Sage Journals: Discover world-class research

Abstract

The successful adoption of Fourth Industrial Revolution (4IR) frontier technologies is crucial for corporate and national sustainability, with knowledge transfer playing a key role in this process. Indeed, past research has demonstrated that global collaboration networks can generate innovative ideas and customised solutions to key development challenges across locations. This study explores global collaborative research networks related to 4IR frontier technologies. Specifically, we leverage the recent launch of ‘ChatGPT’, as a quasi-experiment to analyse globally collaborative networks among researchers. To do this we utilise bibliometric data from the Web of Science (WOS) over the period January to November 2023, 1 year after the launch of ChatGPT and apply social network analysis (SNA) to identify key knowledge disseminators, gatekeepers and ‘bridgers’ at the country, institutional and researcher levels. Our findings highlight the prominent role of the United States, United Kingdom, China and India, but also the increasing dominance of emerging nations like Saudi Arabia, Jordan and Malaysia indicating the importance of geographical and cultural proximity in these relationships. Prestigious institutions such as the National University of Singapore, Imperial College London and Stanford University are found to be central hubs focusing on AI, natural language processing and chatbots. We also find the emergence of regional hubs focusing on specialised areas of research related to ChatGPT in the areas of health, education, communications and linguistics. Our findings provide new insights into how collaborations between developed and emerging regions facilitate 4IR frontier technology adoption and suggest strategies to enhance these connections.

Keywords

ChatGPT collaborative research networks Frontier technologies Industry 4.0 technology adoption

1. Introduction

The Organisation for Economic Co-operation and Development (OECD) defines frontier technologies as technologies that can significantly reshape industry and communications but also have the potential to displace existing systems and processes [1]. Examples of more advanced Fourth Industrial Revolution (4IR) frontier technologies include artificial intelligence (AI), blockchain technologies, virtual reality (VR), 3-D printing, robotics, gene-editing, Internet-of-things (IoT), autonomous or near-autonomous vehicles, nanotechnologies and big data analytics [2]. Currently, many researchers and innovators are utilising these frontier technologies to address sustainable development goals (SDGs) priorities such as mitigating climate change, reducing poverty, eliminating hunger, increasing life expectancy, and increasing productivity and economic growth [3 –5]. Businesses are also utilising frontier technologies to enable faster prototyping, lower production costs and expand their global presence [6,7]. However, much of this depends on the capabilities of countries, corporations and researchers to understand, integrate and adapt new frontier technologies to their unique circumstances. While many developed countries, such as the United States, Canada, United Kingdom, Germany and Japan, continue to lead both the development and application of 4IR frontier technologies, many emerging economies are not far behind. China, India and South Korea remain notable examples [8,9]. Indeed, a lot depends on the collaborative research networks that are established between key stakeholders operating across localities. This will be essential for adopting, diffusing and customising frontier technologies, ensuring their benefits are realised by communities worldwide.

Past research has shown that diverse academic networks are key to generating innovative ideas and solutions, benefitting communities globally [10 –12]. Indeed, key institutions and individuals can also emerge as important gatekeepers and brokers of knowledge. Gatekeepers are actors that are characterised by a high degree of network centrality [13,14], which allows them to act as ‘bridgers’ connecting with other stakeholders operating in distinct and diverse networks, thereby facilitating both the gathering and diffusion of new knowledge in these communities [15 –17]. Brokers on the other hand are actors with ties to other collaborators who are not tied to other actors in a network [18]. In this way, they are in the best position to generate new combinations across networks, as well as exploit or manipulate the flow of information across network members. These actors tend to have relatively higher scores in betweenness centrality [19].

To date, few researchers have been able to examine the role of collaborative research networks in explaining the adoption of the diffusion of new 4IR frontier technologies. This article attempts to contribute to the literature on collaborative research networks, by focusing on research networks that exist in one of the latest 4IR frontier technologies, ChatGPT. Specifically, this article seeks to leverage the recent launch of the generative AI chatbox ChatGPT, to identify key research leaders and knowledge brokers, that have influenced both the growth and dissemination of research on this new technology globally. To do this we utilise information from ‘results-based’ collaborative relationships proxied by published peer-reviewed articles and combine bibliometric and social network analysis (SNA) methodologies to analyse the collaborative relationships that exist at a country, institutional and individual level. Based on our approach, we aim to answer several key research questions:

Research Question 1 (RQ1). Which are the leading countries, institutions and researchers that collaborate, particularly in high-impact research on ChatGPT?

Research Question 2 (RQ2). Who are the main knowledge brokers and gatekeepers, that can influence the dissemination of knowledge in this field?

Research Question 3 (RQ3). What are the key research communities and associated themes these stakeholders are focused on pursuing?

In this way, we can examine possible areas of specialisation across locations, institutions and among researchers. Finally, based on this analysis, we also hope to identify research gaps and opportunities for future research. Overall we aim to take the first steps in understanding research collaboration patterns on new frontier technologies, specifically ChatGPT, and in so doing we aim to understand, how technology adoption and diffusion can be increased across locations.

Our article is divided into six sections. In section 2, we review the literature on collaborative research networks and technology adoption, particularly of frontier technologies. In section 3, we provide an outline of the data and methodology. Results are provided in section 4 and discussions and implications for policy are in section 5. Finally, conclusions, study limitations and areas of further research are outlined in section 6.

2. Review of the literature

2.1. The rise of new Frontier technologies

The rise of the 4IR has seen the emergence of new technological breakthroughs, which increasingly take advantage of the expansion of digital technologies, faster and cheaper computer processing power, and increasing communications capabilities globally [20]. Notable breakthroughs in 4IR technologies include AI, 3D printing, robotics, nanotechnologies, gene-editing, satellite-based imaging sensing, Internet-of-things, autonomous or near-autonomous vehicles, nanotechnologies, big data analytics and distributed ledger technologies. These new technologies are also profoundly interlinked with advances in one, leading to technological jumps in the application of others [21]. Indeed, two notable characteristics of 4IR frontier technologies have been the rate of advancement and the scope of uses to which these new technologies are applied [22,23]. More recently, researchers have highlighted the next evolution in industry innovation called Industry 5.0, which entails humans utilising their expertise and creativity, co-working with powerful, smart and accurate machinery [24,25]. Recent examples of Industry 5.0 applications include cloud manufacturing, bionics and leveraging AI in such areas as intelligent healthcare and sustainable agriculture [25]. Due to the speed of innovation and wide scope of applications, these technologies have the potential to radically disrupt traditional business models, across multiple sectors and industry value chains, while simultaneously increasing the connectivity, collaboration and productivity of companies globally [4,26,27].

The growth and development of new frontier technologies have also been associated with a concomitant change in profitability and income, particularly among countries and companies that have successfully integrated these technologies [22,28]. However, the benefits of frontier technologies are not unequivocal, as deploying these new technologies also brings potential challenges. For countries that have not yet been able to integrate these new technologies, there is the potential to generate significant disparities in earnings across locations [28,29]. Uneven adoption of frontier technologies also has the potential to expand within-country inequalities [30,31]. This is particularly relevant for even the most developed countries, such as the United States, and emerging economies, such as China and India. Related to this is the increasing threat of job polarisation, and job displacement primarily among routine and low-skilled jobs, which can be easily automated [32,33]. Finally, there is the fear that 4IR and more recently Industry 5.0 technologies have the potential to upend traditional industrial development processes of emerging countries by making it more difficult for these economies to progress up the value chain from low value-added to higher value-added activities, which integrate these new technologies [34,35]. A lot therefore depends on countries and firms’ ability to successfully integrate new frontier technologies, to ensure that the benefits of these technologies can filter throughout their respective economies.

2.2. The rise of ChatGPT

One of the innovations coming out of the 4IR was AI-driven chatbots. Chatbots are computer programme algorithms, in which the main goal is communication with a person using text-based or auditory methods [36]. Early chatbots were developed to communicate by replying to similar keywords or interacting with users supporting specific topics or purposes. These early chatbots were frequently used in customer service, e-commerce, e-eduction, marketing, advertising, the entertainment industry and also for data collection [37]. Despite early difficulties in communicating their user acceptance and integration by companies and households alike has increased considerably. Indeed, such chatbots as Apple’s Siri and Amazon’s Alexa are now commonly used in daily life [37].

On 30 November 2022, Open AI launched one of the most advanced chatbots, ChatGPT 3.5, to the global public, made available through the Internet in multiple languages, at minimal costs to users. ChatGPT3.5 or ‘ChatGPT’ as it is commonly known falls in the family of Artificial Intelligence Generated User Content (AIGUC) frontier technologies. It utilises multiple programming technologies, such as deep learning, unsupervised learning, reinforcement learning from human feedback (RLHF), and instruction fine-tuning to process user queries, questions, commands or classification tasks and produces coherent, contextual suitable responses in natural language text in multiple languages. However, unlike other AIGUC chatbox solutions, what sets ChatGPT apart is its human-like ability to ‘talk’ and respond in smooth, natural, instant dialogues with the public through a free and easy-to-use interface [38]. In addition, ChatGPT can remember dialogue, detect, and understand human responses, and respond to follow-up questions, as is common in everyday human conversations [39]. The platform also performs exceptionally well, frequently generating highly accurate reactions to complex and ambiguous contexts, admitting mistakes, challenging incorrect premises, and rejecting inappropriate requests [8]. The latest version ChatGPT 4.0 launched in March 2023, offers new and advanced features that allow users to input text and visual images simultaneously and allow for more advanced multimodal functions [8,40].

