Positioning Artificial Intelligence Research in East Asia and Latin America: A Comparative Bibliometric Analysis

Abstract

This study aims to provide a comprehensive cross-regional bibliometric analysis of artificial intelligence (AI) research in East Asia and Latin America from 2020 to 2025. By quantifying publication trends, authorship, institutional productivity, collaboration networks, and citation impact, the research seeks to identify regional leaders, thematic clusters, and disparities in visibility and impact between these two regions. Design/methodology/approach; Scopus-indexed publications containing the phrases “artificial intelligence research” or “artificial intelligence innovation” in their title, abstract, or keywords were retrieved for the period 2020–2025. Inclusion criteria required at least one author’s affiliation in any of the fourteen specified countries across East Asia or Latin America. All document types (articles, reviews, conference papers, book chapters) were considered. Metadata were manually extracted from Scopus database ranking to identify the top-cited papers, most prolific authors, leading institutions, thematic and subject-area concentrations, and crossnational collaboration patterns. Findings; this bibliometric review clarifies the dynamic trajectory of AI research in East Asia and Latin America, revealing significant disparities in productivity, visibility, and thematic focus. The findings underscore the need for targeted investments in research capacity building, strategic international partnerships, and thematic realignment particularly for Latin America to enhance global visibility and align with emerging AI trends. Originality; by contrasting two understudied regions (East Asia vs. Latin America), we capture shifts in the AI landscape—specifically, the generative AI boom across subfields and regions that no single region or pre 2022 study can. By highlighting structural disparities in productivity, citation impact, and institutional support, it offers policymakers, funding agencies, and academic leaders novel insights.

1. Introduction

Artificial Intelligence (AI) has rapidly evolved from a specialized field of computer science into a broad sociotechnical phenomenon reshaping economies, institutions, and public policy worldwide. Advances in algorithmic design, scalable computation, and data-intensive methods have enabled novel applications that touch healthcare, agriculture, manufacturing, finance, and public administration. Yet the diffusion of these capabilities is uneven: the trajectory of AI development and deploment is mediated by differences in infrastructure, human capital, institutional arrangements, and national strategic choices [1].

As a consequence, the global rise of AI has coincided with the amplification of regional asymmetries—a reality that requires systematic comparative inquiry if policy interventions are to be effective and equitable [2]. Dual-region comparison is under-studied in the literature, and it offers novel in-sights by contrasting a high-capacity AI region with a developing one, thereby identifying unique challenges and opportunities for Latin America.

Comparative bibliometric analysis can serve as a powerful diagnostic tool in revealing the structural contours of such asymmetries. Bibliometrics does not merely quantify publications and citations; when combined with network analysis and thematic mapping, it exposes the architectures of knowledge production and the channels through which expertise and resources circulate across institutional and national boundaries. In particular, comparing East Asia—where state-led strategies and concentrated investments have produced high volumes of frontier methodological work, with Latin America—where research is frequently more fragmented and application oriented, allows an assessment of how divergent innovation models translate into different scientific footprints and global influence [2,3].

Comparing these particular regions allows us to observe how a highly industrialized, investment rich context (East Asia) differs from a developing, fragmented research context (Latin America) in AI, a contrast that can yield lessons for other regions and for global AI policy. This study adopts a reproducible bibliographic approach to compare the production and structure of AI research in East Asia and Latin America. The comparative lens is motivated by three complementary considerations. First, the political economy of innovation differs substantially across regions: East Asia displays coordinated public private investment and institutional consolidation that foster cumulative advantage; Latin America exhibits more heterogeneous funding landscapes and weaker concentration of high-capacity centers, which affect both scale and thematic orientation of research. Second, thematic emphases—foundational versus applied research—carry implications for long-term agency in standard setting, technology governance, and the translation of research into locally relevant solutions. Third, collaboration topologies and publication practices mediate visibility and impact: dense, hub-based networks promote rapid knowledge diffusion and methodological leadership, while modular, dispersed networks may excel at contextual application but be less visible in global citation economies [1,2].

