dimensions of regional innovation ecosystems. However, further research is still needed in the following areas: first, the conceptual definition, theoretical basis, and research framework of the impact mechanism of regional artificial intelligence regarding regional innovation ecosystems need to be further defined. Second, as a regional characteristic development indicator, the development process of regional artificial intelligence will inevitably be subject to policy intervention. Therefore, it is essential to explore the impact mechanism of artificial intelligence regarding the resilience of regional innovation ecosystems under policy-driven approaches.

Based on the above review and commentary on existing research, this study integrates the policies of innovative province construction pilots, artificial intelligence, and the resilience of regional innovation ecosystems into a unified research framework. A quasi-natural experiment is designed for empirical research, aiming to explore the joint impact mechanism of innovative province construction and artificial intelligence regarding the development and evolution of the resilience of regional innovation ecosystems. This not only extends the specific paradigm of studying the co-evolution and development of policy-driven regional innovation ecosystems based on causal inference methods but also provides theoretical guidance and policy references for policy-driven regional artificial intelligence development and the facilitation of the construction of an innovative country.

3. Mechanism Hypothesis and Model Construction

3.1. Innovative Province Construction and Regional Innovation Ecosystem Resilience

From the perspective of the “multidimensional hypertrophic niche” theory, innovative province construction pilots, as a form of driving regional innovation development at the provincial level, encompass a comprehensive, integrated, and coordinated “package” of innovative policies [19] that can enhance the resilience of regional innovation ecosystems through mechanisms such as expanding the scale of innovation subjects, optimizing the supply of innovation resources, factors, and services, and integrating the mechanisms of innovation chains, industrial chains, talent chains, capital chains, and policy chains [20]. Therefore, from the research perspective of this study, the rise of the resilience of regional innovation ecosystems is considered a concentrated manifestation of the expansion of innovation subjects, the consolidation of innovation support, the stimulation of innovation vitality, the influx of innovation resources, and the optimization of the innovation environment.

(1)

Alleviation Effect of Innovation Subject Entry Constraints

First, local governments in the process of innovative province construction will concentrate on increasing regional knowledge supply, adopting policies to support and introduce high-level universities and high-energy research institutions, expanding the scale of innovation producers in the regional innovation ecosystem, enhancing the innovation capability and efficiency of innovation producers, and guiding the formation of a collaborative governance mechanism and coupling relationship between the region and universities and research institutions [21]. Thus, the construction policy for innovative provinces effectively alleviates constraints on the entry of innovation entities, unleashing innovative vitality. At the same time, innovative province construction can also generate the effect of financial subsidies and support the rapid development of high-tech enterprises and related industrial chains [22]. Local governments will also drive the entry of high-tech industries and other innovation consumers into the region and market by alleviating financing constraints, aggregating risk investment, and providing fund rewards and expert guidance [23]. Consequently, the construction of innovative provinces enhances the rapid development and agglomeration effects of high-tech enterprises through economic benefits, invigorating the innovative vitality and developmental potential of the regional innovation ecosystem. This also fosters deep integration and collaborative innovation between academia and industry, thereby enhancing the overall innovative capability of the regional innovation ecosystem and strengthening its resilience.

(2)

Optimization Effect of Resource Allocation

Second, innovative province construction can enhance the level of innovation support and innovation resource supply in the region. On one hand, as the functions of Chinese local governments gradually shift from regulatory and developmental governments to empowering and service-oriented governments, they are more focused on indirectly providing decentralized innovation incentives by providing innovation services and platforms to provide corresponding support for the evolution of the regional innovation ecosystem and strengthen its resilience [24]. On the other hand, under the policy drive of innovative province construction, local governments implement housing projects, household registration policies, and research funding to create an innovative talent highland, develop an innovative economy and increase the contribution rate of scientific and technological progress, and increase R&D investment. At the same time, they form an innovative policy combination represented by “publicizing the list”, organizing major technical joint tackling, and “attracting phoenixes to build nests” to continuously empower the resilience of the regional innovation ecosystem at the level of resource supply. Moreover, traditional methods of resource allocation often suffer from issues like information asymmetry and inefficiency, preventing innovation resources from being matched effectively with the most promising innovative projects [25]. The policies guiding the construction of innovative provinces optimize resource allocation by combining market mechanisms with governmental guidance. On one hand, they leverage the decisive role of the market in resource allocation, promoting the flow of resources toward efficient innovation entities through competitive mechanisms [26]. On the other hand, the government intensifies support for key areas and weak links through policy guidance, ensuring the effective supply of innovation resources and thereby enhancing the stability and resilience of the regional innovation ecosystem.

(3)

Effect of Multi-Subject Innovation Incentives

In addition, innovative province construction has broken through the original institutional and mechanistic barriers, connected the communication channels among innovation subjects, innovation support, and innovation resources, and promoted the integration and coordination of the innovation chain, industrial chain, talent chain, and capital chain. The “Guidelines” state the following: “Improve the existing mechanism for obtaining and using innovation elements at the regional level, reduce the hidden threshold for innovation and entrepreneurship, and various institutional transaction costs”. The breakthrough of existing institutional and mechanistic barriers and the deep integration of various development links in the regional innovation ecosystem can greatly unleash the vitality of regional innovation and entrepreneurship [27]. On one hand, it stimulates knowledge production models dominated by universities and research institutions to focus on original innovation and basic research, empowering the region to become a source of innovation; on the other hand, it drives revolutionary breakthroughs in industrial technology, enhancing the value of the industrial chain supply chain. It is necessary to point out that innovative province construction is first and foremost a “disruptive innovation” in terms of institutions, and through specific measures, it necessarily pursues unconventional means of innovation element allocation and disruptive institutional arrangements, which will greatly stimulate innovation vitality and strengthen the innovation potential of the region. From the experience of implementing various pilot region projects, whether it is the coordinated development of the “Science and Technology Valley”, “Power Valley”, and “Intelligent Manufacturing Valley” in Changsha–Zhuzhou–Xiangtan in Hunan or the construction of a military–civilian integrated technology industry base in Shaanxi and the “Silk Road” international technology cooperation base and the recent approval of the construction of innovative provinces, Jilin has vigorously promoted the construction of the Changchun Regional Innovation Center, the Changjitu Innovation Uplift Belt, and the city (state) county-level innovation collaborative network, all based on promoting the aggregation of innovative subjects, injecting high-level innovation vitality into strategic, professional, and characteristic science and technology teams, and nurturing technology-based small- and medium-sized enterprises to facilitate breakthroughs in key core technologies and the eruption of original innovation. The policies guiding the construction of innovative provinces establish a comprehensive innovation incentive system, invigorating the innovative enthusiasm of diverse entities. Policies such as R&D tax deductions and preferential recognition for high-tech enterprises encourage the deep integration of industry, academia, and research while supporting universities and research institutions through project funding and achievement transformation rewards. This fosters the organic combination of knowledge, technology, and capital, providing robust support for the regional innovation ecosystem and enhancing its resilience.

(4)

Soft Environment Support Effect

In conclusion, the support of innovative province construction for an increase in regional innovation ecosystem resilience is manifested not only in the provision of hardware such as support platforms and innovation resources but also in the optimization of the soft environment for the evolution of the innovation ecosystem. In addition to providing routine support to innovation subjects, local governments, relying on precise positioning and efficient support for innovation processes, have achieved a transformation from technological governance to empowering governance. Institutional innovations such as “publicizing the list” and “innovation vouchers” not only reflect the progress of government governance technology and the optimization of governance methods but also indicate effective intervention by the government in the pain points, breakpoints, and bottlenecks of industrial chains, innovation chains, and other processes based on a stance of market neutrality and respect for market resource allocation. Furthermore, under the impetus of innovative province construction, the optimization of financial, cultural, and technological market environments creates institutional conditions for the rapid circulation and innovative allocation of innovation resources, such as funds, knowledge, and technology. For example, in the top-level design and target planning of innovative province construction in provinces such as Zhejiang, Anhui, Shandong, and Hunan, the development of science and technology finance and the promotion of the “integration of finance and technology” are emphasized as important approaches. In addition, innovative province construction also advocates for strengthening intellectual property protection and calls for adjustments to and innovations in existing intellectual property system arrangements, particularly in support of data intellectual property and regional characteristic industry intellectual property, which have become important measures in innovative province construction in recent years. The policies guiding the construction of innovative provinces emphasize the creation and enhancement of a favorable soft environment. By refining the system of innovation policies and regulations, the legitimate rights and interests of innovation entities are safeguarded. Simultaneously, a societal atmosphere that respects knowledge, talents, and innovation is cultivated, igniting a passion for innovation throughout society. A comprehensive innovation service system is established to provide one-stop services, including information consultation, technical support, and market development, thereby promoting the evolution of the regional innovation ecosystem and enhancing its resilience.

Based on the above analysis, this study proposes the following hypothesis:

Hypothesis 1 (H1).

Innovative province construction can significantly promote the resilience of the regional innovation ecosystem.

3.2. Artificial Intelligence and Regional Innovation Ecosystem Resilience

Regional intelligence based on the access, aggregation, and efficient processing of massive data can break through the data barriers among innovation entities, driving bidirectional activation among the digital economy, virtual economy, artificial intelligence, and real economy and injecting momentum into the enhancement of regional innovation ecosystem resilience [28]. Regional intelligence not only surpasses traditional artificial intelligence in terms of the scale of data collection and processing but also plays a new intelligent shaping role in regional development, especially injecting a new generation of information technology into the regional innovation process and profoundly impacting the resilience of regional innovation ecosystems [29].

(1)

Innovation Community Generation Effect

Looking at the specific mechanisms, first, the development of intelligence gives birth to novel networked innovation communities supported by emerging technologies, such as artificial intelligence, big data, virtual reality, and large language models. These innovation communities revolve around a series of intelligent technologies, such as e-commerce, industrial Internet, industrial metaverse, and smart terminals, forming extensive horizontal and vertical industrial networks [30]. In the specific process, intelligence itself can provide massive data support for enterprise decision-making, management, and market entry for start-ups, significantly deepening the alignment between enterprise innovation development and the market, enhancing enterprise vitality, and expanding the scale of innovative consumers and the precision of knowledge consumption [31]. Moreover, innovative consumers empowered by intelligence are the first to enter the phase of increasing returns to scale in the evolution process of a regional innovation ecosystem [32]. Compared to traditional innovation entities, these innovative consumers operate at a higher level, expand more rapidly, have tighter integration, and evolve more swiftly; they are even capable of replacing or upgrading traditional innovation entities, guiding the development and evolution of the entire regional innovation ecosystem. For example, in the industrial metaverse scenario, suppliers, innovation entities, and partners in the industrial chain supply chain are converged in intelligent spaces for collaboration, innovation, and production, significantly surpassing the efficiency of innovation and the structural upgrading of traditional industrial manufacturing chain supply chains. Furthermore, the process of intelligence will compel universities, research institutions, and other innovation producers to increase the supply of intelligence-related knowledge and talent to the regional innovation ecosystem and promote disciplinary transformation against the intelligence background, such as by integrating intelligent technologies with traditional disciplines such as engineering, medicine, and economics [33], embedding disciplinary development and basic research into massive data resources and intelligent technology empowerment, accelerating the evolution of technology life cycles, and providing more valuable knowledge more efficiently to the regional innovation ecosystem. The close collaboration among artificial intelligence firms, universities, and research institutions fosters joint research and development, talent cultivation, and result transformation, creating synergistic innovation effects. Moreover, the widespread application of regional artificial intelligence technology propels the development of upstream and downstream industrial chains, forming a complete industrial ecosystem that enhances the resilience of the supply chain [34]. Additionally, the formation of innovation communities creates a vibrant atmosphere for innovation in the region, attracting more innovative resources and talent [35] and further spurring the development of the regional innovation ecosystem.