The gradual evolution of ChatGPT has had significant impacts on multiple industries, particularly technology, health care and education, as users continue to find new solutions and applications where this new technology can be applied [41]. The demand and use of ChatGPT continue to grow exponentially, indeed ChatGPT amassed over 1 million registered users just a week after its launch [42]. There have been also a growing number of publications on ChatGPT and its application in many diverse themes ranging from health, medical sciences, information technology and education [43 –48]. However, little research has been completed focusing exclusively on the role of collaborative research networks in fostering adoption, and dissemination of this new technology across locations.

2.2. Technology adoption through collaborative research networks

One way technology adoption and ‘catch-up’ can be achieved is through collaborative research networks, both within and across locations. Indeed, in today’s modern world, scientists are no longer independent and isolated individuals working in faraway labs but rather communities and networks of experts seeking answers to shared problems, relevant to the societies that they serve. These issues frequently require a multidisciplinary approach with specialised skills or resources that only cross-disciplinary networked groups, involving government, industry and academia can provide [49]. Researchers from different nations who collaborate increase the capacity for knowledge exchange, research and development, industrial innovation, and ultimately economic growth [50,51]. This exchange is key to facilitating the diffusion of knowledge among researchers in emerging economies, particularly by introducing or adapting the latest technologies [52,53]. For these reasons, increasing attention is being placed on examining the structure and operation of collaborative research networks. One way to examine these networks is through ‘results-based’ peer-reviewed research, which can be used to trace co-authorship networks among these publications. Co-authorship network analysis facilitates understanding of the linkages among researchers and helps to identify dominant researchers and their relationship with other key researchers in their field [54]. Indeed, by combining bibliometric information with SNA methodologies we can examine the nature and structure of research networks and relate this to innovation performance proxied by research output [55,56]. Through this approach, we can better understand the growth and dynamics of collaborative research networks, identify critical focal nodes of research networks, and identify key nodes that act as gatekeepers or knowledge brokers of information whether at the level of individual researchers, institutions, or countries [57 –59]. When it comes to research on such topics related to 4IR frontier technologies such as AI, blockchains and robotics, understanding the nature of research networks becomes particularly important, not merely to identify thought leaders and influencers but also to understand possible drivers of collaboration between researchers from more developed, well-established institutions and those from emerging economies. Indeed, understanding the nature of these collaborative relationships can help explain the extent to which researchers from emerging economies increasingly work with their counterparts in developed countries to facilitate knowledge transfer and to create new ways these technologies can be applied in their communities.

2.3. Research on collaborative research networks on 4IR frontier technologies

Recent literature on collaborative research networks, centred on the adoption of frontier technologies, provides a nuanced perspective on emerging trends within this dynamic field. Mizukami and Nakano [60] recently traced coauthor research networks to analyse the cross-disciplinary integration of research across key 4IR technologies, AI, big data and the Internet of Things (IoT). A significant contribution of this study lies in its use of bibliometric, cluster and principal component analysis to identify researchers’ areas of expertise and how these specialisations were applied across key 4IR technologies. This analysis was conducted at both the institutional and national levels to reveal cross-disciplinary research themes. For example, in the field of AI, the authors observed similar cross-disciplinary patterns among researchers from China, Japan and South Korea, who focused on areas such as chemistry, clinical medicine, computer science and engineering. Likewise, in the domain of big data, comparable patterns were identified among researchers from Malaysia, Iran, Taiwan, Singapore, Brazil, Pakistan and Saudi Arabia. However, the study offers a limited explanation for why these cross-disciplinary trends emerge.

Other research papers focused on specific fields of 4IR technologies. For instance, Bindu et al. [61] builds on prior research on collaborative research networks in e-commerce, focusing on temporal changes in co-author networks and research themes from 2010 to 2017. Through cluster and network analysis of over 16,000 articles, the authors track shifts in co-authorship patterns and the evolution of major research themes over time. They confirm the central role of researchers from the United States and the United Kingdom as key nodes in co-author networks while highlighting the growing significance of Chinese researchers who collaborated with peers globally. However, the study offers limited insights into the underlying drivers of these collaboration patterns during the study period.

Moosavi et al. [62] conducted a systematic review using bibliometric and network analysis to explore key authors and collaboration trends in research focused on blockchain applications in supply chain management. Analysing 763 articles from 2010 to 2019, they identified the most productive author networks and core research themes associated with them. The most networked countries included the United States, United Kingdom, China, Hong Kong, Germany and France. A unique aspect of their research was the identification of primary funding sources, which were largely concentrated in institutions in mainland China. Their findings demonstrated blockchain’s potential to enhance transparency, traceability, efficiency and information security in supply chain management. However, a limitation of their study is the omission of key knowledge brokers in their co-author network analysis, which could have provided valuable insights into how information flows across the network’s main nodes.

One article most relevant to our approach is Maghsoudi et al. [63], who undertook a co-author network analysis at a country, institutional and researcher level focusing on the research theme of the use of AI in sustainable supply chain networks. Based on their analysis of 499 articles published over the period 2004 to 2023, they highlight key author networks and associated sub-themes researched during the survey period. They repeat this network analysis at the institutional and country level to identify centrally placed nodes. By associating key network nodes with research themes, the researchers are also able to identify areas of research specialisation across researcher, institutional and country-level networks. In addition, they utilise key measures of betweenness and closeness centrality to identify key knowledge brokers well placed to facilitate the dissemination of research to more distant network nodes.

Most recently there have been several reviews tracing the bibliometric footprint on ChatGPT and tangentially the research networks included therein [43]. One notable example is Mubin et al. [41] who examined 103 research papers published on the theme of ChatGPT up to March 2023. Based on bibliometric analysis themes examined by researchers included technology, education, health care and ethical and legal implications. Researchers based in the United States specifically, John Hopkins University and Massachusetts Institute of Technology (MIT) dominated publications and collaborations globally, although the authors highlighted that the limited period examined may not have allowed for sufficient collaborative networks to be established. A second example is Farhat et al. [43] who examined the scholarly footprint of 533 research articles on ChatGPT published over the period November 2022 to early June 2023. As part of their bibliometric review, they examined key collaborative networks at the country, institutional and researcher levels. They highlighted the importance of the United States, China, and India and such institutions of such institutions as Duke University, Tianjin Medical University China, and such researchers as Wang FY and Wang X. Unfortunately, given their bibliometric approach, they were unable to utilise the full ambit of social network measures, to identify key network brokers, ‘bridgers’, or research communities. Our article, therefore, offers an opportunity to build on this research by examining research published up to 1 year after the launch of ChatGPT, and also utilising key social network measures, to identify key network leaders, brokers and research communities.

Table 1.

Key research articles of collaborative research networks on 4IR Frontier technologies.

Article	Period	Focus of study	Number of papers	Methodology
Bindu et al. [61]	1999 to 2017	This article examines temporal changes in co-author networks researching the core theme of e-commerce.	16,203	Data visualisation, social network, and cluster analysis based on author, institution, and country-level publication data.
Moosavi et al. [62]	2010 to 2019	Bibliometric and co-author analysis on research on the use of blockchain supply chain management	763	Bibliometric analysis, co-citation, and co-author analysis
Mizukami and Nakano [60]	2000 to 2018	To analyse and compare the cross-disciplinary integration of research in key fields of 4IR technologies among, researchers, institutions, and countries.	Over 40,000	Bibliometric, cluster, principal component and social network analysis
Maghsoudi et al. [63]	2004 to 2023	Co-author network analysis on the core theme of the use of AI in sustainable supply chains	499	Bibliometric and social network analysis
Mubin et al. [41]	2023	To provide a comprehensive analysis of the current research landscape surrounding ChatGPT.	103	Bibliometric and data analysis based on Google Trends search patterns
Farhat et al. [43]	November 2022 to June 2023.	To complete a comprehensive bibliometric analysis of the scholarly footprint of ChatGPT during its early outbreak phase. Identifying top journals in the field, countries contributing to research, collaboration networks among researchers, and key thematic areas of study.	533	Bibliometric and scientometric review

IR– Industrial Revolution; AI – Artifical Intelliegence

Source: Authors Interpretation.