Beyond documenting differences, our inquiry is oriented toward empirical diagnosis that can inform targeted policy responses. The unequal distribution of AI capabilities intersects with broader social and developmental goals; persistent disparities in research capacity risk reproducing technological dependency and obstructing progress toward several Sustainable Development Goals (SDGs) if left unaddressed [4]. Moreover, addressing these asymmetries is not only a matter of equity but of global technological governance: a more plural and geographically distributed base for AI research can contribute to more diverse value judgments embedded in systems, more contextually appropriate solutions, and stronger regional resilience.

Bibliometric analysis has become an essential methodology for studying scholarly production across disciplines. Öztürk et al. [5] highlight its growing relevance, noting that it has evolved into a rigorous tool for evaluating bodies of literature, particularly in fields such as business and management, but increasingly across multiple areas of inquiry. As Donthu et al. [6] explain, it involves the systematic application of quantitative methods to bibliometric datasets, which allows scholars to assess productivity, thematic development, and research impact. Similarly, Passas [7] frames bibliometric studies as structured investigations designed to identify patterns, trends, and influence within a field, underscoring their usefulness for handling large volumes of scientific information. Several factors have contributed to the wider adoption of bibliometric approaches. Among them are the accessibility of software packages such as R and VOSviewer, along with the growth of major scholarly databases such as Google Scholar, Scopus, and Web of Science [7]. In addition, their interdisciplinary appeal—spanning domains from data science to operations research has further expanded their application [7]. Zupic and Čater [8] provide a methodological framework that details techniques such as citation analysis, co-citation analysis, bibliographic coupling, co-authorship analysis, and co-word analysis. They also describe science mapping as a means of visualizing how disciplines, fields, subfields, and individual studies interconnect, likening it to a geographic map of knowledge structures. Within this framework, bibliometric studies support researchers in detecting trends related to journal and article influence, collaboration dynamics, and the intellectual foundations of a research area [7]. Despite its methodological strengths, the approach is not without challenges. Öztürk et al. [5] caution that much of the current literature still lacks the ability to generate practical insights for improving research design and execution. Mukherjee et al. [9] reinforce this view, noting that while bibliometric studies have proven highly effective for mapping the evolution of fields and identifying emerging themes, they are sometimes criticized for insufficiently linking their analytical and visual outputs to theoretical development or practical applications.

It is pertinent to note that the bibliometric analysis was grounded in the principles formulated by Price [10], who established the foundations of the quantitative study of science by conceptualizing scientific literature as a cumulative and structured system, amenable to analysis through patterns of growth, productivity, and citation. In addition, the use of impact indicators and citation analysis was based on the seminal contribu-tions of Garfield [11], as the study sought to demonstrate that citation networks make it possible to identify intellectual cores, dominant streams, and cognitive hierarchies within a scientific field. These principles underpin the comparative analysis of visibility and influence applied in this study. Finally, the analytical framework was validated through the contributions of Leydesdorff [12], consolidating the integration of bibliometrics, network analysis, and systems theory, and emphasizing that science should be examined as a complex system of interdependent communications.

In recent years, artificial intelligence has rapidly expanded, reshaping diverse fields and consolidating its role as a disruptive technological driver. Dwivedi et al. [13] describe AI as a fast evolving suite of technologies that is redefining processes across business, society, and the environment. The parallel growth of big data and advanced computing has amplified its transformative potential, opening opportunities for both organizations and broader society. Rodriguez et al. [14] emphasize AI’s profound effect on scientific inquiry, noting that its data-processing capacity has transformed both natural and social sciences by enabling the detection of patterns, the formulation of predictions, and the development of new methodological and theoretical models. This multidisciplinary character is further reinforced by Dwivedi et al. [13], who note the increasing diversity and volume of studies addressing AI’s implications for organizational decision-making.