(2)

Resource Platform Agglomeration Effect

Second, regional intelligence can empower the construction of innovation platforms, provide more convenient and intelligent innovation support for innovation entities, and offer new quality innovation resources to the innovation ecosystem. First, intelligent technologies, such as the Internet of Things, big data, and large language models, enhance the operational efficiency of innovation platforms such as technology enterprise incubators, university science and technology parks, innovation corridors, key laboratories, and industrial science and technology parks, enabling the digital management of processes such as industry–university–research cooperation, technology enterprise incubation, major technological breakthroughs, and innovation project selection, evaluation, investment, follow-up, and acceptance to ensure the smooth, efficient, and sustainable progress of such projects. Second, intelligent technologies strengthen innovation services by improving the efficiency of innovation resource allocation. The intelligent innovation network formed by intelligence connections among enterprises, between enterprises and innovation platforms, and between enterprises and other innovation entities can achieve the precise matching of innovation elements between supply and demand parties [36]. Similarly, intelligence leads universities, research institutions, and other innovation producers to achieve the open sharing of research instruments, equipment, data, and other innovation resources through intelligent platforms, which has become an important way for innovation entities to participate in major scientific and technological joint research efforts [37]. Third, intelligence can drive the transformation and upgrading of innovation platforms and innovation services, enabling them to possess more efficient empowering and service capabilities. Therefore, in terms of innovation support and innovation resources, intelligence has a constructive impact on strengthening the resilience of regional innovation ecosystems. Moreover, resource platforms within the realm of artificial intelligence reduce innovation costs and enhance efficiency through data and computing power sharing [38]. These platforms provide robust technical support for the research and application of AI technologies, promoting ongoing breakthroughs and innovations [39]. The aggregation effect of these resource platforms attracts AI enterprises to settle, forming industrial clusters that further stimulate regional economic development, thus creating a synergistic effect among regional artificial intelligence, resource platforms, and the regional industrial economy.

(3)

Empowerment Effect of Diversified Innovation Processes

Simultaneously, the new technologies, new formats, new industries, and new models triggered by regional intelligence can stimulate the vitality of regional innovation and entrepreneurship. Regional intelligence, relying on artificial intelligence, blockchain, big data, and the Internet of Things, penetrates various aspects of innovation entities, innovation elements, and innovation targets, promoting the formation of new quality productivity at the innovation level. Particularly in the field of engineering and technological innovation, intelligence can efficiently drive the research and development of new materials, promote new developments at a high level, and foster the application of new methods at a high quality, integrating massive data, generative intelligent decision-making, and human labor substitution with engineering innovation at a deep level [40]; meanwhile, in the fields of original innovation and basic research, a high-information-flow innovation network and social network constructed by regional intelligence can maximize organized scientific research, organize interdisciplinary joint efforts, and strengthen the collaborative integration and integrated connectivity capabilities of innovation entities [41], enhancing their ability to break through their own technical expertise areas, conduct cross-border searches, improve the breakthrough capability of major critical core technological research, and increase the likelihood of success in basic research; in terms of industrial formats and new product innovation, regional intelligence can empower data elements to drive the emergence of new industrial formats and innovations in new products. Additionally, regional artificial intelligence encourages open innovation through cross-sector collaboration and the integration of industry, academia, and research, thus introducing external innovation resources that promote technological innovation and industrial upgrading [42]. The profound integration of artificial intelligence technology with traditional industries accelerates the transformation and upgrading of these sectors, as well as the rapid development of emerging industries, culminating in a diversified innovation ecosystem [43]. The swift advancement of regional artificial intelligence technology promotes the continual iteration and enhancement of products and services, thereby augmenting the self-repair and renewal capabilities of the regional innovation ecosystem. Consequently, regional artificial intelligence enhances the resilience and adaptability of the innovation ecosystem by introducing diverse innovation models and methodologies.

(4)

Soft Environment Technological Transformation Effect

In terms of optimizing the innovation environment, intelligence is profoundly changing governance methods and concepts of government. The government can utilize disruptive governance methods such as big data, intelligent government affairs, and virtual digital participation in the intelligence transformation to enhance the efficiency of translating governance guidelines into action, evolving into an enabling government with a key influence on the evolution and development of the innovation chain. At the same time, the “data element x” effect and the expansion effect of computing power derived from regional intelligence can optimize the intensity and breadth of intellectual property protection. Especially in recent years, innovation in the form of data intellectual property has not only provided legal protection for innovation achievements and intellectual property definition in the field of intelligence but has also provided opportunities for the development of new formats and models. In addition, the development of soft environments, such as finance and technology markets, has become increasingly intertwined with regional intelligence. For instance, technology finance relies on important technological aspects of regional intelligence, such as automated data processing, the Internet of Things, blockchain, and online payments, forming innovative financial models that provide a new impetus for the improvement of the regional innovation ecosystem environment. Under the multi-tiered guarantees provided by government, legal frameworks, and financial services driven by advancements in artificial intelligence, regional economies are enabled to sustain healthy development.

Based on the above analysis, this study proposes the following hypothesis:

Hypothesis 2 (H2).

Intelligence can significantly promote the resilience of the regional innovation ecosystem.

3.3. The Construction of Innovative Provinces Drives the Empowerment of Regional Innovation Ecosystem Resilience through Artificial Intelligence

Under the guidance of the “Guidelines”, pilot regions’ local governments often allocate a significant amount of attention to artificial intelligence in the process of promoting the construction of innovative provinces [44]. For instance, the Zhejiang Provincial Committee clearly stated in its deployment of innovation, deepening reform, and opening up enhancement that it aims to lead the construction of a modern industrial system as a breakthrough for the construction of innovative provinces through elements of digital development such as the digital economy. It accelerates the realization of digital empowerment and data value surge by embracing ubiquitous sensing, interconnection, and intelligence.

Initially, from a technological perspective, the local governments of pilot regions need to anchor the development and enhancement of regional computing power to achieve regional informatization and digital transformation. Therefore, local governments should increase research and development investment in the digital field and focus on nurturing new formats and models of artificial intelligence. In response to this demand, the policies guiding the construction of innovative provinces promote the development of regional computing infrastructure, facilitate the conditional opening of public and social data to enterprises, and encourage companies to participate in the establishment of innovative service platforms for data aggregation, algorithm consolidation, and computing power testing. This approach provides abundant data resources for the research and application of artificial intelligence technologies. Thus, the support of artificial intelligence by local governments can strengthen the surge in resilience in the digital field itself.

Second, based on the “construction of innovative provinces—artificial intelligence” transmission pathway, this study analyzed the specific dimensions of resilience within the regional innovation ecosystem, including the buffering capacity, fluidity, evolutionary potential, diversity, and coordination. First, support for research and development investments in artificial intelligence through innovative province construction policies enables a region to have a stronger technological reserve and response capability when facing external economic fluctuations or technological transformations, thereby enhancing the buffering capacity of the innovation ecosystem. Simultaneously, the establishment of a stable mechanism for technological innovation investment promotes the increase in societal R&D expenditures as a proportion of regional GDP, providing ample resource reserves for the innovation ecosystem to address uncertainty risks. Second, policies promoting deep integration among industry, academia, and research strengthen knowledge exchange and collaboration among enterprises, universities, and research institutions, accelerating knowledge flow and technological diffusion in the field of artificial intelligence and thereby enhancing the fluidity of the regional innovation ecosystem. Third, policies that facilitate the deep integration of artificial intelligence technology with traditional industries promote the optimization and upgrading of industrial structures, thereby enhancing the evolutionary capability of the regional innovation ecosystem. Fourth, the policies support various scales and types of enterprises in the artificial intelligence sector, including startups, small- and medium-sized enterprises, and leading firms, forming a diverse landscape of innovation entities that enhances the diversity of the innovation ecosystem. Finally, by establishing and improving deliberative coordination mechanisms, policies guide the resolution of significant issues in the construction of innovative provinces and encourage the establishment of enterprise-centered, collaborative innovation mechanisms that promote synergy among innovation entities, creating a coordinated regional innovation ecosystem.

In conclusion, local governments can strengthen regional digital development, thereby empowering the expansion and improvement of regional innovation subjects, efficiently allocating innovation resources, transforming and optimizing innovation support, and creating a sound, stable, and vibrant innovation environment. This, in turn, promotes an overall surge in the resilience of a regional innovation ecosystem. In essence, this constitutes a transmission mechanism path of “innovative province construction → digital development → enhancement of regional innovation ecosystem resilience.” Therefore, this article presents the following hypothesis:

Hypothesis 3 (H3).

Regional artificial intelligence development can play a positive mediating effect between the construction of innovative provinces and the resilience of the regional innovation ecosystem.

3.4. Quasi-Natural Experiment Design and Model Construction

3.4.1. Spatial Double-Difference Model Construction

In order to preliminarily verify the impact mechanism of innovative provincial construction and artificial intelligence regarding the resilience of the regional innovation ecosystem, this study first employs the classic causal inference method—the double-difference model for a quasi-natural experiment. However, the conventional double-difference model often overlooks the spatial correlation of variables in spatial panels, violating the Gauss–Markov classic assumption. Therefore, this study utilizes a spatial double-difference model for examination, which enables the exploration of the spatial spillover effects of core variables on regional ecosystem resilience. The core advantage of this model lies in its ability to significantly eliminate or reduce systematic and random errors, thereby highlighting authentic signals or patterns. In the context of continuous space, the model enhances the precision and reliability of data analysis through the meticulous processing of spatial data [45]. Consequently, this paper employs the model to accurately quantify the spatial effects arising from the transformation of the ecological niche suitability within high-tech industrial innovation ecosystems, validating its impact on green energy efficiency. To assess the separate impacts of innovative provincial construction and artificial intelligence and their spatial effects on the resilience of the regional innovation ecosystem, this study constructs a spatial double-difference model (Models 1–3) based on the research of Chagas et al. [46].