3. Data and methodology

Figure 1 outlines our research methodology adapted from Munoz et al. [64] and Maghsoudi et al. [63]. There are generally four stages in co-authorship analysis:

Retrieval of scientific publication

Standardisation of entries for authors and organisations

Bibliometric and SNA

Interpretation of results

The first stage is focused on the retrieval of the data to be used for the study. This entails building the keywords used for the search query and selecting the journal database for the search. During the first stage, we used WoS reference databases to generate our sample of articles. The WOS (Science Citation Index, Social Sciences Citation Index, and Arts and Humanities Index) has existed since the 1970s. It is considered one of the most comprehensive and prominent databases, as it covers journals and citations from more than 34,000 professional journals in multiple languages and practically all critical research fields [65,66]. It also offers advanced search parameters features including searches based on month of publication. In this case, the keyword used in the search is ‘ChatGPT’. Given that we wanted to focus on research papers that examined the application of this recently launched AI frontier technologies We limited our search to peer-reviewed journal articles, book chapters and reviews published in multiple languages during 2023 starting from 2 January to 30 November 2023, exactly 1 year after the launch of ChatGPT 3.5 [67,68]. To ensure that scientific articles relevant to our study focus and to reduce the number of duplicates, we follow Moher et al. [69] and Ramsawak et al. [70] by applying the four-stage Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guidelines for selecting articles. Our search resulted in 1731 articles. After applying our inclusion and exclusion criteria, we excluded 138 articles. In stage 3, we reviewed titles, abstracts and keywords to determine the articles’ eligibility, removing papers not aligned with our overall research theme and eliminating duplicates and false positives. Based on this, our final sample was reduced to 1227.

Figure 1.

Methodological approach.

During the second stage, we cleaned the data. Two of the critical issues that need to be addressed when cleaning the data are homonymy (two authors with the same name) and synonymy (the existence of different variations on an author’s name) [71]. Indeed, large differences have been found in metrics between co-authorship networks when comparing network patterns based on full and short names [72,73]. There can also be similar inconsistencies in the naming of institutions and countries. A common example is the use of country abbreviations like US, USA and United States, which, if not standardised, can be treated as distinct country locations in the analysis. To address this issue we utilise Microsoft Excel and Bibliometrix R-Package developed by Aria and Cuccurullo [74] to disaggregate and clean the data. Next, we follow Wagner and Leydesdorff [58] and Wagner [75] by creating symmetric co-occurrence tables, where cells outside the main diagonal reflect the relationship between two nodes. Co-occurrence tables were created at the country, institutional and researcher levels to facilitate SNA.

In the third stage, we apply SNA methods and metrics to explore the network of researchers, institutions and countries, each representing a unique node within the research network. SNA serves as a critical tool for identifying the characteristics of these nodes, as well as the relationships and connections between them. It enables the quantification of network structures, helping to determine the most influential actors, key information gatekeepers, and the flow of both tangible and intangible resources across the network. SNA also allows for visualisations of social network patterns among target groups [76,77].

3.1. Utilising SNA measures

Common metrics used in SNA include measures of density, centrality and visual representations of the network structure.

3.1.1. Density

Network density measures the level of connectivity among nodes by calculating the percentage of existing links relative to the maximum possible links in the network. The density value can be normalised to range from zero, indicating minimal connections, to one, representing a fully connected network [78]. Density measures are typically used to estimate the extent nodes can communicate directly with others.

3.1.2. Centrality

Centrality measures the extent to which links are concentrated in one or a few nodes in the network. There are several measures of centrality utilised in co-author analysis.

Degree centrality measures the number of connections a node has with other nodes. The more ties a node possesses, the greater its power, prestige and influence within the network. When two nodes exhibit a higher degree of centrality, the likelihood of being linked to another node through a shared connection increases [79].

Betweenness centrality measures how often a node lies on the shortest path between other nodes, effectively acting as a bridge to other nodes. Nodes with high betweenness centrality are considered central, as they play a key role in influencing and controlling the flow of information by connecting different groups within the network [79 –81].

Closeness centrality is based on the degree to which a node is close to all other nodes in a network. Nodes with high closeness centrality tend to obtain information better than other nodes or can also influence other nodes more easily [82].

Eigenvalue centrality emphasises nodes connected to other influential nodes, showing how well-integrated a node is within the larger network.

Network structure and identification of network communities – Finally visualisation of networks, remains a key element of SNA. To do this we utilise Gephi software to build network maps. An essential aspect of our network analysis is identifying communities within networks. Communities are defined as networks of entities that are more closely connected than other entities within the dataset [83]. The intuition behind the formation of research communities is based on the observation that some entities may tend to interact more relative to other entities in the group. These communities in social networks are analogous to clusters in research networks. We follow Chhonker et al. [84] by utilising the Newman-Girvan community detection algorithm [85] to identify dominant communities within the network structure, at the country, institutional and researcher levels. As a means to assess the extent of thematic specialisation, as an extension of the research we will also examine communities of keywords, to identify possible network linkages among core research themes.

In the final stage, we analyse the results with an emphasis on addressing our primary research questions. This involves identifying key influential nodes, knowledge gatekeepers and research communities to explore how research collaborations can facilitate the adoption and dissemination of innovation related to 4IR frontier technologies across different communities.

4. Results

4.1. Descriptive statistics

Based on our preliminary results we confirm a total of 1227 documents of which 18.8% (231) were single-authored articles, with an average of 3.9 co-authorships per article. Roughly, 26.6% of the articles included international co-authorships (see Table 2).

Table 2.

Summary statistics of articles.

	N
Total number of documents	1227.0
Authors	4086.0
Number of Single-authored documents	231.0
Co-Authors per Doc	3.9
International co-authorships %	26.6

Source: Authors Estimate.

4.2. Country-level analysis

Table 3 summarises the results of key country-level network measures. We begin our analysis by identifying the top 10 countries, where researchers on ChatGPT were located based on each of our key SNA measures. Next, we examine our network graph and identify key country-level network communities to uncover possible spatial patterns in collaborative research.

Degree Centrality : This metric reflects the overall importance of these countries, in generating research with other countries in the network on ChatGPT. United Kingdom, China, India, United States and France were found to score the highest in terms of network connection measured by degree centrality and emerged as the most prolific collaborators in this group.

Closeness Centrality : This indicator reflects the average geodesic distance (shortest path) between an individual and all other members within the network’s largest connected component (giant component). Essentially, it identifies individuals strategically positioned to efficiently access and disseminate information. The United States, the United Kingdom and France are found to be key developed countries included in this network. However, India, China, Qatar and the United Arab Emirates are also network disseminators.

Betweenness Centrality : This metric captures the relative importance of countries as knowledge gatekeepers, acting as bridges or brokers to other network members. Not surprisingly, the United States, United Kingdom, India, Germany and China scored high in terms of betweenness centrality.

Table 3.

Network measures for the top 10 countries.

	Degree	Score	Betweenness	Score	Closeness	Score	Eigencentrality	Score
1	United Kingdom	57	India	0.159	United Kingdom	0.761	United Kingdom	1.000
2	United States	55	United Arab Emirates	0.126	United States	0.748	United States	0.971
3	India	53	United States	0.101	India	0.736	India	0.937
4	China	46	United Kingdom	0.100	China	0.692	China	0.931
5	United Arab Emirates	45	France	0.049	France	0.686	Australia	0.904
6	France	45	Qatar	0.047	United Arab Emirates	0.680	France	0.895
7	Australia	43	Saudi Arabia	0.041	Australia	0.675	Spain	0.852
8	Qatar	40	Jordan	0.036	Qatar	0.659	Qatar	0.849
9	Spain	39	China	0.034	Spain	0.654	United Arab Emirates	0.821
10	Germany	35	Spain	0.032	Saudi Arabia	0.634	Italy	0.814

Countries in bold appear in the top ten of all key social network measures.

Source: Authors Estimate.

Country-level network linkages are presented in Figure 2 and illustrate the collaborative landscape among countries in producing articles related to the ‘ChatGPT’. It comprises 84 nodes representing different countries and 701 edges denoting collaborative ties between them, this network embodies a collaborative structure that showcases the interconnectivity among nations in generating scholarly work on ChatGPT. With a graph density of 0.201, the network highlights a substantial level of interconnectedness among the countries involved in this domain. Moreover, the average clustering coefficient, standing at 0.729, suggests a high tendency for nodes to form clusters or communities within the network, indicative of cohesive collaboration among countries with shared research interests in ChatGPT research.

Figure 2.

Co-authoring network of countries.

The colours present in Figure 2 depict the results obtained from the detection community, summarised features of which can be observed in Table 4.

Table 4.

Information about communities in countries’ networks.

Main countries	Community colour	Community name	Number of nodes	Number of edges	Density
United Kingdom, United States, China, France, Australia, Spain, Germany, Italy, Switzerland, Netherlands, Canada, New Zealand	Purple	Global AI Powerhouses	28	176	0.47
India, Norway, Nigeria, Russia, Thailand, Peru, Nepal, Kazakhstan	Blue	Diverse Global Contributors	15	23	0.22
Arab Emirates, Qatar, Saudi Arabia, Jordan, Egypt, Oman, Malaysia, Pakistan, Iraq	Green	Middle East & Asia Collaborators	24	60	0.22
Singapore, Mexico, Greece, Japan, Argentina, Iran, Portugal, Slovenia, Colombia, Ethiopia, Vietnam, Indonesia	Orange	Emerging AI Hubs	14	43	0.47
Poland, Lithuania, Hungary	Dark green	Eastern European AI Cooperation	3	2	0.66

Source: Authors Estimate.