The development of AI research spans several decades, beginning in the mid-twentieth century. According to Dwivedi et al. [13], sustained academic engagement with the subject has fostered a wide array of theories and approaches for addressing societal and business challenges through intelligent systems. Dwivedi et al. [13] Also, identify a set of core topics that have gained momentum, including healthcare applications, sustainability, supply chain management, consumer adoption, and decision-support systems. Their findings show that AI is exerting a strong influence across numerous disciplines, including management, the social sciences, engineering, computer science, and mathematics. Recent bibliometric studies add further nuance to these observations. Valencia-Arias et al. [15] report that AI research output surged significantly in 2022 and 2023, with China, the United States, and India leading global production. Their study highlights an increasing orientation toward Sustainable Development Goals and neural network approaches, with major thematic clusters forming around sustainability and higher education. They also note that AI applications in sustainable practices constitute a rapidly emerging research frontier, leveraging AI to address environmental, social, and economic challenges across sectors.

Methodologically, the study compiles and analyzes a corpus of AI publications affiliated with institutions in East Asia and Latin America, applying bibliometric indicators of productivity (e.g., documents per institution, documents per researcher), impact (citations per document), thematic structure (topic clustering), and collaboration topology (co-authorship and institutional networks). The temporal scope is chosen to capture the recent acceleration of AI research and its emergent subfields; where appropriate, we highlight temporal dynamics reflecting the rapid adoption of methods such as deep learning and large language models.

To help contextualize the study, the introduction now concludes with a brief overview of the article’s structure. Section 2 details the research design and bibliometric methods used to collect, process, and analyze the dataset. Section 3 presents the results on publication productivity, citation impact, researchers’ h-index, universities’ ranking, thematic clusters, and collaboration patterns across both regions. Section 4 discusses these findings from a comparative perspective, emphasizing the regional disparities and their broader implications. Section 5 concludes the article by summarizing the main contributions, noting the study’s limitations, and outlining directions for future research.

2. Materials and Methods

2.1. Data Collection

Bibliometric analysis is considered a quantitative method because it relies on the systematic collection and analysis of data (such as citations and publication counts). Bibliometrics is an established method that uses data from scientific publications to map the structure and dynamics of specific scientific fields [8].

Identifying national research output using bibliometric databases is not methodologically neutral and may introduce systematic bias if not carefully executed. Huang et al. [16] demonstrate that common retrieval strategies in Scopus and the Web of Science Core Collection—such as searching for country names directly or applying country filters can yield inconsistent results due to variations in country name conventions and indexing practices. These discrepancies may affect the reliability and reproducibility of spatial and regional bibliometric analyses, underscoring the need for transparent reporting of data retrieval strategies and cautious interpretation of national level results. We queried the Scopus database for publications with either “artificial intelligence research” OR “artificial intelligence innovation” in the Title, Abstract, or Keywords fields. All document types indexed by Scopus were considered (i.e., articles, reviews, books, chapters, conference papers, etc.).

This search was limited to the publication years 2020–2025 and applied a filter requiring at least one author affiliated with one of our target East Asian or Latin American countries. The Scopus advanced query was: TITLE-ABS-KEY (“artificial intelligence research” OR “artificial intelligence innovation”) AND AFFILCOUNTRY (China OR South Korea OR Japan OR Taiwan OR Hong Kong OR Singapore OR Macao OR Brazil OR Mexico OR Colombia OR Peru OR Chile OR Ecuador OR Argentina) AND PUBYEAR > 2019 AND PUBYEAR < 2026. The year 2020 marks the onset of the modern acceleration of AI research, characterized by the emergence of foundational models and the precursors of today’s large language models.