Model 1:

$${Res}_{it}=\rho \mathit{W}\cdot {Res}_{jt}+{{\alpha }_1{DID}_{it}+{\alpha }_2{AI}_{it}+\sum \beta {\mathit{X}}_{\mathit{i}\mathit{t}}+\gamma }_t+u_i+{\epsilon }_{it}$$

(1)

Model 2:

$${Res}_{it}={{\alpha }_1{DID}_{it}+{\alpha }_2{AI}_{it}+\sum \beta {\mathit{X}}_{\mathit{i}\mathit{t}}+\gamma }_t+{\lambda \mathit{W}\cdot v_{jt}+u}_i+{\epsilon }_{it}$$

(2)

Model 3:

$$\begin{gathered} {Res}_{it}=\rho \mathit{W}\cdot {Res}_{jt}+{{\alpha }_1{DID}_{it}+{\alpha }_2{AI}_{it}+\sum \beta {\mathit{X}}_{\mathit{i}\mathit{t}}+\theta \mathit{W}\left({DID}_{jt}+{Di}_{jt}+\sum {\mathit{X}}_{\mathit{j}\mathit{t}}\right)+\gamma }_t \\ +u_i+{\epsilon }_{it} \end{gathered}$$

(3)

Models 1–3 consist of Spatial Double-Difference Models (SDID) based on the Spatial Autoregressive Model (SAR), Spatial Error Model (SEM), and Spatial Durbin Model (SDM). In Equations (1)–(3), ${Com}_{it}$ represents the dependent variable of regional innovation ecosystem resilience, where ${DID}_{it}$ and ${AI}_{it}$ denote the policy treatment variables for innovative provincial construction in region i at time t and the regional artificial intelligence index, respectively, with coefficients of ${\alpha }_1$ and ${\alpha }_2$. W is the spatial weight matrix underlying the Spatial Double-Difference Model. The sum of $\sum \beta {\mathit{X}}_{\mathit{i}\mathit{t}}$ represents the product of control variables and their coefficients, ρ stands for the spatial autoregressive coefficient, and θ represents the coefficient of spatial spillover effects. $v_{jt}$ denotes external shock variables, while $u_i$ and ${\gamma }_t$ indicate the fixed effects of region and time, and ${\epsilon }_{it}$ denotes the random disturbance term.

Additionally, to preliminarily explore the synergistic effects of core variables on the resilience of the regional innovation ecosystem, this study introduces the interaction term of the policy treatment variables for innovative provincial construction and the regional artificial intelligence index (${DID}_{it}$ × ${AI}_{it}$) into three sets of Spatial Double-Difference Models.

3.4.2. Dual Machine Learning Model Construction

To address the issues of the curse of dimensionality, the difficulty in exhaustively enumerating key control variables, and the inability to precisely fit the regression model with linear combinations of control variables in traditional double-difference models, this study refers to the research of Chernozhukov et al. [47]. In the quasi-natural experiment, Dual (Unbiased) Machine Learning (DML) is introduced for causal inference analysis. To verify the mechanism hypotheses H1 and H2 and similarly introduce interaction terms like Cho, KA et al. to examine synergistic effects, the baseline regression models based on partial linear regression models are constructed as shown in Equations (4)–(6) [48].

Model 4:

$${Res}_{it+1}={\alpha }_1{DID}_{it}+g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)+U_{it},E\left(U_{it}|{DID}_{it},{\mathit{X}}_{\mathit{i}\mathit{t}}\right)=0$$

(4)

Model 5:

$${Res}_{it+1}={\alpha }_2{AI}_{it}+g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)+U_{it},E\left(U_{it}|{AI}_{it},{\mathit{X}}_{\mathit{i}\mathit{t}}\right)=0$$

(5)

Model 6:

$${Res}_{it+1}={\alpha }_3{DID}_{it}\times {AI}_{it}+g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)+U_{it},E\left(U_{it}|{DID}_{it}\times {AI}_{it},{\mathit{X}}_{\mathit{i}\mathit{t}}\right)=0$$

(6)

In Equations (4)–(6), the function $g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)$ represents a multidimensional control variable, the specific functional form of which $\widehat{g}\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)$ will be obtained through a machine learning model. The control variables in Model 9 include ${DID}_{it}$ and ${AI}_{it}$. Subsequently, to correct for endogeneity bias, it is necessary to construct auxiliary equations regarding the core explanatory variables and obtain unbiased estimates of the coefficients using the instrumental variable method.

Moreover, to examine the mediating mechanism of hypotheses H3(a-e), this study, drawing on Cho K A et al. [49], establishes a dual machine learning mediation model (Model 7) based on a stepwise regression approach.

Model 7:

$$\left\{\begin{array}{l}Equation1:{Res}_{it+1}\left({Sub}_{it+1},{Sur}_{it+1}\dots \right)={\alpha }_1{DID}_{it}+g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)+U_{it},E\left(U_{it}|{DID}_{it},{\mathit{X}}_{\mathit{i}\mathit{t}}\right)=0\\ Equation2:{Di}_{it+1}={\beta }_1{DID}_{it}+g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)+U_{it},E\left(U_{it}|{DID}_{it},{\mathit{X}}_{\mathit{i}\mathit{t}}\right)=0\\ Equation3:{Res}_{it+1}\left({Sub}_{it+1},{Sur}_{it+1}\dots \right)={\alpha }_1^{\prime }{DID}_{it}+{\alpha }_2{AI}_{it}+g\left({\mathit{X}}_{\mathit{i}\mathit{t}}\right)+U_{it},E\left(U_{it}|{DID}_{it},{\mathit{X}}_{\mathit{i}\mathit{t}}\right)=0\end{array}\right.$$

(7)

In Equation (7), ${Res}_{it+1}({Sub}_{it+1},{Sur}_{it+1}\dots )$ represents the regional innovation ecosystem resilience and its sub-indicators (innovation subject, innovation support, innovation vitality, innovation resources, innovation environment).

3.5. Explanation of Variables and Data Sources

3.5.1. Interpreted Variable: Regional Innovation Ecosystem Resilience (Res)

Drawing on the research findings of Liu Hedong and Lu Chenxi [50], the resilience of regional innovation ecosystems is primarily categorized into four dimensions: diversity, buffering capacity, evolutionary capacity, and fluidity, referred to as the four-dimensional resilience characteristics of regional innovation ecosystems. Building on this foundation, this study posits that, in addition to the fundamental dimension (diversity), external variation dimension (buffering capacity), representational dimension (evolutionary capacity), and internal driving dimension (fluidity), the integration dimension should also be considered. This dimension characterizes the degree of interaction, integration, and collaboration among internal producers (such as universities, research institutions, and enterprise R&D departments), consumers (entities applying innovative technologies), and disassemblers (knowledge transfer entities like bases, parks, and incubators) within the innovation ecosystem. Consequently, this study incorporates the symbiotic dimension (coordination) into the resilience characterization framework of regional innovation ecosystems.

(1)

Diversity (Div): Diversity is primarily reflected in the types, quantities, and scales of innovation entities, populations, and communities within the regional innovation ecosystem. A higher degree of diversity enhances the system’s resilience against external disturbances.

(2)

Buffering Capacity (Buf): Buffering capacity refers to the ability of the regional innovation ecosystem to withstand external shocks [51]. The inherent resource endowments and innovative accumulations within the region, along with their complex structural compositions, can strengthen the buffering capacity of the ecosystem.

(3)

Flowability (Flo): Fluidity is characterized by the interwoven flows of capital, information, talent, and technology both within and outside the regional innovation ecosystem.

(4)

Evolutionary Capacity (Evo): Evolutionary capacity relies on the continuous growth in scale, structural upgrades, and energy enhancement of innovation entities, populations, and communities within the regional innovation ecosystem through the exchange of information, materials, and energy with the external environment.

(5)

Coordination (Coo): Coordination reflects the complex relationships formed among innovation entities, populations, and communities within the regional innovation ecosystem through knowledge spillovers, technological collaboration, and information sharing, as well as the empowering effects of knowledge and technology on local industries.

Based on the above analysis, this study establishes an indicator system for the resilience of regional innovation ecosystems, as shown in

Table 1. Comprehensive evaluation index system for regional innovation ecosystem resilience (Res).

Level II Evaluation Items	Level III Evaluation Items	Proxy Data	Indicator Attributes
Diversity (Div)	Diversity in Higher Education	Number of Higher Education Institutions	+
	Diversity of Scientific Research Institutions	Number of Scientific Research Institutions	+
	Diversity of Innovative Entities in the Market	Number of High-tech Enterprises	+
Buffer Capacity (Buf)	Accumulation of Innovative Knowledge	Cumulative Number of Scientific Papers	+
Evolvability I—R&D Dimension (Evo-I)	Technological Accumulation	Cumulative Number of Valid Inventions	+
	Investment in R&D Innovation	Full-Time Equivalent of R&D Personnel	+
		Industry–Academia–Research Funding	+
Evolvability II—Commercialization Dimension (Evo-II)	Output from R&D Innovation	Number of Patent Applications	+
	Investment in Commercialized Innovation	End-of-Year Employment Number	+
		Enterprise Technology Import Funding	+
		Enterprise Technology Digestion Funding	+
		Enterprise Technology Acquisition Funding	+
		Enterprise Technology Renovation Funding	+
	Output from Commercialized Innovation	New Product Sales Revenue	+
		Technology Market Transaction Volume	+
Flowability (Flo)	Talent Mobility	Number of Students Enrolled	+
	Capital Mobility	Government Science and Technology Investment	+
		Enterprise Science and Technology Investment	+
		Foreign Science and Technology Investment	+
	Technology Mobility	Technology Market Technology Inflow Regional Amount	+
		Technology Market Technology Outflow Regional Amount	+
	Information Mobility	Number of Internet Broadband Access Ports	+
Coordination (Coo)	Knowledge-Driven Industrial Development	Value Added of High-tech Industries	+
		Total Power of Agricultural Mechanization	+
		Knowledge Empowerment of Emerging Industries and Service Development	+
		Number of Graduating Enterprises from Provincial Science and Technology Business Incubators in the Current Year	+
	Regional Integration of Industry, Academia, and Research	Number of Graduating Enterprises from National University Science and Technology Parks	+
		Full-Time Equivalent of Local University R&D Results Application and Science and Technology Service Personnel	+
		Funding for Local University Results Application and Science and Technology Service Projects	+

.