Considering the arrangement of countries within each of the communities, the following analyses are presented for each of the communities:

4.2.1. Community 1: ‘global AI powerhouses’ (purple)

This community represents a coalition of major countries actively contributing to AI research, development and publications related to ChatGPT. The high density suggests strong collaboration among these countries. It’s noteworthy that the United Kingdom, United States, China and other leading nations in AI are part of this community, indicating a global synergy in advancing AI technologies.

4.2.2. Community 2: ‘diverse global contributors’ (blue)

This community comprises a diverse set of countries with varied contributions to ChatGPT-related articles. The lower density suggests less interconnectedness compared with the first community, but it highlights the global reach and inclusivity of the research network. Countries like India, Russia and Nigeria in this community showcase a broad spectrum of AI research engagement.

4.2.3. Community 3: ‘Middle East & Asia collaborators’ (green)

This community represents collaboration among Middle Eastern and Asian countries in the AI domain, showing regional synergy. The moderate density suggests a balanced level of cooperation. The presence of countries like Saudi Arabia, United Arab Emirates and Pakistan emphasises the growing significance of AI research in the Middle East and Asia.

4.2.4. Community 4: ‘emerging AI hubs’ (orange)

This community signifies a mix of countries from different continents, collaborating in the realm of AI research. The high density implies strong ties among these nations, fostering a vibrant exchange of ideas. Singapore, Japan and Iran contribute to the community’s dynamism, highlighting the diverse perspectives shaping AI development.

4.2.5. Community 5: ‘Eastern European AI cooperation’ (dark green)

This community reflects a compact group of Eastern European countries with a high density, indicating close collaboration in AI research related to ChatGPT. Despite having fewer nodes, the high density suggests a focused and efficient network, with Poland, Lithuania and Hungary actively engaged in joint endeavours within the AI landscape.

4.3. Network of scientific institutions

Table 5 provides a summary of key network measures at the institutional level. We start by identifying the leading institutions within the network, followed by an analysis of network graphs and communities.

Table 5.

Key institutions for collaboration.

	Degree	Score	Betweenness	Score	Closeness	Score
1	NATIONAL UNIVERSITY OF SINGAPORE*	127	NATIONAL UNIVERSITY OF SINGAPORE ^a	84.00	MONASH UNIVERSITY ^a	1.000
2	UNIVERSITY OF COLORADO	97	MAYO CLINIC ^a	76.00	PENINSULA HEALTH	1.000
3	UNIVERSITY OF CENTRAL FLORIDA	97	STANFORD UNIVERSITY ^a	74.42	CHANDIGARH UNIVERSITY	1.000
4	QATAR UNIVERSITY	97	IMPERIAL COLLEGE LONDON ^a	68.17	JOSEPH AYO BABALOLA UNIVERSITY	1.000
5	UNIVERSITY OF NEW SOUTH WALES ^a	96	ICAHN SCHOOL OF MEDICINE AT MOUNT SINAI ^a	36.00	VANDERBILT UNIVERSITY MEDICAL CENTRE	1.000
6	UNIVERSITY OF BRISTOL ^a	90	TSINGHUA UNIVERSITY ^a	30.08	SICHUAN UNIVERSITY	1.000
7	UNIVERSITY OF READING	88	INSTITUTE OF AUTOMATION	23.24	IMPERIAL COLLEGE LONDON ^a	0.021
8	NEWCASTLE UNIVERSITY	86	UNIVERSITY OF CALIFORNIA ^a	22.59	STANFORD UNIVERSITY ^a	0.021
9	INDIAN INST TECHNOL DELHI	85	UNIVERSITY OF MARYLAND ^a	20.00	NATIONAL UNIVERSITY OF SINGAPORE ^a	0.020
10	FORE SCHOOL OF MANAGEMENT	85	MASSACHUSETTS GENERAL HOSPITAL	19.00	TSINGHUA UNIVERSITY ^a	0.019

Source: Authors Estimate.

Indicates that the institution is in the top 100 Universities Globally based on US News and World Report University Ranking.

Degree Centrality – We note that the University of Singapore tops the list based on degree measures highlighting the high number of collaborative linkages associated with this University. This is followed by Universities coming from the developed country contexts, including the University of Colorado, Central Florida (USA), and Qatar University (Qatar).

Betweenness Centrality – Interestingly, the University of Singapore (Singapore) also ranks within the top ten of betweenness network measures scores, highlighting their pivotal role as knowledge gatekeepers, but also knowledge brokers based on their ability to influence and direct information in this new emerging field. Other universities from key emerging economies such as Tsinghua University (China), fall in the top 10 institutions in terms of high betweenness measures. Other key institutions from developed countries include The Mayo Clinic – College of Medicine (USA), Standford University (USA), Imperial College of London (UK) and the University of California (USA). It is noteworthy to mention that many high-ranking universities at the global level are included in this list, suggesting that institutional prestige plays a significant role in shaping universities as key brokers of knowledge exchange.

Closeness Centrality – Institutions from developed emerging economies tend to dominate in terms of measures of closeness (Monash University (Australia), Vanderbilt University (USA) and Imperial College of London (United Kingdom)). However, there are several institutions based in emerging countries that also are included in the top 10 list Chandigarh University (India), Joseph Ayo Babalola University (Nigeria) and Sichuan University (China).

Interestingly, the University of Singapore (Singapore) stands out in terms of being listed in the top 10 of all three network measures, highlighting the relative importance of this institution, not only in terms of collaboration but also the dissemination and linkage with other universities globally also involved in researching topics related to ChatGPT. Similarly, Imperial College of London (UK), Standford University (USA), Chandigarh University (India) and Tsinghua University (China), fall into the top 10 of both betweenness and closeness measures, indicating the relative importance of these institutions in acquiring and dissemination of information.

Figure 3 illustrates the collaborative network, highlighting the extent of collaboration among universities and scientific institutions involved in ChatGPT research. With 1283 members and 7765 edges, the network demonstrates an average degree of 13.99 and a density index of 0.009, indicating substantial collaboration within the community. Each member is connected to approximately 14 others, reflecting a tightly woven network of interactions.

Figure 3.

Cooperation network of scientific institutions.

The use of colours in Figure 3 reflects the results of community detection algorithms, revealing clusters of institutions with shared research interests and methodologies. Based on this information, we can provide the following naming and analysis about each of the communities also outlined in Table 6:

The Global Consortium (Purple): This prominent community boasts the largest diverse membership (107 universities) with a remarkable density of connections (0.42). This suggests a highly collaborative environment. The international composition (University of Central Florida (USA) and Qatar University (Qatar)) indicates a focus on a broad range of global applications of ChatGPT

The Florida Focus Group (Blue): This geographically concentrated community (108 members) centres on the Florida State University System. With a lower connection density (0.07), collaboration might be less extensive, potentially suggesting a regional focus on ChatGPT applications or a preference for independent research.

The Medical Pioneering Group (Red): This distinguished community of 118 members includes prominent institutions such as Harvard University, Johns Hopkins Medicine and the National University of Singapore. Their involvement suggests a focus on healthcare applications of ChatGPT, potentially exploring areas like chatbot-based diagnostics, patient education and therapeutic interventions.

The US Powerhouse Group (Green): This group (118 members) comprises leading US universities like the University of California System and MIT. Their high connection density (0.13) signifies a strong collaborative network. This suggests a potential national focus on ChatGPT research within the United States, with this group acting as a driving force.

The International Collaborative Group (Orange): This community (78 members) fosters collaboration between universities from Europe (Utrecht University, King’s College London), Canada (Western University) and the United States (University of North Carolina Charlotte). Their moderate connection density (0.11) indicates a well-connected but potentially less centralised group. This international composition suggests a focus on fostering global partnerships and knowledge exchange in ChatGPT research.

Table 6.

Main communities in the university network.

Community colour	The main members	Number of members	Number of connections	Density	Community name
Purple	University Of North Carolina, University Of Colorado System, University System Of Ohio, University Of New South Wales Sydney, University Of Central Florida, Qatar University, Liverpool John Moores University, University Of Bristol, Indian Institute Of Technology System (IIT System), Aston University	107	2382	0.42	The Global Consortium
Blue	State University System Of Florida, Florida State University, Florida International University, University Of Hawaii System, Varna University Of Management, Yeshiva University, University Of Florida, Bogazici University, University Of Massachusetts System, University Of Nevada Las Vegas	108	383	0.07	The Florida Focus Group
Red	National University Of Singapore, Harvard University, University Of London, Post Graduate Institute Of Medical Education & Research (Pgimer), Chandigarh, Johns Hopkins Medicine, Cleveland Clinic Foundation, Singapore National Eye Centre, Nanyang Technological University, Nanyang Technological University & National Institute Of Education (Nie) Singapore, Erasmus Mc	118	615	0.09	The Medical Pioneering Group
Green	University Of California System, University Of Washington, University Of Washington Seattle, United States Department Of Energy (Doe), Massachusetts Institute Of Technology (MIT), Lawrence Berkeley National Laboratory, Pennsylvania Commonwealth System Of Higher Education (Pcshe), the University Of Alabama System, University Of California San Diego, University Of Colorado Anschutz Medical Campus	118	902	0.13	The US Powerhouse Group
Orange	Utrecht University, King’s College London, Monash University, Rutgers New Brunswick, Rutgers System, George Washington University, University Of North Carolina Charlotte, University Of Birmingham, Western University (University Of Western Ontario), Trinity College Dublin	78	321	0.11	The International Collaborative Group

Source: Authors Estimate.