The years 2022–2025 correspond to the global boom in generative AI, which fundamentally reshaped research priorities, publication volume, and thematic trends. In addition, most existing bibliometric reviews of AI cover periods ending before 2022, so they do not capture the transformative impact of generative AI. By focusing on 2020–2025, the primary motive was to capture the latest surge in AI research (notably the generative AI boom in 2022–2023) and provide up-to-date insights, even if it means some citation counts are low. In line with bibliometric best practices [8], we cross-checked institutional affiliations of each author at the time of publication with Google Scholar and author identifiers available in Scopus (such as the Scopus Author ID) to ensure they correspond to East Asian and Latin American institutions and to distinguish authors with similar names. For our analysis of prolific authors and collaboration networks, we cross-verified author identities by checking profiles (including affiliations and publication records) to reduce the risk of conflating different individuals. The result is a top list of higher-cited works and contributors (the query was executed in August 2025) that collectively define East Asia and Latin America’s footprint in artificial intelligence research.

Table 1. Inclusion and exclusion criteria for the Scopus-based bibliometric dataset (2020–2025).

Criteria	Description
Inclusion criteria	Publication years: 2020–2025. Subject categories: All. Document types: Articles (reviews, applied, surveys), review, conference paper, book chapter. Keywords: “artificial intelligence research” or “artificial intelligence innovation” in the title/abstract/keywords (topically relevant to the AI focus). Affiliation: At least one author affiliated with institutions from the 14 target countries: China, South Korea, Taiwan, Japan, Brazil, Hong Kong, Singapore, Mexico, Colombia, Peru, Ecuador, Chile, Macao, Argentina (East Asia or Latin America as defined in this study).
Exclusion criteria	Documents were excluded if they: (a) were outside the 2020–2025 publication year range, (b) lacked the specified AI-related keywords in the title or abstract, (c) were topically unrelated upon screening, or (d) did not include at least one author affiliated with an institution from the target East Asian or Latin American countries.
Documents	A total of 35,890 publications met these criteria, reflecting the inclusion of multiple document types and the rapid growth of AI-related publications in the past five years, especially in China.

synthesizes the inclusion and exclusion criteria applied to construct the Scopus-based bibliometric dataset on artificial intelligence publications for the 2020–2025 period, considering multiple document types and institutional affiliations across 14 East Asian and Latin American countries. This table provides the methodological framework that delimits the universe of analysis and ensures the thematic coherence of the selected documents.

2.2. Data Analysis

From this dataset, we aggregated metrics by author, paper, journal, and university affiliation. Authors were ranked by total citations of their artificial intelligence-related works; papers were ranked by citations and h-index; journals were ranked by citations of the included papers; and, on institutional productivity, universities were ranked by the number of times they appeared as an affiliation on any author. The list of the top 27 most-cited AI papers in the two regions is an illustrative output of the citation analysis, not a final inclusion set, and was derived from the Scopus full dataset ranking specifically from the “Analyze Results” function. The Reference list is an index of the analyzed articles and was also selected to provide context, for example, methodological guides for bibliometrics.

3. Results

This section presents the main findings from a bibliometric analysis of the scientific output of East Asia and Latin America in Artificial Intelligence research and innovation during the 2020–2025 period. The results are organized around the study’s four specific objectives: most-cited authors, most-cited articles, most-cited journals, and universities with the highest number of affiliations.

3.1. Taxonomy of AI Research Papers

To contextualize the thematic structure of the literature,

Table 2. Summary of Taxonomy.

Category	Subcategories	Representative Works
Methodological Foundations	Deep Learning and Neural Architectures Explainable AI and Interpretability Security, Privacy, and Federated Learning Foundations of Anomaly Detection and Swarm Intelligence	“A Survey of Transformers” “Explainable AI: Toward Medical XAI”
Application Domains	Healthcare and Bioinformatics Sustainable Development and Environmental AI Education and Human-Centered AI Metaverse, IoT, and Industry 4.0	“Large Language Models in Medicine” “AI in Metaverse: Security & Privacy”
Interdisciplinary Topics	AI in Business, Management, and Marketing AI and Social Sciences/Ethics AI and Sustainability	“AI in Marketing” “Ethical AI and Governance” “Challenges and Opportunities of Generative AI for Higher Education as Explained by ChatGPT”

Own Source.

summarizes the taxonomy of artificial intelligence research papers identified in the dataset, distinguishing methodological foundations, application domains, and interdisciplinary topics.