Referencing the research findings of Liu Hedong and Lu Chenxi [50], the resilience of regional innovation ecosystems is primarily categorized into diversity, buffering capacity, evolutionary potential, and fluidity, collectively referred to as the four-dimensional resilience characteristics of regional innovation ecosystems. Building upon this foundation, this study posits that, in addition to the fundamental dimension (diversity), external variation dimension (buffering capacity), representational dimension (evolutionary potential), and internal driving dimension (fluidity), the integration dimension should also be considered. This dimension characterizes the degree of interaction, integration, and collaboration among internal producers (such as universities, research institutions, and corporate R&D departments), consumers (entities applying innovative technologies), and decomposers (knowledge conversion entities like bases, parks, and incubators) within the innovation ecosystem. Consequently, this study incorporates the symbiotic dimension (coordination) into the resilience characterization framework of regional innovation ecosystems.

In line with the research of Liu Hedong and Lu Chenxi, this study constructs an index system for assessing the resilience of regional innovation ecosystems (see Table 1) and employs the entropy method to calculate the comprehensive index of resilience as well as its sub-indicators.

Data sources span from the years 2009 to 2021, encompassing the “China Science and Technology Statistical Yearbook”, “China High-tech Industry Statistical Yearbook”, “China Statistical Yearbook”, “China Torch Statistical Yearbook”, and “Compilation of Statistical Data on Science and Technology from Chinese Universities”. Data regarding knowledge empowerment for the development of emerging industries and services is derived from data mining, referencing the “14th Five-Year National Science and Technology Innovation Plan”. A selection of 57 keywords, including deep learning, machine learning, blockchain, artificial intelligence, and big data, is established, and through advanced Baidu searches, relevant occurrences of these keywords in relation to various prefecture-level cities from 2009 to 2021 are acquired. Utilizing Python software 3.8.5, the frequency of these keywords associated with each city is crawled and summarized by province, with the total word frequency serving as an indicator of the development of knowledge-empowered emerging industries and services within the region.

3.5.2. Interpret Variable I: Policy Handling Variable for Innovative Province Construction (DID)

According to the requirements of quasi-natural experimental design, the variable for the group undergoing the national-level innovative province construction policy pilot during the period 2009–2021 is assigned a value of 1, while the others are assigned a value of 0. The pilot regions for innovative province construction approved by China include eleven provinces: Jiangsu (2013), Anhui (2013), Shaanxi (2013), Zhejiang (2013), Hubei (2016), Guangdong (2016), Fujian (2016), Sichuan (2017), Shandong (2017), Hunan (2018), and Jilin (2021).

3.5.3. Interpret Variable II: Regional Artificial Intelligence Index (AI)

Among the key indicators for measuring a region’s potential for development in artificial intelligence is intelligent input, which encompasses four primary aspects: financial investment, talent reserves, infrastructure development, and policy support [52].

Financial investment reflects the region’s emphasis on artificial intelligence research and development, as well as its economic strength [53]. Talent reserves are crucial in determining the sustainability and depth of technological innovation [54]. Infrastructure development, including high-performance computing platforms and data centers, serves as the foundational support for the application of artificial intelligence [55]. Policy support, through guiding policies and incentive mechanisms, creates a favorable external environment for the advancement of artificial intelligence [56].

Production application directly measures the degree of penetration and practical impact of artificial intelligence technologies within the regional economy, encompassing the breadth and depth of AI applications as well as their actual effects on corporate productivity, service quality, and cost control. Innovation and benefits assess the innovative potential driven by the development of artificial intelligence within the region and its future developmental tendencies [57].

Based on the study design and referencing the research of Xin Yongrong et al., the regional artificial intelligence index is evaluated in three dimensions: intelligent input, production application, and innovation and benefits [58]. Drawing on the studies of Li Lianhui et al. and Meng Xiaona et al. for indicator selection [59,60], the comprehensive evaluation index system for the regional artificial intelligence index is established as shown in

Table 2. Comprehensive evaluation index system for regional artificial intelligence index (Ai).

Level II Evaluation Items	Level III Evaluation Items	Proxy Data	Indicator Attributes
Intelligent Investment	Internet Infrastructure Investment	Optical Cable Length per Provincial Area	+
	Intelligent Funding Investment	High-tech Manufacturing R&D Funding	+
	Intelligent Talent Investment	High-tech Manufacturing R&D Personnel	+
	Intelligent Equipment Investment	Fixed Asset Investment in Information Transmission, Software, and Information Technology Services	+
Intelligent Application	Software Development and Application Status	Software Product Revenue/Industrial Enterprise Main Business Revenue	+
	Intelligent Product Development Status	Embedded Systems Business Revenue/Industrial Enterprise Main Business Revenue	+
	Development Status of Intelligent Enterprises	Number of Artificial Intelligence Enterprises in Various Regions	+
	Degree of Intelligent Technology Application	Industrial Robot Penetration Rate in Various Regions	+
Innovation and Market Benefits	Innovative Capacity	Number of Artificial Intelligence Patents in Various Regions	+
	Profits in the Intelligent Market	Total Profits of High-tech Manufacturing	+
	Efficiency in the Intelligent Market	Main Business Revenue of High-tech Manufacturing/Number of High-tech Manufacturing Employees	+
	Social Benefits	Energy Consumption per Unit GDP (Coal, Electricity)	+

.

In Table 2, the degree of application of intelligent technologies is represented by the penetration of industrial robots across different regions. The method for measuring the installation density of industrial robots refers to the study by Acemoglu et al. [61], primarily by matching the usage of robots in various industries in China disclosed by IFR with the industry sector categories in the “China Labor Statistics Yearbook” and then calculating the industrial robot penetration in each region on a yearly basis as a proxy variable for the installation density of industrial robots; the number of intelligent patents applied for is obtained by searching for relevant IPC codes related to artificial intelligence, machine learning, algorithms, etc. on the Yipin platform, with the following specific search formula provided: “Patentee province = (Beijing) AND (IPC = (G05D1/02) OR IPC = (G05D1/08) OR IPC = (G05D1/12) OR IPC = (G06F1/16) OR IPC = (G05B15/02) OR IPC = (G06K9/66) OR IPC = (G07C9/00) OR IPC = (G08B19/00) OR IPC = (G08B25/10) OR IPC = (G06F3/01) OR IPC = (G06F9/44) OR IPC = (G06F9/455) OR IPC = (G06N3/12) OR IPC = (G06N5/00) OR IPC = (G06N5/02) OR IPC = (G06N5/04) OR IPC = (G06K9/00) OR IPC = (G06K9/62) OR IPC = (G06N3/02) OR IPC = (G06N3/08) OR IPC = (G60F40*) OR IPC = (A61B5/0467) OR IPC = (A61B5/0478)) AND Application date = (2009.01.01:2009.12.31)”.

The remaining data are derived from the “China Electronic Information Industry Statistics Yearbook”, the “China Science and Technology Statistics Yearbook”, the “China Logistics Yearbook”, the “China Information Industry Yearbook”, the “Peking University Artificial intelligence Inclusive Finance Index”, and the EPS database, with missing data filled using linear interpolation to form a balanced panel data for 30 provinces (excluding Tibet) from 2009 to 2021.

This study also utilizes the entropy method to comprehensively evaluate the indicator system presented in Table 2, resulting in an index for the level of regional artificial intelligence development and its sub-indicators.

3.5.4. Control Variables

Prior studies have shown that local government intervention, environmental regulations, green finance, and other economic variables can drive knowledge production, unlock innovation technology barriers, and stimulate regional innovation vitality [62,63]. Furthermore, the depth adjustment of the industrial structure has been proven to affect the aggregation of regional innovation elements and optimize the innovation environment [64].

Therefore, this study selects the government scale (GI), government environmental attention (Att), level of green finance (GF), and rationalization of industrial structure (RIS) as control variables. The scale of government is represented by the ratio of the general public budget expenditure of local governments to the actual GDP of the region; the reciprocal of the Theil index of industrial structure rationalization proposed by Hu Longwei et al. is used as the measure of industrial structure rationalization [65]. To measure government environmental attention and the level of green finance, this study constructs an indicator system, as shown in

Table 3. Comprehensive evaluation index system for government environmental attention (Att) and green finance level (GF).

Variable Name	Sub-Indicators	Evaluation Items	Measurement Methods	Indicator Attributes
Government Environmental Attention (Att)	Policy Planning Attention	Green Development Focus	-	+
		Intensity of the “Five-in-One” Ecological Civilization Layout	-	+
	Resource Allocation Attention	Environmental Governance Intensity	Expenditure on Environmental Pollution Control/General Public Budget Expenditure	+
		Strength of Environmental Protection Infrastructure Construction	Investment in Environmental Infrastructure/GDP	+
	Legislative Attention	Regional Ecological Civilization Construction Legislation	Number of Ecological and Environmental-related Local Legislations/Total Number of Local Legislations	+
Green Finance Level (GF)	Green Credit	Proportion of Environmental Project Loans	Credit Amount for Environmental Protection Projects/Total Credit Amount	+
	Green Investment	Level of Environmental Governance Investment	Investment in Environmental Pollution Control/GDP	+
	Green Insurance	Comprehensive Level of Pollution Liability Insurance	Income from Liability Insurance for Environmental Pollution/Total Premium Income	+
	Green Bonds	Issuance Level of Green Bonds	Total Issuance of Green Bonds/Total Issuance of All Bonds	+
	Green Funds	Proportion of Green Funds	Market Value of Green Funds/Total Market Value of All Funds	+
	Green Equity	Development Level of Green Equity	Transaction Volume of Carbon, Energy Usage Rights, and Pollution Rights/Total Transaction Volume of Equity Market	+

.

In Table 3, the focus on green development is measured by the total frequency of key words related to green development (including “environmental protection”, “ecology”, “blue skies and white clouds”, and “low carbon”, among 94 other words) in the annual government work reports of all prefecture-level cities under each province, divided by the total frequency of all words in the government work reports of prefecture-level cities under the province for that year. On the other hand, the intensity of the “Five-in-One” ecological civilization layout is calculated by measuring the frequency of five key words (“ecology”, “economy”, “society”, “politics”, “culture”) individually in the aforementioned government work reports and then expressing the ratio of the frequency of “ecology” to the total frequency of the five key words. The data related to the number of local legislations were obtained from searching the PKULaw platform.

The remaining data mentioned above are derived from the “China Statistical Yearbook”, “China Environmental Statistical Yearbook”, and statistical yearbooks of various provinces and cities, and the entropy method is likewise employed to comprehensively evaluate the index system of Table 3.