4.4. Researcher network

4.4.1. Uncovering the most influential authors

Table 7 highlights the top performers across four key graph metrics – Degree Centrality, Closeness Centrality, Betweenness Centrality, and Eigen Centrality – within the global collaborative research networks on ChatGPT. These metrics help assess the significance, influence, and connectivity of researchers in the network. The scores indicate the relative importance of individuals in driving collaboration, spreading knowledge, and facilitating the adoption and diffusion of cutting-edge technologies, especially ChatGPT.

Degree Centrality: This metric quantifies the number of co-authors each individual collaborates with. High scores indicate prominent collaborators who frequently engage with diverse researchers within the network. Okumus F (University of Central Florida, USA), Chowdhury S (Fondation Toulouse TBS Business School, France), Papagiannidis S (Newcastle University, UK), Dwivedi Y (Swansea University, UK) and Pandey N (National Institute of Industrial Engineering (NITE), India), emerge as the most prolific collaborators in this group.

Closeness Centrality (Giant Component): This indicator reflects the average geodesic distance (shortest path) between an individual and all other members within the network’s largest connected component (giant component). Essentially, it identifies individuals strategically positioned to efficiently access and disseminate information. Liu J (Sichuan University, China), Wang C (Sichuan University, China), Ong J (University of Michigan, USA), Wang X (Chinese Academy of Sciences, China) and Wu Y (Tianjin University, China) exhibit the highest levels of centrality within the giant component.

Betweenness Centrality (Giant Component): This metric captures the frequency with which an individual acts as a bridge between different subgroups within the giant component. High scores signify individuals who play a crucial role in facilitating information flow and collaboration across diverse research communities. Wang C (Sichuan University, China), Ong J (University of Michigan, USA), Liu J (Sichuan University, China) and Sarker P (University of Nevada, USA) stand out as prominent bridge builders in the network.

Table 7.

Top scorers in key graph indicators.

Rank	Degree centrality	Score	Closeness centrality (GC)	Score	Betweenness centrality (GC)	Score	Eigen centrality	Score
1	Okumus F	91	Liu J	0.224	Wang C	0.021	Okumus F	1.000
2	Chowdhury S	84	Wang C	0.223	Ong J	0.020	Dwivedi Yk	0.994
3	Papagiannidis S	84	Ong J	0.216	Sarker P	0.018	Pandey N	0.994
4	Dwivedi Yk	81	Wang X	0.210	Liu J	0.016	Kar Ak	0.994
5	Pandey N	81	Wu Y	0.207	Sharma A	0.011	Raman R	0.994
6	Kar Ak	80	Sarker P	0.207	Sharma R	0.011	Rana Np	0.994
7	Raman R	80	Chen L	0.206	Wang X	0.010	Chowdhury S	0.988
8	Rana Np	80	Li H	0.204	Prasad R	0.009	Papagiannidis S	0.988
9	Kraus S	69	Liu S	0.202	Kim J	0.008	Kraus S	0.967
10	Kshetri N	67	Guo J	0.201	Li H	0.007	Kshetri N	0.966

Source: Authors Estimate.

Figure 4 illustrates a co-authorship network, visually showcasing the collaborative relationships among a group of authors. By examining the properties of this graph, we can derive valuable insights into the collaboration patterns and network structure within a particular research domain.

Network Size and Connectivity: The graph comprises 3899 nodes, representing individual authors, interconnected by 15,190 edges, signifying co-authorship relationships.

Network Sparsity: The graph density of 0.002 indicates a sparse network, where the number of edges is considerably lower compared with the maximum possible connections between nodes. This sparsity suggests that the majority of authors collaborate with a relatively small subset of other authors within the network.

Community Structure: The high modularity score of 0.948 implies a well-defined community structure within the network. This suggests that authors tend to cluster with others they collaborate with more frequently, potentially indicating the presence of distinct research communities or subfields within the broader domain.

Clustering and Collaboration Patterns: The average clustering coefficient of 0.96 signifies a high degree of local clustering within the network (Table 8). This implies that authors tend to collaborate with others who also share common co-authors, potentially reflecting the formation of close-knit collaboration groups around specific research topics. However, the presence of 658 connected components indicates that the network is not fully connected, suggesting the existence of isolated groups of authors with limited collaboration with the broader network.

Figure 4.

Co-authorship network.

Table 8.

Summary of network relationships.

Property	Value
Nodes	3899
Edges	15,190
Graph Density	0.002
Modularity	0.948
Connected Components	658
Avg. Clustering Coefficient	0.96

Source: Authors Estimate.

4.5. Keywords network

Analysing the keyword network offers valuable insights into the wide array of topics and categories related to ChatGPT applications. Figure 5 presents the keyword network generated from the articles in our sample.

Figure 5.

Keywords network.

The colours in Figure 5 represent the results obtained from the community detection, which provides a sort of categorisation of the studied domains. The main communities (comprising at least 5% of the total network members) are as follows.

4.5.1. Orange community: advanced GPT applications

This community encompasses advanced applications and research related to ChatGPT chat applications. It includes topics such as the utilisation of large language models in education, healthcare and technology. The keywords suggest a focus on training, decision-making and engineering prompts for ChatGPT models. In addition, it highlights the use of transformers and deep learning techniques in various domains, including healthcare (Ophthalmology) and digital health.

4.5.2. Red community: medical chatbot development

This community revolves around the development and utilisation of chatbots in the medical domain. It includes keywords related to medical education, practice and the use of language models to create conversational agents. The focus seems to be on integrating chatbots into medical curricula and practice, possibly for patient examination and interaction.

4.5.3. Blue community: generative AI and linguistic analysis

This community explores the realm of generative AI and linguistic analysis, focusing on topics such as bias, transparency and the behaviour of generative pre-trained transformers like ChatGPT. It delves into issues like the black-box nature of AI models, linguistic styles and the memorisation capabilities of AI systems. The keywords suggest a focus on understanding and improving the quality and reliability of generative AI outputs.

4.5.4. Purple community: intelligent systems and medicine

This community revolves around the intersection of AI, particularly ChatGPT models, and medicine. It discusses the application of intelligent systems in healthcare, including patient safety, performance comparison and the mitigation of biases such as gender bias. Reinforcement learning and autonomous vehicles are also mentioned, indicating a broader discussion on intelligent systems beyond healthcare.

4.5.5. Green community: ethics and higher education

This community focuses on ethical considerations and issues related to higher education, research and innovation. It addresses topics such as plagiarism, academic integrity and the ethical use of large language models in research. The keywords also touch upon digital literacy, productivity and the impact of technology on mental health, highlighting the broader societal implications of advanced AI applications.

4.5.6. Dark green community: assessment and communication

This community centres on communication, assessment and educational curriculum development. It addresses topics such as testing, writing and problem-solving in undergraduate education, with a focus on disciplines like urology and health. The keywords suggest an emphasis on effective communication strategies and the importance of addressing misconceptions in public understanding.

4.5.7. Pink community: data analysis and knowledge management

This community focuses on data analysis, knowledge management and the intersection of AI with various domains. It addresses topics such as sentiment analysis, data modelling and the management of multimodal data. The keywords suggest an interest in explainable AI, computational modelling and the application of AI techniques in fields like journalism and software engineering. In addition, it emphasises the importance of data visualisation and knowledge organisation in handling complex datasets.

5. Discussion and implications for policy

Based on our analysis, we can take the first steps to answer our key research questions. First, at the country level similar to past research on collaborative research networks on frontier technologies we confirm the dominance of key developed nations United States, United Kingdom, Germany, Australia and France as key countries where leading researchers, connectors and bridgers focusing on the theme ChatGPT are located. This is also reflected in the presence of an A.I. Powerhouse and Global Consortium community, made-up of researchers from top-developed countries. However, it is also important to mention the emergence of other network leaders and connectors from emerging economies, principally India, China, Qatar and the United Arab Emirates. These key nodes act as both connectors and bridgers to researchers in both developed and emerging countries. The presence of the Florida Focus Community (USA), Eastern European AI Cooperation Community (Eastern European), and the Middle Eastern, and Asian research communities also confirms the importance of geography and cultural proximity as key factors associated with research collaboration in this field [86 –88]. We also note that collaborators among researchers from key emerging economies such as India and Singapore have established research networks across other diverse and quite distinct emerging economies such as Norway, Nigeria, Russia and Thailand, which suggests other factors may be at play in terms of influencing collaboration patterns at a country level.