Citation analysis by category highlights that “Methodological Overviews” continue to attract substantial citations, but “Application Domains”, generative-AI, and medical AI are leading, reflecting the 2022–2023 surge. In “Interdisciplinary Topics”, marketing and service-oriented AI remain highly cited, suggesting that business-oriented AI has a strong academic impact. The coding scheme was obtained by assigning each document to its primary category based on title/abstract/keywords; multi-category papers are counted under “Interdisciplinary”.

3.2. Most Cited Papers

Table 3. Top-cited AI-related publications (Part I).

No.	Title	Authors	Journal	Year	Citations
[1]	“So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy	Dwivedi, Y.K.; Kshetri, N.; Hughes, L.; Wirtz, J.; Wright, R.	International Journal of Information Management	2023	2067
[2]	A review of uncertainty quantification in deep learning: Techniques, applications and challenges	Abdar, M.; Pourpanah, F.; Hussain, S.; Makarenkov, V.; Nahavandi, S.	Information Fusion	2021	1624
[3]	A Review of Yolo Algorithm Developments	Jiang, P.; Ergu, D.; Liu, F.; Cai, Y.; Ma, B.	Procedia Computer Science	2021	1568
[4]	Large language models in medicine	Thirunavukarasu, A.J.; Ting, D.S.J.; Elangovan, K.; Tan, T.F.; Ting, D.S.W.	Nature Medicine	2023	1318
[5]	Are Transformers Effective for Time Series Forecasting?	Zeng, A.; Chen, M.; Zhang, L.; Xu, Q.	Proceedings of the AAAI Conference on Artificial Intelligence	2023	1252
[6]	A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI	Tjoa, E.; Guan, C.	IEEE Transactions on Neural Networks and Learning Systems	2021	1161
[7]	A Metaverse: Taxonomy, Components, Applications, and Open Challenges	Park, S.-M.; Kim, Y.-G.	IEEE Access	2022	1157

lists the 25 most-cited East Asian and 2 Latin American papers on artificial intelligence (2020–2025). Titles cover a broad range of topics, including the implications of generative AI for research, techniques, applications in medicine and education, the metaverse, deep learning, and the Internet of Things. The top-cited article (“So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy) discusses the multidisciplinary impacts, opportunities, and challenges of ChatGPT, highlighting areas for further research in ethics, digital transformation, and education. The leading papers of systematic surveys and systematic reviews of the field, which tend to attract extensive citations are: “A Survey on Explainable Artificial Intel-ligence (XAI): Toward Medical XAI”, “A survey on deep learning and its applications”, “A survey on security and privacy of federated learning”, “A survey of transformers”, “A Survey on Metaverse: Fundamentals, Security, and Privacy”, and “The strategic use of artificial intelligence in the digital era: Systematic literature review and future re-search directions” top cited article from Latin America. Notably, the 25th most-cited paper in our list has 620 citations, which is only about $30\%$ of the citations of the top-cited paper (2067 citations). This steep drop-off illustrates how a small handful of publications capture a disproportionate share of total citations.

Note: Table 3,

Table 4. Top-cited AI-related publications (Part II).