3.5.5. Spatial Weight Matrix

The spatial weight matrix describes the dependency relationships between geographical regions and serves as a medium for variables to exhibit spatial effects in the spatial Durbin model. Given the close economic linkages between the subject of this study and geography, an economic spatial weight matrix is constructed in this study as the spatial weight matrix used for empirical analysis, with a specific form, as shown in Equation (8):

$$\mathit{W}=\left[\begin{array}{ccc}\begin{array}{cc}1& {\mathit{W}}_{12}\\ {\mathit{W}}_{21}& 1\end{array}& \cdots & \begin{array}{c}{\mathit{W}}_{1\mathit{N}}\\ {\mathit{W}}_{2\mathit{N}}\end{array}\\ ⋮& \ddots & ⋮\\ \begin{array}{cc}{\mathit{W}}_{\mathit{N}1}& {\mathit{W}}_{\mathit{N}2}\end{array}& \cdots & 1\end{array}\right],$$

$$W_{i\theta }=\left\{\begin{array}{c}{\displaystyle \frac{1}{|y_i-y_{\theta }|}}i\ne \theta \\ 1i=\theta \end{array}\right.$$

(8)

In the above equation, W represents the economic spatial weight matrix, $W_{i\theta }$ denotes an element in W, and $y_i$ and $y_{\theta }$ respectively represent the average actual GDP of regions i and θ.

4. Empirical Analysis

4.1. Causal Double-Difference Analysis Based on Spatial Effects

4.1.1. Spatial Autocorrelation Test Based on Moran’s I

Spatial autocorrelation is one of the core concepts in spatial econometrics, referring to the inherent relationships between data at adjacent geographic locations. When data exhibit spatial autocorrelation, traditional statistical methods such as Ordinary Least Squares (OLS) may underestimate standard errors, leading to erroneous estimations of parameter significance and ultimately affecting the predictive and explanatory power of the model. Moran’s I serves as an effective indicator for determining whether empirical data display spatial heterogeneity.

As a method for measuring the degree of spatial autocorrelation, Moran’s I test can detect spatial patterns within a dataset by quantifying the spatial correlation through the average similarity between each observation and its neighboring observations. If the Moran’s I statistic is significantly greater than zero, it indicates the presence of positive spatial autocorrelation, meaning that observations in neighboring areas tend to be similar (i.e., exhibiting “high–high” or “low–low” clustering). Conversely, if it is significantly less than zero, it signifies negative spatial autocorrelation, suggesting that neighboring observations tend to differ. A value close to zero implies weak or nonexistent spatial autocorrelation.

Therefore, prior to conducting spatial econometric modeling, Moran’s I test can determine whether spatial autocorrelation is present, which is crucial for selecting an appropriate model. If spatial autocorrelation is detected, it necessitates the use of spatial econometric models capable of addressing this phenomenon, such as the Spatial Autoregressive Model (SAR), the Spatial Error Model (SEM), or the Spatial Durbin Model (SDM), which simultaneously consider both spatial lags and spatial errors. These models are better suited to capture spatial dependency and can provide more accurate parameter estimates, thereby enhancing the explanatory and predictive capabilities of the model.

Table 4. Global Moran’s I for regional innovation ecosystem resilience (IER) from 2009 to 2021.

Year	Moran’s I	p-Value	Year	Moran’s I	p-Value
2009	0.186 ***	0.000	2016	0.174 ***	0.000
2010	0.185 ***	0.000	2017	0.169 ***	0.000
2011	0.190 ***	0.000	2018	0.154 ***	0.000
2012	0.185 ***	0.000	2019	0.151 ***	0.000
2013	0.191 ***	0.000	2020	0.151 ***	0.000
2014	0.184 ***	0.000	2021	0.148 ***	0.000
2015	0.176 ***	0.000

*** p < 0.01.

reports the annual Moran’s I values for the explained variable, regional innovation ecosystem resilience (IER), under the economic spatial weight matrix from 2009 to 2021, revealing that the resilience of the regional innovation ecosystem is significantly positive each year. Figure 1 presents the Moran scatter plots of IER for 2009 and 2021, demonstrating clear instances of “high–high” and “low–low” clustering in regional innovation ecosystem resilience. These results indicate the presence of spatial autocorrelation in the explained variable, thus necessitating the use of a spatial double-difference model for causal inference.

4.1.2. Model Selection for Spatial Double-Difference Analysis

In the preceding sections, this paper designed three alternative spatial double-difference models: the Spatial Autoregressive Double-Difference Model (SAR-DID), the Spatial Error Double-Difference Model (SEM-DID), and the Spatial Durbin Double-Difference Model (SDM-DID). The SAR model directly incorporates a spatial autoregressive term in the explained variable, positing that the dependent variable is influenced not only by its own explanatory variables but also by the dependent variables of neighboring regions. Consequently, the SAR model is suited for studying inter-regional interactions. The SEM model, on the other hand, introduces spatial correlation in the model’s error term, suggesting that the correlation between observations is transmitted through the error term. The SDM model can be viewed as a hybrid of the SAR and SEM models, encompassing both spatial autoregressive terms and spatially lagged independent variables, thereby capturing the spatial spillover effects of the dependent variable as well as the independent variables on the dependent variable.

To select the most suitable model for causal inference within the spatial double-difference framework, this paper first employs the LM test and its robust form to determine whether spatial lag and spatial error terms should be incorporated in the ordinary least squares estimation. The results of the tests indicate that the LM statistics formed based on the spatial lag and error terms reject the use of ordinary least squares estimation; thus, this paper adopts the SDM-DID as the empirical model for causal inference.

Additionally, while the LM test serves as a preliminary check for model specification, it is also essential to conduct post hoc tests using the Wald test and LR test to assess the relative merits of the SDM-DID, SAR-DID, and SEM-DID models. The results reveal that, whether through the Wald test or the LR test, the SDM-DID consistently outperforms both the SAR-DID and SEM-DID models. Therefore, this paper ultimately confirms the use of the Spatial Durbin Double-Difference Model for causal inference. The test results are reported in

Table 5. Selection and testing of the spatial double-difference model.

Test Item	Statistic	p-Value
LM (SAR)	3.930 **	0.047
Robust LM (SAR)	20.470 ***	0.000
LM (SEM)	15.477 ***	0.000
Robust LM (SEM)	32.017 ***	0.000
Wald (SAR)	22.53 ***	0.0010
Wald (SEM)	11.26 *	0.0807
LR (SAR)	12.88 **	0.0449
LR (SEM)	13.88 **	0.0310

* p < 0.10, ** p < 0.05, *** p < 0.01.

.

4.1.3. Parameter Estimation Based on the Spatial Durbin Double-Difference Model (SDM-DID)

This study employs the Spatial Durbin Double-Difference Model (SDM-DID) for parameter estimation. Following the baseline regression, the estimated results of the spatial spillover effects of the SDM-DID model are decomposed into direct and indirect effects. The direct effect refers to the change in the endogenous variable of a region that directly impacts the dependent variable within that region; in other words, the direct effect measures the extent to which changes in the independent variable affect the dependent variable without considering influences from other regions, thus termed the “local effect.” Conversely, the indirect effect represents the influence of changes in the endogenous variable of one region on the dependent variable of neighboring regions through spatial spillover effects, reflecting spatial interdependence—specifically, how activities in one region affect surrounding areas, thus referred to as the “neighboring effect”. Decomposing the spatial spillover effects of the SDM-DID helps clarify the contributions of each effect, facilitating better interpretation of the model’s outputs. For policymakers, distinguishing between direct and indirect effects aids in understanding the potential scope of policy implications, allowing for the assessment of local direct impacts and possible cross-regional influences.

After processing the model through the aforementioned steps, the final results are reported in

Table 6. Parameter estimation results for Model 3 and Model 6.

	Model 3			Model 6
	IER			IER
	Local Effects	AdjacentEffects	Total Effects	Local Effects	AdjacentEffects	Total Effects
DID	0.0134 ***	0.0410	0.0544	−0.00128	0.00690	0.00561
	(3.79)	(1.16)	(1.51)	(−0.27)	(0.15)	(0.11)
AI	0.768 ***	−0.262 *	0.505 ***	0.668 ***	−0.172	0.496 **
	(44.40)	(−1.89)	(3.54)	(26.23)	(−0.81)	(2.35)
DID × AI				0.104 ***	−0.0382	0.0656
				(5.05)	(−0.20)	(0.34)
GI	0.0264	−0.0854	−0.0590	0.00119	−0.147	−0.146
	(0.88)	(−0.43)	(−0.29)	(0.04)	(−0.78)	(−0.75)
Att	−0.00350	−0.0234	−0.0269	−0.000887	−0.0471	−0.0480
	(−0.26)	(−0.34)	(−0.38)	(−0.07)	(−0.73)	(−0.71)
RIS	−0.000209 *	0.00110	0.000894	−0.000223 *	0.000694	0.000471
	(−1.75)	(0.84)	(0.66)	(−1.91)	(0.53)	(0.35)
GF	−0.00921	0.0560	0.0468	−0.00429	0.0911 **	0.0868 **
	(−0.76)	(1.26)	(1.05)	(−0.35)	(2.15)	(2.01)
ρ	0.276 *			0.282 *
	(1.71)			(1.77)
Variance
lgt_theta	−2.399 ***			−2.485 ***
	(−15.44)			(−16.12)
sigma2_e	0.000213 ***			0.000197 ***
	(12.49)			(12.52)
N	390			390
${\mathrm{R}}^2$	0.870			0.863

t statistics in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01.

. The findings indicate that the direct effect of the innovative provinces policy variable (DID) is 0.0134, which rejects the null hypothesis of a direct effect coefficient of zero at the 1% significance level. This suggests that the innovative provinces policy can exert a significant positive direct effect on the resilience of regional innovation ecosystems. By optimizing resource allocation, strengthening industry–academia–research collaboration, and enhancing knowledge flow, this policy can improve the adaptive capacity and resilience of regional innovation ecosystems, enabling them to rebound swiftly and sustain development even in the face of challenges such as shortages of innovative elements, barriers to knowledge flow, and heightened technological obstacles. This result confirms the mechanism hypothesis H1.

Simultaneously, the direct effect coefficient of the regional artificial intelligence development index (AI) is 0.768, which also passes the 1% significance test, indicating that regional artificial intelligence development significantly promotes the enhancement of the resilience of local innovation ecosystems. Artificial intelligence enhances information-processing capabilities, accelerates knowledge dissemination and application, and improves production efficiency through automation and intelligence, enabling regional innovation entities to better adapt to rapidly changing technological environments and market demands and thereby strengthening the entire system’s capacity to withstand external shocks and recover quickly. This result validates the mechanism hypothesis H2.