Our institutional-level analysis confirms the dominance of key prestigious institutions such as the National University of Singapore, Stanford University, Harvard University, Johns Hopkins University, Tsinghua University and Imperial College London as leaders and key connectors in the network, spanning both developed and emerging economies (Appendix 1 Table 9). This highlights the importance of institutional proximity or factors related to common institutional form, and the availability of key competencies, resources or endowments may be key contributing factors influencing collaborative relationships [88 –90]. Interestingly, The National University of Singapore is found in the top 10 of key network measures, which suggests the relative importance of this University, not only in generating research but in connecting and disseminating knowledge among other institutions globally. A closer examination of our community-level analysis suggests reasons for their dominance. The National University of Singapore is found to be one of the leaders of the Medical Pioneering Group communities, focusing on ChatGPT applications related to the medical field, and also one of the premier research institutions globally.

At the researcher level, we note the importance of key researchers such as Okumus F (University of Central Florida, USA) and Chowdhury S (Fondation Toulouse TBS Business School, France), as key focal nodes in terms of the number of connections, but significant differences exist among researchers, who are key disseminators and bridgers of networks, with significant overlaps among researchers from emerging countries such as Liu J (Sichuan University, China), Wang C (Sichuan University, China), Wang X (Chinese Academy of Sciences, China) and Wu Y (Tianjin University, China), all of whom ae based in China. We also note the relative concentration of researchers in two to three major ‘giant’ networks of collaborators networks, highlighting the intensity of collaboration among a large body of researchers.

Finally, we observe thematic specialisation among researchers. Key themes include advanced ChatGPT applications in sectors such as healthcare, education, communication, linguistics and data analytics. Notably, there is a strong focus on ethics and assessments in higher education, reflecting efforts by researchers to address the risks associated with ChatGPT’s use in educational contexts. This suggests that both global and local research collaborations may be formed around specialised research themes [91].

Overall, this research highlights the importance of establishing research links with key thought leaders both locally and internationally. What is now obvious is the importance of geographic and cultural proximity in supporting the growth of collaborative research networks, a finding supported by such researchers as de Dominicis et al. [92] and Santamaría et al. [93]. This can help inform outreach strategies for researchers, who might opt to join networks that are geographically and culturally closer to advance their research agendas more effectively. In addition, fostering a certain level of institutional specialisation is crucial. By building a critical mass of expertise, institutions can enhance their ability to share knowledge, develop new capabilities, and establish partnerships with other academic or industry entities. This specialisation can also improve access to grant funding and resources, positioning institutions as leaders in specific fields. Moreover, it supports researchers in joining global networks, especially in collaboration with prestigious institutions, opening up more opportunities for funding, cutting-edge technology, and engagement with thought leaders in their respective fields. Finally, being a globally recognised research institution is important, but being part of a collaborative network with other similar institutions that focus on common topics and issues can enhance the impact and distribution of research across these communities.

6. Conclusion

This article aims to enrich the literature on collaborative research networks, specifically focusing on those investigating ChatGPT, a cutting-edge frontier technology in AIGUC. Our goal is to identify key network leaders, disseminators, bridges and research communities at the country, institutional and researcher levels. By doing so, we explore how collaborative network relationships can facilitate both the adoption and dissemination of this emerging technology.

Collectively, our results highlight the growing importance of collaborative research networks in generating new innovative peer-reviewed research on an emerging AI-based frontier technology, ChatGPT. In keeping with past research and our core research questions we confirm the importance of researchers from key developed countries, specifically the United States, United Kingdom, Germany and France, or what can be considered Global AI powerhouses, as key collaborators, disseminators and bridgers of research and knowledge on ChatGPT. Interestingly, we also note the growing importance of researchers from other key emerging locations such as China, India, Singapore, and Qatar, which suggests that these locations are becoming centres of research in their own right, bridging and disseminating knowledge among researchers operating in other emerging economies. Just one key example of this is researchers from the University of Singapore which has been able to position this institution as a key node as a bridger and disseminator on research of medical applications of ChatGPT. Second, we confirm the importance, of geography and cultural proximity as important drivers for collaborative research patterns among researchers, specifically, among researchers operating in the United States, the Middle East and Eastern Europe. This is also confirmed by intense intra-country relationships, such as the US Powerhouse Group, and the Florida Focus Group. However, many far-reaching global and diverse research networks exist, such as the Global Consortium and the International Collaborative Group which suggests other factors matter. One explanation can be the network of researchers’ from key prestige made up of leading institutions from both developed and emerging economies, who become key disseminators of information, also acting as bridges across networks. This suggests institutional proximity, access to resources and specialised capabilities become important drivers in establishing research network relationships. We also note specialisation among researchers based on key themes, related to the application and impact of ChatGPT, in such fields as technology, health, education, communication and linguistics. This specialisation is also reflected institutionally, as in the case of the Medical Pioneering Group, made up of key health-based research institutions located throughout the globe.

One of the key limitations of the study remains the survey period, which extends up to 1 year after the launch of ChatGPT 3.5. This may be a relatively short period in terms of the time it takes for research networks to develop and be fully operationalised, which past research has suggested to be between 3 and 5 years, and up to 10 years [58,94]. We, therefore, believe our research to be indicative of what can be considered research networks among first-mover researchers on ChatGPT, and as such provide insight into which researchers and networks, may have the capability to move first on frontier technologies such as ChatGPT.

Second, our analysis and results are based on peer-reviewed articles and books (including book chapters) sourced exclusively from the WOS database, rather than other bibliometric databases like Scopus. We chose WOS as our preferred bibliographic database not only for its extensive and diverse range of journals but also for its advanced search features, which allowed us to tailor our search for specific types of research outputs and time periods in months. In addition, we excluded research from conference proceedings, which, while significant in fields like computer science, may not have undergone the same rigorous peer review process. This article aimed to focus on research outputs that have undergone peer review, with copyrights and intellectual property rights assigned, serving as our standardised measure of quality.

Third, this article focuses on collaborative research patterns on one of the latest frontier technologies ChatGPT. Recently, many other generative AI applications have emerged such as Google’s BARD (now Gemini), and Bing AI, with an increasing number of specialised AI and chatbox solutions being developed for specific fields and purposes [95]. In addition, numerous other forms of 4IR and Industry 5.0 frontier technologies have not yet been examined by researchers, such as 3D printing, IoTs and robotics which can also be opportunities for extending research in this domain.

Another opportunity for further research relates to uncovering detailed drivers of research collaboration patterns on ChatGPT. While not an explicit objective of this article, it would be good not only, to analyse possible research collaboration patterns of frontier technology, but also to understand possible determinants of the underlying research patterns. Indeed, we have attempted to uncover some elements of this, based on the information that is available from bibliometric and web-based sources. However, other researchers have been able to delve deeper, using micro-level data on researchers’ profiles and detailed information on the research projects completed (see Wang et al. [96], Jeong et al. [97], Chang [98] and Jeong et al. [99] as useful examples). Indeed, we believe that this can be an opportunity for further research, by focusing on uncovering key drivers for collaborative research on a larger number of 4IR and Industry 5.0 frontier technologies, and over a long period where network research collaborations will probably be more established.

Footnotes

Appendix 1. Key Universities and Global Rank

Table 9.

Selected universities ranked in the top 100 globally based on US news and world report rankings.

	Institution	Country	Global Rank
1	HARVARD UNIVERSITY	United States	1
2	STANFORD UNIVERSITY	United States	3
3	UNIVERSITY OF LONDON	United Kingdom	7
4	UNIVERSITY OF CALIFORNIA	United States	11
5	IMPERIAL COLLEGE LONDON	United Kingdom	12
6	JOHN HOPKINS UNIVERSITY	United States	13
7	TSINGHUA UNIVERSITY	CHINA	16
8	NATIONAL UNIVERSITY OF SINGAPORE	SINGAPORE	22
9	NANYANG TECHNOLOGICAL UNIVERSITY	SINGAPORE	27
10	MONASH UNIVERSITY	AUSTRALIA	35
11	UNIVERSITY OF NEW SOUTH WALES	AUSTRALIA	36
12	ICAHN SCHOOL OF MEDICINE AT MOUNT SINAI	United States	40
13	VANDERBILT UNIVERSITY MEDICAL CENTRE	United States	63
14	UNIVERSITY OF MARYLAND	United States	72
15	UNIVERSITY OF BRISTOL	United Kingdom	96

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship and/or publication of this article.

ORCID iD

Richard Ramsawak

References

OECD. OECD science, technology and industry scoreboard 2017. Paris: OECD, 2017.

ESCAP

Frontier technologies for sustainable development in Asia and the Pacific. 2018, https://repository.unescap.org/bitstream/handle/20.500.12870/295/ESCAP-2018-RP-Frontier-technologies-sustainable-development-Asia-Pacific.pdf?sequence=1&isAllowed=y

Zhenmin

Frontier technologies: a window of opportunity for leapfrogging! UN Chronicle

2019; 55(4): 39–43.