No.	Title	Authors	Journal	Year	Citations
[8]	A survey on deep learning and its applications	Dong, S.; Wang, P.; Abbas, K.	Computer Science Review	2021	1031
[9]	A survey on security and privacy of federated learning	Mothukuri, V.; Parizi, R.M.; Pouriyeh, S.; Dehghantanha, A.; Srivastava, G.	Future Generation Computer Systems	2021	951
[10]	ChatGPT: Bullshit spewer or the end of traditional assessments in higher education?	Rudolph, J.; Tan, S.; Tan, S.	Journal of Applied Learning and Teaching	2023	909
[11]	Machine Learning	Zhou, Z.-H.	Machine Learning	2021	908
[12]	Federated Learning for Internet of Things: A Comprehensive Survey	Nguyen, D.C.; Ding, M.; Pathirana, P.N.; Li, J.; Poor, H.V.	IEEE Communications Surveys and Tutorials	2021	890
[17]	What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education	Tlili, A.; Shehata, B.; Adarkwah, M.A.; Huang, R.; Agyemang, B.	Smart Learning Environments	2023	882
[18]	Artificial intelligence: A powerful paradigm for scientific research	Xu, Y.; Liu, X.; Cao, X.; Wang, F.; Zhang, J.	Innovation	2021	867
[16]	A survey of transformers	Lin, T.; Wang, Y.; Liu, X.; Qiu, X.	AI Open	2022	745
[19]	Roles of artificial intelligence in construction engineering and management: A critical review and future trends	Pan, Y.; Zhang, L.	Automation in Construction	2021	722

and

Table 5. Top-cited AI-related publications (Part III: East Asia and Latin America).

No.	Title	Authors	Journal	Year	Citations
[20]	A strategic framework for artificial intelligence in marketing	Huang, M.-H.; Rust, R.T.	Journal of the Academy of Marketing Science	2021	721
[21]	Scientific discovery in the age of artificial intelligence	Wang, H.; Fu, T.; Du, Y.; Bengio, Y.; Zitnik, M.	Nature	2023	715
[22]	A Unifying Review of Deep and Shallow Anomaly Detection	Ruff, L.; Kauffmann, J.R.; Vandermeulen, R.A.; Dietterich, T.G.; Muller, K.-R.	Proceedings of the IEEE	2021	705
[23]	A Survey on Metaverse: Fundamentals, Security, and Privacy	Wang, Y.; Su, Z.; Zhang, N.; Luan, T.H.; Shen, X.	IEEE Communications Surveys and Tutorials	2023	701
[24]	Study on artificial intelligence: The state of the art and future prospects	Zhang, C.; Lu, Y.	Journal of Industrial Information Integration	2021	686
[25]	A Review on Representative Swarm Intelligence Algorithms for Solving Optimization Problems: Applications and Trends	Tang, J.; Liu, G.; Pan, Q.	IEEE/CAA Journal of Automatica Sinica	2021	636
[26]	Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence	Ali, S.; Abuhmed, T.; El-Sappagh, S.; Díaz-Rodríguez, N.; Herrera, F.	Information Fusion	2023	631
[27]	Pre-trained models: Past, present and future	Han, X.; Zhang, Z.; Ding, N.; Zhao, W.X.; Zhu, J.	AI Open	2021	630
[28]	Engaged to a Robot? The Role of AI in Service	Huang, M.-H.; Rust, R.T.	Journal of Service Research	2021	620
[29]	The strategic use of artificial intelligence in the digital era: Systematic literature review and future research directions	Borges, A.F.S.; Laurindo, F.J.B.; Spínola, M.M.; Gonçalves, R.F.; Mattos, C.A.	International Journal of Information Management	2021	457
[30]	Challenges and Opportunities of Generative AI for Higher Education as Explained by ChatGPT	Michel-Villarreal, R.; Vilalta-Perdomo, E.; Salinas-Navarro, D.E.; Thierry-Aguilera, R.; Gerardou, F.S.	Education Sciences	2023	313

Source: Scopus.

presents the top most-cited AI research papers in our dataset (2020–2025), as identified by total citation count in the Scopus query results. This list is purely determined by citation frequency. It is noteworthy that many of these highly cited works are comprehensive surveys or reviews, which tend to accumulate citations quickly, alongside a few highly cited applied research articles. Papers “Hunger Games Search: Visions, conception, implementation, deep analysis, perspectives, and towards per-formance shifts” and “IOBR: Multi-Omics Immuno-Oncology Biological Research to Decode Tumor Microenvironment and Signa-tures” were excluded because they did not meet the inclusion criteria regarding artificial intelligence research and innovation.