In the estimation of indirect effect coefficients, the indirect effect coefficient of AI is −0.262, which passes the 10% significance test. This finding suggests that regional artificial intelligence development has a significant negative spatial spillover effect on the resilience of regional innovation ecosystems. Possible reasons for this include the fact that an increase in the resilience of a regional innovation ecosystem implies that a region can better cope with challenges such as shortages of innovative elements, barriers to knowledge flow, and technological obstacles, quickly restoring normal innovation activity after encountering these external disturbances. However, when a region’s innovation capacity significantly strengthens, it may create a “siphoning effect” on neighboring areas, attracting talent, capital, and resources to concentrate in that area, which could lead to a more severe shortage of innovative elements in surrounding regions. Additionally, highly developed artificial intelligence technologies and applications may raise the entry barriers for the entire industry, making it challenging for neighboring areas to keep pace with technological innovations and thereby exacerbating the technological gap and developmental imbalances among different regions and further weakening the innovation resilience of these areas.

On the other hand, as leading regions continue to advance technological innovation and establish more sophisticated innovation ecosystems, they often create relatively closed knowledge networks and collaborative relationships, which may limit the speed and scope of knowledge and technology diffusion to other regions, preventing the effective utilization of advanced technologies and concepts to bolster their own innovation ecosystems. Thus, in this context, while the development of artificial intelligence in one region may enhance its internal innovation resilience, it may simultaneously exert a negative spatial spillover effect on other regions, inhibiting the healthy development of their innovation ecosystems.

Moreover, the empirical analysis reveals that the indirect effect of DID is not significant, indicating that the construction of innovative provinces struggles to exert notable spatial spillover effects on the resilience of regional innovation ecosystems. Possible reasons include the fact that innovative provinces often emphasize leveraging local characteristics and the development of advantageous industries, which may result in innovation models and experiences that are challenging to replicate in other regions. Significant differences in economic structure, industrial foundation, and cultural backgrounds among regions can pose adaptability issues when successful innovation models are disseminated across regions. Additionally, despite the aim of innovative provinces being to enhance the openness and inclusivity of local innovation systems, the existence of administrative boundaries may still present certain institutional barriers that hinder deeper cross-regional cooperation and the free flow of knowledge and technology. Furthermore, the focus of innovative province construction may lean more toward building tightly knit internal innovation networks rather than broadly expanding cross-regional partnerships. This relatively closed network structure, while enhancing local innovation resilience, restricts interaction with external regions, thereby affecting the emergence of spatial spillover effects.

To further explore the joint impact of innovative province policies and artificial intelligence on the resilience of regional innovation ecosystems, this study introduces an interaction term (DID × AI) within the original SDM-DID model for parameter estimation. The empirical results indicate that the direct effect coefficient of DID × AI is 0.104 and is significantly positive at the 1% level. This finding suggests that the policy for constructing innovative provinces not only has a direct positive influence on the resilience of regional innovation ecosystems but also positively moderates the beneficial effects of artificial intelligence development on the resilience of these ecosystems.

4.1.4. Parallel Trend Test

The assumption of parallel trends is a prerequisite for using the double-difference model. Therefore, this study refers to Zhu Chen et al. to set up a parallel trends test, as defined in Equation (9) [66]:

$$\begin{gathered} {Res}_{it}={\lambda }_{it}+{{\theta }_1{policy}_{i\left(t-12\right)}+{\theta }_2{policy}_{i\left(t-11\right)}+\dots +{\theta }_{12}{policy}_{it}+\dots +{\theta }_{19}{policy}_{i\left(t+7\right)} \\ +{\theta }_{20}{policy}_{i\left(t+8\right)}+\alpha {AI}_{it}+\sum \beta {\mathit{X}}_{\mathit{i}\mathit{t}}+\gamma }_t+u_i+{\epsilon }_{it} \end{gathered}$$

(9)

In Equation (9), ${policy}_{i(t\pm n)}$ represent the virtual variables for the i region before and after n years of being approved for the pilot project of innovative province construction (with t−7 as the base period). If the coefficient θ of ${policy}_{i(t-n)}$ is not significant, while the coefficient θ of ${policy}_{i(t+n)}$ is significant, it indicates the existence of parallel trends between the treatment and control groups. Moreover, the construction of innovative provinces can generate significant policy effects on the regional innovation ecosystem resilience. Figure 2 illustrates the coefficients of ${policy}_{i(t\pm n)}$, demonstrating that this study’s quasi-natural experiment passes the parallel trends test.

4.2. Causal Inference Analysis Results Based on Double Machine Learning

4.2.1. Baseline Regression

Following the spatial double-difference analysis, this study employs a double machine learning model for baseline regression to further mitigate the curse of dimensionality in causal inference. In the parameter estimation using the double machine learning model, the model algorithm is set to random forests, with a data split ratio of 1:4, ultimately yielding the results shown in

Table 7. Baseline regression based on difference-in-difference model and double machine learning.

	Difference-in-Difference		Double Machine Learning
	IER	IER	IER	IER	IER
DID	0.0136 **	−0.00251	0.0582 ***		0.0582 ***
	(2.18)	(−0.34)	(5.58)		(5.58)
AI	0.753 ***	0.647 ***		0.818 ***	0.818 ***
	(25.07)	(13.35)		(13.79)	(13.79)
DID × AI		0.111 ***			0.281 ***
		(3.07)			(2.80)
_cons	0.00113	0.00116	0.00113	−0.000473	−0.000492
	(0.53)	(0.55)	(0.53)	(−0.35)	(−0.30)
Control Variables	yes	yes	yes	yes	yes
Fixed Region	yes	yes	yes	yes	yes
Fixed Time	yes	yes	yes	yes	yes
N	390			390
${\mathrm{R}}^2$	-			-

t statistics in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01.

. Additionally, to cross-validate the parameter estimation outcomes of the double machine learning model, this paper also reports the parameter estimates from the ordinary causal double-difference model in Table 7 to preliminarily explore the robustness of the empirical findings.

Table 7 indicates that in the baseline regression model, the coefficient of the innovative provinces policy variable (DID) regarding the explained variable, the resilience of regional innovation ecosystems (IER), is 0.0136, passing the 5% significance test. In the double machine learning model, this coefficient is 0.0582, significant at the 1% level. This result demonstrates that the pilot policy for innovative provinces significantly promotes the resilience of regional innovation ecosystems. By providing financial support, optimizing the innovation environment, and fostering collaboration among industry, academia, and research, the policy enhances the innovation capabilities and collaborative networks of enterprises within the region. This not only aids in bolstering the resilience of the regional innovation ecosystem against external shocks but also accelerates recovery from adverse impacts while continuously driving technological innovation and development. This outcome reaffirms the mechanism hypothesis H1.

Simultaneously, in the double-difference baseline regression model, the coefficient for artificial intelligence (AI) regarding the explained variable IER is 0.753, which passes the 1% significance test. In the double machine learning model, the coefficient for AI is 0.818, also significant at the 1% level. This finding indicates that artificial intelligence significantly enhances the resilience of regional innovation ecosystems. It further illustrates that AI can assist enterprises and government agencies in rapidly processing vast amounts of data, providing deeper insights, and facilitating more informed strategic decision-making, thereby reducing the negative impacts of uncertainty. Additionally, by automating repetitive tasks, AI can significantly improve production and service efficiency, freeing up human resources for more creative activities and thus strengthening the overall resilience and flexibility of the ecosystem. Furthermore, AI technology promotes cross-industry collaboration, stimulating new business models and technological solutions, enhancing the diversity and complexity of regional innovation ecosystems, and making them more resilient to shocks from any singular domain. Consequently, hypothesis H2 is again validated.

After conducting the baseline regression analysis, this study introduces the interaction term between the innovative provinces policy variable and artificial intelligence (DID × AI) into the model. The coefficients in the double-difference model and the double machine learning model are 0.111 and 0.281, respectively, both passing the 1% significance test. This result further confirms that the innovative provinces policy can exert a positive moderating effect on the resilience impact of artificial intelligence within regional innovation ecosystems.

4.2.2. Mechanism Path Analysis

After conducting the baseline regression, this study seeks to validate the mechanism path “Construction of Innovative Provinces → Artificial Intelligence → Resilience of Regional Innovation Ecosystems”. In this context, resilience is understood as the ability of regional innovation ecosystems to withstand external disturbances caused by shortages of innovative elements, barriers to knowledge flow, and increased technological obstacles, as well as their capacity to rapidly recover after short-term evolutionary stagnation. Therefore, this study conceptualizes resilience as a composite indicator based on the diversity, evolutionary potential, buffering capacity, fluidity, and coordination of regional innovation ecosystems. In this section, the analysis will further discuss the mechanism by which the construction of innovative provinces influences the various dimensions of resilience via artificial intelligence, specifically empirically validating the paths “DID → AI → IER_Diversity” “DID → AI → IER_Evolutionary”, “DID → AI → IER_Buffering” “DID → AI → IER_Mobility”, and “DID → AI → IER_Coordination”.

During the empirical process, this study continues to utilize the double machine learning model based on the random forest algorithm and employs stepwise regression for parameter estimation of the mechanism paths. Subsequently, the validity of the mediation effect paths is verified through Sobel tests, Aroian tests, and Goodman tests, with the results reported in

Table 8. Results of Mechanistic Pathway Examination.