Ramalingam

Hernandez

Prieto Martín

, et al. Ten Frontier Technologies for international development. 2016, https://opendocs.ids.ac.uk/articles/online_resource/Ten_Frontier_Technologies_for_International_Development/26440834

Kharas

McArthur

Ohno

Breakthrough: the promise of Frontier technologies for sustainable development. Washington, DC: Brookings Institution Press, 2022.

Kohnová

Papula

Salajová

. Internal factors supporting business and technological transformation in the context of Industry 4.0. Bus: Theor Pract 2019; 20: 137–145.

Horváth

Szabó

RZ.

Driving forces and barriers of Industry 4.0: do multinational and small and medium-sized companies have equal opportunities?

Technol Forecast Soc Change 2019; 146: 119–132.

Liu

, et al. A brief overview of ChatGPT: the history, status quo and potential future development. IEEE CAA J Autom Sin 2023; 10(5): 1122–1136.

Dominguez Lacasa

Jindra

Radosevic

, et al. Paths of technology upgrading in the BRICS economies. Res Pol 2019; 48(1): 262–280.

10.

Lee

Bozeman

The impact of research collaboration on scientific productivity. Soc Stud Sci 2005; 35(5): 673–702.

11.

Adams

Collaborations: the rise of research networks. Nature 2012; 490(7420): 335–336.

12.

van Rijnsoever

Hessels

. Factors associated with disciplinary and interdisciplinary research collaboration. Res Pol 2011; 40(3): 463–472.

13.

Giuliani

Bell

The micro-determinants of meso-level learning and innovation: evidence from a Chilean wine cluster. Res Pol 2005; 34(1): 47–68.

14.

Messeni Petruzzelli

Albino

Carbonara

, et al. Leveraging learning behavior and network structure to improve knowledge gatekeepers’ performance. J Knowl Manag 2010; 14(5): 635–658.

15.

Persson

Are highly cited papers more international?

Scientometrics 2010; 83(2): 397–401.

16.

Wagner

Brahmakulam

Jackson

, et al. Science and technology collaboration: building capability in developing countries. No RAND/MR-13570-WB, Report. Santa Monica, CA: Rand Corporation, 2001.

17.

Tatsuro

Caroline

Anny

, et al. Science & technology collaboration: building capacity in developing countries?Santa Monica, CA: RAND Corporation. 2001.

18.

Fleming

Mingo

Chen

Collaborative brokerage, generative creativity, and creative success. Admin Sci Quart 2007; 52(3): 443–475.

19.

Banerjee

Chandrasekhar

Duflo

, et al. Gossip: identifying central individuals in a social network. National Bureau of Economic Research, 2014, https://www.nber.org/system/files/working_papers/w20422/w20422.pdf

20.

UNIDO. Industrial development report 2020: industrializing in the digital age. New York: UN, 2019.

21.

Sanders

Falcão

Haider

, et al. World economic and social survey 2018: Frontier technologies for sustainable development, 2018, https://www.un.org/development/desa/dpad/publication/world-economic-and-social-survey-2018-frontier-technologies-for-sustainable-development/

22.

Schwab

The fourth industrial revolution. Danvers, MA: Crown Currency, 2017.

23.

Molontay

Nagy

Twenty years of network science: a bibliographic and co-authorship network analysis. In: Çakırtaş

Ozdemir

(eds) Big data and social media analytics: trending applications. Cham: Springer, 2021, pp. 1–24.

24.

Maddikunta

PKR

Pham

Q-VBP

Deepa

, et al. Industry 5.0: a survey on enabling technologies and potential applications. J Ind Inform Integ 2022; 26: 100257.

25.

Demir

Döven

Sezen

Industry 5.0 and human-robot co-working. Proc Comput Sci 2019; 158: 688–695.

26.

Kolluru

Gupta

The prospects and risks of Industry 4.0: issues and implications. In: Al Mawali

Al Lawati

Ananda

(eds) Fourth industrial revolution and business dynamics: issues and implications. Singapore: Springer, 2021, pp. 223–239.

27.

Lorenz

Kraemer-Mbula

Chapter 14: Measuring frontier technology adoption in developing countries. In: Gault

Arundel

Kraemer-Mbula

(eds) Handbook of innovation indicators and measurement. Cheltenham: Edward Elgar Publishing, 2023, pp. 260–277.

28.

Zervoudi

EK.

Fourth industrial revolution: opportunities, challenges, and proposed policies. In: Grau

Wang

(eds) Industrial robotics: new paradigms. London: Intech Open, 2020, pp. 1–25.

29.

Mellacher

Scheuer

Wage inequality, labor market polarization and skill-biased technological change: an evolutionary (agent-based) approach. Comput Econ 2021; 58(2): 233–278.

30.

Connor

Kemeny

Storper

Frontier workers and the seedbeds of inequality and prosperity. J Econ Geogr 2023; 24: lbad018.

31.

Capello

Lenzi

The regional economics of technological transformations: Industry 4.0 and servitisation in European regions. London: Routledge, 2021.

32.

Acemoglu

Restrepo

The wrong kind of AI? Artificial intelligence and the future of labour demand. Cambr J Reg Econ Soc 2019; 13(1): 25–35.

33.

Acemoglu

Restrepo

Robots and jobs: evidence from US labor markets. J Pol Econ 2019; 128(6): 2188–2244.

34.

Archibugi

Pietrobelli

The globalisation of technology and its implications for developing countries: windows of opportunity or further burden?

Technol Forecast Soc Change 2003; 70(9): 861–883.

35.

Pietrobelli

Soete

The role of foreign technology and indigenous innovation in the emerging economies: technological change and catching-up. World Dev 2011; 39(7): 1204–1212.

36.

Dharani

Jyostna

JVSL

Sucharitha

Likitha

, et al. Interactive transport enquiry with AI chatbot. In: Proceedings of the 2020 4th international conference on intelligent computing and control systems (ICICCS), Madurai, India, 13–15 May 2020. New York: IEEE.

37.

Skrebeca

Kalniete

Goldbergs

, et al. Modern development trends of chatbots using artificial intelligence (AI). In: Proceedings of the 2021 62nd international scientific conference on information technology and management science of Riga Technical University (ITMS), Riga, 14–15 October 2021. New York: IEEE.

38.

Taecharungroj

‘What can ChatGPT do?’ Analyzing early reactions to the innovative AI chatbot on Twitter. Big Data Cogn Comput 2023; 7(1): 35.

39.

Guo

. Harnessing the power of ChatGPT in medical education. Med Teach 2023; 45: 1063.

40.

Achiam

Adler

Agarwal

, et al. (eds). GPT-4 technical report. 2023, https://cdn.openai.com/papers/gpt-4.pdf

41.

Mubin

Alnajjar

Trabelsi

, et al. Tracking ChatGPT research: insights from the literature and the Web. IEEE Access 2024; 12: 30518–30532.

42.

Hassani

Silva

ES.

The role of ChatGPT in data science: how AI-assisted conversational interfaces are revolutionizing the field. Big Data Cogn Comput 2023; 7(2): 62.

43.

Farhat

Silva

Hassani

, et al. The scholarly footprint of ChatGPT: a bibliometric analysis of the early outbreak phase. Front Artif Intel 2024; 6: 1270749.

44.

Farhat

Sohail

Madsen

DO.

How trustworthy is ChatGPT? The case of bibliometric analyses. Cogent Eng 2023; 10(1): 2222988.

45.

Oliński

Krukowski

Sieciński

Bibliometric overview of ChatGPT: new perspectives in social sciences. Publications 2024; 12(1): 9.

46.

Baig

Yadegaridehkordi

ChatGPT in the higher education: a systematic literature review and research challenges. Int J Educ Res 2024; 127: 102411.

47.

Guenier

AW.

ChatGPT and health communication: a systematic literature review. Int J e-Health Med Commun 2024; 15(1): 1–26.

48.

Tian

Wang

, et al. Bibliometric analysis of research on ChatGPT: using VOSViewer. In: Proceedings of 2023 6th international conference on educational technology management (ICETM’2023), Guangzhou, China, 2–5 November 2023, pp. 235–243. New York: ACM.

49.

Beaver

DD.

Reflections on scientific collaboration (and its study): past, present, and future. Scientometrics 2001; 52(3): 365–377.

50.

Luis Toral

Bessis

del Rocío Martínez Torres

-M

. External collaboration patterns of research institutions using shared publications in the Web of Science. Program 2013; 47(2): 170–187.

51.

Sharma

Thomas

VJ.

Inter-country R&D efficiency analysis: an application of data envelopment analysis. Scientometrics 2008; 76(3): 483–501.

52.

Guerrero Bote

Olmeda-Gómez

de Moya-Anegón

Quantifying the benefits of international scientific collaboration. J Am Soc Inform Sci Technol 2013; 64(2): 392–404.

53.

Collaboration with foreign universities for innovation: evidence from Chinese manufacturing firms. Int J Technol Manag 2016; 70(2–3): 193–217.

54.

Liu

Bollen

Nelson

, et al. Co-authorship networks in the digital library research community. Inform Process Manag 2005; 41(6): 1462–1480.

55.

Acedo

Barroso

Casanueva

, et al. Co-authorship in management and organizational studies: an empirical and network analysis*. J Manag Stud 2006; 43(5): 957–983.

56.

Katz

JS.

Bibliometric assessment of intranational university-university collaboration. Brighton: University of Sussex, 1992.

57.

Gazni

Sugimoto

Didegah

Mapping world scientific collaboration: authors, institutions, and countries. J Am Soc Inform Sci Technol 2012; 63(2): 323–335.

58.

Wagner

Leydesdorff

Network structure, self-organization, and the growth of international collaboration in science. Res Pol 2005; 34(10): 1608–1618.

59.

Wagner

CS.

Six case studies of international collaboration in science. Scientometrics 2005; 62(1): 3–26.

60.

Mizukami

Nakano

International comparison of cross-disciplinary integration in industry 4.0: a co-authorship analysis using academic literature databases. PLOS ONE 2022; 17(10): e0275306.

61.

Bindu

Sankar

Kumar

KS.

Research collaboration and knowledge sharing in e-governance. Transform Govern: People Proces Pol 2019; 13(1): 2–33.

62.

Moosavi

Naeni

Fathollahi-Fard

, et al. Blockchain in supply chain management: a review, bibliometric, and network analysis. Env Sci Poll Res. Epub ahead of print 27 February 2021. DOI: 10.1007/s11356-021-13094-3.

63.

Maghsoudi

Shokouhyar

Ataei

, et al. Co-authorship network analysis of AI applications in sustainable supply chains: key players and themes. J Clean Prod 2023; 422: 138472.

64.

Munoz

Queupil

Fraser

Assessing collaboration networks in educational research. Int J Educ Manag 2016; 30(3): 416–436.

65.

Dahlander

Gann

DM.

How open is innovation?

Res Pol 2010; 39(6): 699–709.

66.

Birkle

Pendlebury

Schnell

, et al. Web of Science as a data source for research on scientific and scholarly activity. Quan Sci Stud 2020; 1(1): 363–376.

67.

Teubner

Flath

Weinhardt

, et al. Welcome to the era of ChatGPT et al. The prospects of large language models. Bus Inform Syst Eng 2023; 65: 95–101.

68.

Kalyan

KS.

A survey of GPT-3 family large language models including ChatGPT and GPT-4. Nat Lang Process J 2024; 6: 100048.

69.

Moher

Liberati

Tetzlaff

, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Int J Surg 2010; 8(5): 336–341.

70.

Ramsawak

Buertey

Maheshwari

, et al. Interlocking boards and firm outcomes: a review. Manag Dec 2023; 62: 1291–1322.

71.

Müller

M-C

Reitz

Roy

Data sets for author name disambiguation: an empirical analysis and a new resource. Scientometrics 2017; 111(3): 1467–1500.

72.

Kim

Diesner

Distortive effects of initial-based name disambiguation on measurements of large-scale coauthorship networks. J Assoc Inform Sci Technol 2016; 67(6): 1446–1461.

73.

Kim

Diesner

The effect of data pre-processing on understanding the evolution of collaboration networks. J Informe 2015; 9(1): 226–236.

74.

Aria

Cuccurullo

Bibliometrix: an R-tool for comprehensive science mapping analysis. J Inform 2017; 11(4): 959–975.

75.

Wagner

CS.

International collaboration in science and technology: promises and pitfalls. In: Box

Engelhard

(eds) Science and technology policy for development, dialogues at the interface. London: Anthem Press, 2006, pp. 165–176.

76.

Wasserman

Social network analysis: methods and applications, vol. 2. Cambridge: Cambridge University Press, 1994.

77.

Stanley

Katherine

Social network analysis: methods and applications. New York: Cambridge University Press, 2009.

78.

Petrescu-Prahova

Belza

Leith

, et al. Peer reviewed: using social network analysis to assess mentorship and collaboration in a public health network. Prevent Chron Dis 2015; 12: E130.

79.

Xie

Liu

, et al. Analysis of scientific collaboration networks among authors, institutions, and countries studying adolescent myopia prevention and control: a review article. Iran J Public Health 2019; 48(4): 621–631.

80.

Freeman

The development of social network analysis: a study in the sociology of science, vol. 1. Laguna Beach, CA: ΣP Empirical Press, 2004.

81.

Darvish

Tonta

(eds). The network structure of nanotechnology research output of Turkey: a co-authorship and co-word analysis study, 2015, https://www.issi-society.org/proceedings/issi_2015/0732.pdf

82.

Salamati

Soheili

Social network analysis of Iranian researchers in the field of violence. Chinese J Traumatol 2016; 19(5): 264–270.

83.

Bedi

Sharma

Community detection in social networks. Wires Data Min Knowl Disc 2016; 6(3): 115–135.

84.

Chhonker

Verma

Kar

, et al. M-commerce technology adoption. Bottom Line 2019; 31(3/4): 208–233.

85.

Newman

MEJ

. Communities, modules and large-scale structure in networks. Nat Phys 2012; 8(1): 25–31.

86.

Katz

JS.

Geographical proximity and scientific collaboration. Scientometrics 1994; 31(1): 31–43.

87.

Hoekman

Frenken

van Oort

The geography of collaborative knowledge production in Europe. Annal Reg Sci 2009; 43(3): 721–738.

88.

Hoekman

Frenken

Tijssen

RJW

. Research collaboration at a distance: changing spatial patterns of scientific collaboration within Europe. Res Pol 2010; 39(5): 662–673.

89.

Hoekman

Frenken

Oort

, et al. Collaboration networks as carriers of knowledge spillovers: evidence from EU 27 regions. 2008, https://www.researchgate.net/publication/23524538_Collaboration_networks_as_carriers_of_knowledge_spillovers_Evidence_from_EU27_regions

90.

Ponds

Van Oort

Frenken

The geographical and institutional proximity of research collaboration. Paper Reg Sci 2007; 86(3): 423–444.

91.

López

Arza

The role of thematic specialization in international scientific collaboration: the case of Chagas research. Sci Publ Pol 2024; 51: scae017.

92.

de Dominicis

Florax

RJGM

de Groot

HLF

. Regional clusters of innovative activity in Europe: are social capital and geographical proximity key determinants? Appl Econ 2013; 45(17): 2325–2335.

93.

Santamaría

Nieto

Rodríguez

Failed and successful innovations: the role of geographic proximity and international diversity of partners in technological collaboration. Technol Forecast Soc Change 2021; 166: 120575.

94.

Luukkonen

Persson

Sivertsen

Understanding patterns of international scientific collaboration. Sci Technol Hum Val 1992; 17(1): 101–126.

95.

Gozalo-Brizuela

Garrido-Merchán

EC.

A survey of generative AI applications (Arxiv preprint Arxiv: 2306; 027812023), https://arxiv.org/abs/2306.02781

96.

Wang

Rodan

Fruin

, et al. Knowledge networks, collaboration networks, and exploratory innovation. Acad Manag J 2013; 57(2): 484–514.

97.

Jeong

Choi

Kim

The determinants of research collaboration modes: exploring the effects of research and researcher characteristics on co-authorship. Scientometrics 2011; 89(3): 967–983.

98.

Chang

S-H.

The technology networks and development trends of university-industry collaborative patents. Technol Forecast Soc Change 2017; 118: 107–113.

99.

Jeong

Choi

Kim

J-Y.

On the drivers of international collaboration: the impact of informal communication, motivation, and research resources. Sci Publ Pol 2013; 41(4): 520–531.

Adoption and diffusion of Frontier technologies: Tracing global collaborative research networks on ChatGPT

Abstract

Keywords

1. Introduction

2. Review of the literature

2.1. The rise of new Frontier technologies

2.2. The rise of ChatGPT

2.2. Technology adoption through collaborative research networks

2.3. Research on collaborative research networks on 4IR frontier technologies

3. Data and methodology

3.1. Utilising SNA measures

3.1.1. Density

3.1.2. Centrality

4. Results

4.1. Descriptive statistics

4.2. Country-level analysis

4.2.1. Community 1: ‘global AI powerhouses’ (purple)

4.2.2. Community 2: ‘diverse global contributors’ (blue)

4.2.3. Community 3: ‘Middle East & Asia collaborators’ (green)

4.2.4. Community 4: ‘emerging AI hubs’ (orange)

4.2.5. Community 5: ‘Eastern European AI cooperation’ (dark green)

4.3. Network of scientific institutions

4.4. Researcher network

4.4.1. Uncovering the most influential authors

4.5. Keywords network

4.5.1. Orange community: advanced GPT applications

4.5.2. Red community: medical chatbot development

4.5.3. Blue community: generative AI and linguistic analysis

4.5.4. Purple community: intelligent systems and medicine

4.5.5. Green community: ethics and higher education

4.5.6. Dark green community: assessment and communication

4.5.7. Pink community: data analysis and knowledge management

5. Discussion and implications for policy

6. Conclusion

Footnotes

Appendix 1. Key Universities and Global Rank

Declaration of conflicting interests

Funding

ORCID iD

References