	DID → AI → IER			DID → AI → IER_Diversity
	IER	AI	IER	IER_Diversity	AI	IER_Diversity
DID	0.0582 ***	0.0435 ***	0.0243 ***	0.0109 *	0.0435 ***	−0.00186
	(5.58)	(4.80)	(2.94)	(1.68)	(4.80)	(−0.29)
AI			0.769 ***			0.288 ***
			(12.06)			(4.87)
_cons	0.00113	0.00116	−0.000603	0.000447	0.00116	−0.000202
	(0.53)	(0.55)	(−0.45)	(0.35)	(0.55)	(−0.18)
Control Variables	yes	yes	yes	yes	yes	yes
Fixed Time	yes	yes	yes	yes	yes	yes
Fixed Individual	yes	yes	yes	yes	yes	yes
IntermediateProportion	57.5%			Complete Mediation
Sobel (Z-value)	4.462 ***			3.420 ***
Aroian (Z-value)	4.449 ***			3.384 ***
Goodman (Z-value)	4.476 ***			3.458 ***
	DID → AI → IER_Evolutionary			DID → AI → IER_Buffering
	IER_Evolutionary	AI	IER_Evolutionary	IER_Buffering	AI	IER_Buffering
DID	0.0462 ***	0.0435 ***	0.0248 ***	0.0509 ***	0.0435 ***	−0.00129
	(5.70)	(4.80)	(3.49)	(3.23)	(4.80)	(−0.09)
AI			0.484 ***			1.185 ***
			(6.61)			(7.13)
_cons	0.000876	0.00116	−0.000214	0.000348	0.00116	−0.00232
	(0.52)	(0.55)	(−0.17)	(0.10)	(0.55)	(−1.05)
Control Variables	yes	yes	yes	yes	yes	yes
Fixed Time	yes	yes	yes	yes	yes	yes
Fixed Individual	yes	yes	yes	yes	yes	yes
IntermediateProportion	45.7%			Complete Mediation
Sobel (Z-value)	3.886 ***			3.984 ***
Aroian (Z-value)	3.857 ***			3.957 ***
Goodman (Z-value)	3.915 ***			4.011 ***
	DID → AI → IER_Mobility			DID → AI → IER_Coordination
	IER_Mobility	AI	IER_Mobility	IER_Coordination	AI	IER_Coordination
DID	0.0511 ***	0.0435 ***	0.0168	0.0841 ***	0.0435 ***	0.0565 ***
	(3.09)	(4.80)	(1.13)	(4.59)	(4.80)	(3.06)
AI			0.778 ***			0.627 ***
			(6.21)			(6.81)
_cons	0.0000376	0.00116	−0.00171	0.0000582	0.00116	−0.00135
	(0.01)	(0.55)	(−0.72)	(0.02)	(0.55)	(−0.58)
Control Variables	yes	yes	yes	yes	yes	yes
Fixed Time	yes	yes	yes	yes	yes	yes
Fixed Individual	yes	yes	yes	yes	yes	yes
IntermediateProportion	Complete Mediation			32.5%
Sobel (Z-value)	3.799 ***			3.926 ***
Aroian (Z-value)	3.769 ***			3.898 ***
Goodman (Z-value)	3.830 ***			3.954 ***
N	390	390	390	390	390	390
${\mathrm{R}}^2$	-	-	-	-	-	-

* p < 0.10, ** p < 0.05, *** p < 0.01, same below.

.

The analysis indicates that the mechanism path “DID → AI → IER” constitutes a positive mediation effect path with a mediation proportion of 57.5%, passing all the significance tests. This result demonstrates that artificial intelligence can significantly mediate the relationship between the construction of innovative provinces and the resilience of regional innovation ecosystems, thereby validating mechanism hypothesis H6.

Additionally, the mechanism paths “DID → AI → IER_Evolutionary” and “DID → AI → IER_Coordination” also pass the Sobel, Aroian, and Goodman tests, representing partial mediation effects with mediation proportions of 45.7% and 32.5%, respectively. These results indicate that artificial intelligence positively mediates the relationship between the construction of innovative provinces and the evolutionary potential and coordination of regional innovation ecosystems. The policies for constructing innovative provinces provide robust support for the development of artificial intelligence, while the application of AI further enhances the evolutionary potential and coordination of regional innovation ecosystems. Specifically, the policies create a favorable environment for the development and application of artificial intelligence technologies through policy guidance and support. As a key technology, AI acts as a bridge in this process, effectively enhancing the evolutionary potential of regional innovation ecosystems by optimizing resource allocation, increasing research and development efficiency, and promoting technological innovation. This means that the application of artificial intelligence helps to increase innovation investment, improve innovation output, and enhance innovation efficiency. Moreover, AI strengthens the coordination of regional innovation ecosystems. By facilitating collaboration among universities, research institutions, and high-tech industries, AI reduces communication costs and improves cooperation efficiency, which, in turn, promotes the transformation of scientific and technological achievements and industry–academia–research collaboration. This synergistic effect helps to build a tighter and more efficient innovation network, enabling participants to better share resources, exchange information, and collectively advance technological innovation. Therefore, the policies for constructing innovative provinces not only directly drive the evolutionary potential and coordination of regional innovation ecosystems but also reinforce these effects through the mediating factor of artificial intelligence. The pivotal role of AI enhances the efficiency and collaboration of innovation activities within the region, laying a solid foundation for achieving sustainable innovation development.

Simultaneously, the mechanism paths “DID → AI → IER_Diversity”, “DID → AI → IER_Buffering”, and “DID → AI → IER_Mobility” also pass the Sobel, Aroian, and Goodman tests, indicating that artificial intelligence can significantly mediate the relationship between the construction of innovative provinces and the diversity, buffering capacity, and fluidity of regional innovation ecosystems. Specifically, the policies for constructing innovative provinces provide both policy support and resource allocation for artificial intelligence technologies, creating an environment conducive to their growth and application, which further promotes the development of various aspects of regional innovation ecosystems.

First, the application of artificial intelligence enhances the diversity of regional innovation ecosystems by introducing more intelligent solutions and technology platforms, which fosters collaboration and development among innovation entities such as universities, research institutions, and high-tech enterprises, thereby increasing the number of innovation entities and diversifying the innovation network. Second, AI enhances the buffering capacity of regional innovation ecosystems; the application of AI technologies helps accumulate more knowledge and skills, providing the ecosystem with the capability to withstand external shocks. Finally, AI promotes the flow of key elements such as technology, capital, knowledge, and talent through intelligent information management and resource-sharing platforms. This flow not only strengthens internal interaction and cooperation within the region but also provides the necessary motivation and support for innovation activities, thereby enhancing the vitality and resilience of regional innovation ecosystems. Additionally, all three of these mechanism paths exhibit complete mediation effects, suggesting that the current implementation of the innovative province pilot policy is still in its early stages, resulting in limited direct effects on the cultivation of innovation entities, the accumulation of technological knowledge, and the flow of innovative elements within regional innovation ecosystems. However, through the mediation of artificial intelligence, the policies can establish effective channels of influence regarding the diversity, buffering capacity, and fluidity of these ecosystems.

4.3. Robustness Testing

This study necessitates a robustness check of the empirical results pertaining to the aforementioned mechanism pathway.

Table 9. Results of the robustness test.

Test Item	Effect/Dependent Variable	IER	AI	IER	Covariate	Fixed Effect	Intermediate Proportion	Effect Test
Exclude Jiangxi Province Sample	DID	0.0499 ***	0.0367 ***	0.0215 ***	yes	yes	56.9%	pass
		(5.28)	(4.54)	(3.07)
	AI			0.774 ***
				(12.21)
	_cons	0.00122	0.00123	−0.000581
		(0.57)	(0.58)	(−0.43)
Exclude the Year of Policy Implementation	DID	0.061 ***			yes	yes	63.8%	pass
		(4.19)
	AI		0.564 ***
			(10.30)
	_cons	0.061 ***	0.564 ***	0.342 **
		(4.19)	(10.30)	(2.57)
Adjust the Sample Split Ratio to 1:3	DID	0.0627 ***	0.0407 ***	0.0317 ***	yes	yes	48.4%	pass
		(5.47)	(3.58)	(4.51)
	AI			0.745 ***
				(10.44)
	_cons	0.0000308	0.000228	−0.000256
		(0.01)	(0.11)	(−0.18)
Adjust the Sample Split Ratio to 1:7	DID	0.0502 ***	0.0360 ***	0.0196 ***	yes	yes	55.1%	pass
		(5.27)	(3.91)	(2.81)
	AI			0.769 ***
				(12.23)
	_cons	0.00153	0.00218	−0.000488
		(0.74)	(1.10)	(−0.36)
Use Lasso Regression algorithm	DID	0.0763 ***	0.0855 ***	0.0131 ***	yes	yes	82.8%	pass
		(6.35)	(6.01)	(2.84)
	AI			0.739 ***
				(26.50)
	_cons	−0.000290	−0.000693	0.000222
		(−0.13)	(−0.25)	(0.22)
Use Gradient Boosting algorithm	DID	0.0463 ***	0.0358 ***	0.0203 ***	yes	yes	56.1%	pass
		(4.89)	(4.05)	(2.65)
	AI			0.725 ***
				(10.22)
	_cons	0.00218	0.00186	0.00104
		(1.19)	(0.99)	(0.84)

t statistics in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01.

presents the robustness test results for the principal mechanism path “DID → AI → IER”. The robustness tests outlined in Table 9 are categorized into four distinct evaluations:

(1)

Exclusion of Controversial Samples: Due to the spontaneous initiation of innovation province policies in Jiangxi Province, China, without endorsement from the central government, there remains ongoing debate regarding Jiangxi’s categorization as an innovation province. Consequently, this study excludes the sample data from Jiangxi and reanalyzes the mechanism pathway “DID → AI → IER”.

(2)

Exclusion of the Policy Implementation Year: In quasi-natural experiments, the treatment regions may not be subject to the policy for the entirety of the implementation year. Classifying the implementation year as part of the treatment period could potentially skew the results of the mechanism pathway examination. Thus, this study eliminates the sample data from the treatment regions for the year of implementation and reassesses the mechanism pathway “DID → AI → IER”.

(3)

Adjustment of Sample Split Ratios: In conducting dual machine learning model analyses, observational data are typically randomized into several distinct subsets or folds. One portion is utilized to estimate nuisance parameters, referred to as “first-stage” data, while another portion estimates the parameters of interest, termed “second-stage” data. In previous analyses, a sample split ratio of 1:4 was employed; for the robustness check, this ratio is adjusted to 1:3 and 1:7, followed by a re-evaluation of the mechanism pathway “DID → AI → IER”.

(4)

Substitution of Machine Learning Algorithms: In prior analyses, the random forest algorithm was utilized as the machine learning technique for the dual machine learning model, adept at capturing complex nonlinear relationships within the data and exhibiting strong resilience to outliers and noise. However, other algorithms, such as lasso regression, can enforce L1 regularization to yield sparsity, thereby excluding insignificant variables from estimation, while gradient boosting demonstrates formidable predictive capabilities and flexibility. To comprehensively assess the robustness of the research model across various algorithms, this study employs both lasso regression and gradient boosting algorithms to re-estimate the parameters of the mechanism pathway “DID → AI → IER”.

4.4. Extended Analysis: Mediating Effects of Intelligent Investment, Intelligent Application, Intelligent Innovation, and the Market

To further validate H3, this study employs a counterfactual framework to examine the resilience-enhancing effects of regional innovation ecosystems driven by the construction of innovation provinces and artificial intelligence. Intelligent investment, intelligent application, and intelligent innovation, along with the market, represent distinct facets of the regional AI development process. They each exert unique influences within the transmission mechanism from “construction of innovation provinces” to “resilience of regional innovation ecosystems”. Given that causal mediation effects can elucidate the specific roles of mediating variables in causal inference analyses and reveal the heterogeneous effects of policy in treatment and control groups through a counterfactual lens, this study adopts this methodology.

Drawing on the approach of Farbmacher et al. [67], this study conducts a causal mediation effect analysis based on dual machine learning. Equation (10) illustrates the principles of causal mediation analysis, positing that the explained variable is simultaneously determined by the policy treatment status and the mediating variables, which in turn depend on the treatment status. This allows us to derive whether the mediating pathways in the treatment group (including both direct and indirect effects) are valid and if the introduction of the policy can establish these paths in the control group as well.

$$\left\{\begin{array}{c}Direct\left(1\right)=E\left(Res\left(1,M\left(1\right)\right)-Res\left(0,M\left(1\right)\right)\right)\\ Direct\left(0\right)=E\left(Res\left(1,M\left(0\right)\right)-Res\left(0,M\left(0\right)\right)\right)\\ Indirect\left(1\right)=E\left(Res\left(1,M\left(1\right)\right)-Res\left(1,M\left(0\right)\right)\right)\\ Indirect\left(0\right)=E\left(Res\left(0,M\left(1\right)\right)-Res\left(0,M\left(0\right)\right)\right)\end{array}\right.$$

(10)

Table 10. Mediation effect analysis.

	Total Effect	Treatment Group Direct Effect	Control Group Direct Effect	Treatment Group Indirect Effect	Control Group Indirect Effect
DID → Input → Com	0.124 ***	0.063 ***	0.023 ***	0.101 ***	0.061 ***
DID → Application → Com	0.047 **	0.063 ***	0.025 ***	0.030 **	0.047 **
DID → I&M → Com	0.082 ***	0.038 ***	0.087 ***	0.043 ***	0.049 ***

** p < 0.05, *** p < 0.01.

reports the results of the causal mediation effect analysis with intelligent investment (Input), intelligent application (Application), and intelligent innovation and market (I&M) as mediating variables, thereby scrutinizing the three mechanism pathways: “DID → Input → IER”, “DID → Application → IER”, and “DID → I&M → IER”. The findings in Table 10 confirm that the direct and indirect effects of all three mechanisms are established not only within the treatment group but also within the control group.

This outcome indicates that, first, the innovation province policy encourages investment in the field of artificial intelligence, encompassing both financial support and advantageous policy conditions for the research and development of AI technologies. An influx of resources into the AI sector stimulates the research and development of related technologies and products, thereby enhancing the resilience and technological capabilities of enterprises within the region. As technological levels rise, local enterprises become better equipped to navigate market uncertainties, such as the risks associated with technological innovation failures or advancements by competitors. Furthermore, the application of AI technologies can enhance production efficiency and reduce operational costs, thereby enabling firms to maintain stability in the face of economic fluctuations and ultimately bolstering the resilience of the entire regional innovation ecosystem.

Second, the innovation province policy advocates for the integration of AI technologies into existing products and services. The application of AI can improve product quality, enhance service efficiency, and give rise to new business models. Such applications not only enhance the resilience of existing industries but also have the potential to spawn new industries and employment opportunities, thereby invigorating the overall regional economy. Moreover, the implementation of AI fosters the diffusion of knowledge and technology, strengthening collaboration and communication among various innovation entities and forming a more cohesive innovation network. The establishment of this network helps mitigate the blockage of knowledge channels and reduces technological barriers, enabling the regional innovation ecosystem to more effectively mobilize resources and respond swiftly to challenges.

Additionally, innovation province policies typically encourage firms to increase R&D investments, support innovative projects, and create favorable opportunities for the commercialization of new technologies. This accelerates the iterative upgrading of AI technologies and stimulates market demand, guiding capital and social resources toward areas with developmental potential. As AI technologies advance and markets expand, interactions between enterprises and research institutions within the region will become increasingly frequent, generating a positive feedback loop. This loop enhances the flexibility and adaptability of the regional innovation ecosystem, enabling it to swiftly adjust strategies in response to external disturbances such as technological changes and shifts in market demand, thereby maintaining or restoring normal operations.

Furthermore, the effective establishment of the institutional mechanism linking “innovative provincial development → artificial intelligence → resilience of regional innovation ecosystems,” along with the emergence of a series of policy pilot experiences, has created favorable conditions for the dissemination of this mechanism to non-pilot regions. Thus, in the analysis based on a counterfactual framework, the three mechanisms—”DID → Input → IER”, “DID → Application → IER”, and “DID → I&M → IER”—are not only valid within the treatment group but also hold true in the control group, which remains unaffected by policy intervention.

5. Conclusion and Policy Recommendations

5.1. Research Findings

• This study investigates the impact of innovation province policies and artificial intelligence on the resilience of regional innovation ecosystems, as well as the mechanisms by which the construction of innovation provinces indirectly influences this resilience via the development of regional AI. The key conclusions are as follows. Impact of Innovation Province Policies on Regional Innovation Ecosystem Resilience: Innovation province policies exert a significant positive influence on the resilience of regional innovation ecosystems. By providing financial support, optimizing the innovation environment, and facilitating collaboration between academia and industry, these policies enhance the innovative capabilities and collaborative networks of enterprises, thereby increasing the system’s capacity to withstand external shocks and recover swiftly.
• Impact of Artificial Intelligence on Regional Innovation Ecosystem Resilience: Artificial intelligence significantly enhances the resilience of regional innovation ecosystems. AI boosts information-processing capabilities, accelerates the dissemination and application of knowledge, and improves production efficiency, thereby augmenting the adaptive capacity of innovation entities regarding rapidly changing environments, which in turn bolsters the resilience of the entire system.
• Synergistic Effects of Innovation Province Policies and Artificial Intelligence: Innovation province policies positively moderate the role of AI in enhancing the resilience of regional innovation ecosystems. The support and promotion of AI technology further amplify its beneficial effects on the resilience of these ecosystems.
• Analysis of Spatial Spillover Effects: The development of artificial intelligence presents a significant negative spatial spillover effect on the resilience of regional innovation ecosystems. This phenomenon may arise from leading regions attracting innovation factors from surrounding areas, exacerbating inter-regional disparities and thus suppressing the resilience of adjacent ecosystems. The difficulty that innovation province construction has in generating notable spatial spillover effects is likely due to the complexities of policy implementation and regional disparities, which restrict the cross-regional flow of knowledge and technology.
• Mechanism Pathway Analysis Conclusions: There exists a significant mediating effect of AI in enhancing the resilience of regional innovation ecosystems through the construction of innovation provinces. AI plays a positive mediating role for aspects such as diversity, evolutionary capacity, buffering ability, fluidity, and coordination of regional innovation ecosystems. Additionally, the extended analysis based on causal mediation effects indicates that intelligent investment, intelligent application, and intelligent innovation, alongside the market, as different facets of AI development, all exert significant mediating effects in the process of enhancing ecosystem resilience driven by innovation province policies.

5.2. Policy Recommendations

Based on the aforementioned research findings, the following policy recommendations are proposed:

• Strengthen the Implementation of Innovation Province Policies: Treat innovation province policies as a crucial lever for enhancing the resilience of regional innovation ecosystems. Continuously optimize the innovation environment to stimulate the innovative vitality of enterprises. Specific measures may include establishing specialized innovation funds to support R&D activities and academia–industry collaboration projects; creating innovation service platforms to provide technical consulting, project matching, and talent training services; and encouraging enterprises to engage in international scientific and technological cooperation and exchanges to elevate their positions within the global innovation network.
• Promote the R&D and Application of Artificial Intelligence Technologies: Position AI technologies as a key force for enhancing the resilience of regional innovation ecosystems, fostering deep integration with traditional industries. The government should formulate development plans for the AI industry, clarifying development objectives and key tasks; establish AI innovation centers to attract talent and resources, promoting breakthroughs in critical technologies; and implement the “AI+” action plan to guide enterprises in utilizing AI technologies for product upgrades and process optimization.
• Optimize the Synergistic Development of Innovation Province Policies and Artificial Intelligence: Emphasize the establishment of effective mechanisms for the collaborative development of innovation province policies and AI technologies, achieving a harmonious integration of policy guidance and technological drive. The government can first create platforms for policy–technology alignment to facilitate communication and collaboration between policymakers and technical experts; develop tax incentives and financial subsidy policies targeting AI technologies to reduce application costs for enterprises; and encourage universities and research institutions to offer AI-related programs and courses to cultivate a larger pool of specialized talent.
• Enhance Inter-Regional Cooperation and Coordinated Development: Strengthen inter-regional cooperation and coordinated development to achieve the optimal allocation of innovative resources and shared innovation outcomes. Concrete actions may include establishing cross-regional innovation cooperation mechanisms to facilitate technology transfer and outcomes conversion between different regions; encouraging enterprises to undertake cross-regional collaborative projects for the co-development of new products and technologies; and enhancing talent exchanges and training collaborations among different regions to promote resource sharing and complementary advantages.

5.3. Innovation and Limitations of This Study

5.3.1. Innovation

In comparison to previous studies, the primary innovations of this research are threefold:

• Integration of Key Elements: This study incorporates innovation province policies, regional AI development, and the resilience of regional innovation ecosystems into a unified analytical framework.
• Multidimensional Assessment of Ecosystem Resilience: The research employs a five-dimensional measurement of regional innovation ecosystem resilience, characterized by diversity, evolutionary capacity, buffering ability, fluidity, and coordination. A comprehensive evaluation of the resilience of regional innovation ecosystems across various provinces in China (excluding Tibet and the regions of Hong Kong, Macau, and Taiwan) is conducted.
• Mechanism Analysis and Empirical Research: Through mechanism analysis and empirical investigation, this study delineates the policy transmission pathway of “innovation province policies → regional AI development → regional innovation ecosystem resilience”. This pathway not only assesses the effectiveness of China’s regional innovation policies but also provides a framework that can be extended to other emerging market nations.

5.3.2. Limitations and Prospects

• Constrained by the research objectives, perspectives, and data availability, this study examines the role of innovative provinces within the framework of China’s provincial administrative divisions, utilizing regional macro data. However, lower administrative tiers and entities remain unexamined in this paper. Future inquiries could harness data scraping, field research, and text analysis to procure additional micro-level data sources, thereby re-evaluating the principal issues discussed herein from a novel and more focused vantage point.
• Grounded in the theory of innovative ecological niches, this paper investigates the impact of innovation-oriented provincial policies and regional artificial intelligence development on the resilience of regional innovation ecosystems. From the perspective of this theory, the regional innovation ecosystem is the highest-level unit of such systems. Future research could delve deeper into other levels of innovation ecosystems, such as industrial and corporate innovation ecosystems, thereby extending the conclusions drawn in this study to encompass heterogeneous units of innovation ecosystems.
• In light of the self-organizing, self-regulating, and self-reinforcing characteristics of regional innovation ecosystems, future studies could employ diverse theories such as system dynamics and dissipative structure theory to explore the abrupt and nonlinear changes in resilience resulting from the dynamic evolution of these systems.