3.3. Most Productive Authors

Table 6. Top productive authors in artificial intelligence research (Part I: East Asia, 2020–2025).

Author	Affiliation	Location	Citations	Docs (20–25)	h-Index
Wang, Feiyue	Institute of Automation, Chinese Academy of Sciences	Beijing, China	52,595	49	100
Niyato, Dusit (Tao)	Nanyang Technological University	Singapore City, Singapore	67,287	43	119
Raja, Muhammad Asif Zahoor	National Yunlin University of Science and Technology	Douliou, Taiwan	17,734	34	67
Chiu, Thomas K.F.	Chinese University of Hong Kong	Hong Kong, Hong Kong	4770	27	37
Hwang, Gwojen	National Taichung University of Education	Taichung, Taiwan	29,652	27	88
Tlili, Ahmed	Beijing Normal University	Beijing, China	3377	26	25
Kang, Jiawen	Guangdong University of Technology	Guangzhou, China	13,083	25	50
Gupta, B. B.	Asia University	Taichung, Taiwan	22,265	24	78
Ting, Daniel Shu Wei	Duke-NUS Medical School	Singapore City, Singapore	18,807	24	62
Kim, Hee-cheol	Inje University	Gimhae, South Korea	2965	23	28
Ashraf, Imran	Yeungnam University	Gyeongsan, South Korea	5758	22	40
Du, Hongyang	The University of Hong Kong	Hong Kong, Hong Kong	3189	22	31
Sun, Bai	State Key Laboratory for Manufacturing Systems Engineering	Xi’an, China	6749	22	48
Li, Rita Yi Man	Hong Kong Shue Yan University	Hong Kong, China	5245	20	37
Xiong, Zehui	Singapore University of Technology and Design	Singapore City, Singapore	15,240	20	61

and

Table 7. Top productive authors in artificial intelligence research (Part II: Latin America, 2020–2025).

Author	Affiliation	Location	Citations	Docs (20–25)	h-Index
Valencia-Arias, Alejandro	Universidad Señor de Sipán	Chiclayo, Peru	1539	19	17
Rodrigues, Joel J.P.C.	Universidade Federal do Piauí	Teresina, Brazil	39,075	15	103
Valle-Cruz, David	Universidad Autónoma del Estado de México	Toluca, Mexico	803	13	14
Queiroz, M. M.	Fundação Getulio Vargas	Rio de Janeiro, Brazil	7214	12	31
Ramirez-Montoya, Maria Soledad	Tecnológico de Monterrey	Monterrey, Mexico	3908	12	32
Alanya-Beltran, Joel	Tecnológico de Monterrey	Monterrey, Mexico	221	11	8
Ovalle, Christian	Universidad Tecnológica del Perú	Lima, Peru	32	10	3
Acosta-Vargas, Patricia	Universidad de las Américas	Quito, Ecuador	944	9	16
Barbosa, Jorge Luis Victória	Universidade do Vale do Rio dos Sinos	Sao Leopoldo, Brazil	3358	9	30

Source: Scopus. Note: Brazilian authors dominate total citations in Latin America. East Asian authors present higher h-index values, indicating longer-term and larger-scale research trajectories in AI.

lists the 15 most productive East Asian and 9 most productive Latin American authors in the field of artificial intelligence research (2020–2025). Chinese, Singaporean, and Taiwanese researchers dominate the list, which does not align with China’s research leadership by country. Although productivity from Latin American researchers often appears in lower-impact journals. The h-index was included as a complementary metric to gauge the influence and sustained productivity of authors.

In contrast, in Latin America recent productivity in artificial intelligence research is noticeably lower, with volumes ranging approximately between 9 and 19 documents per author over the 2020–2025 period. Moreover, bibliometric impact appears more concentrated and asymmetric: the presence of one or two authors with extraordinarily high citation counts and h-index values can be observed, alongside a broader group of researchers displaying patterns typical of emerging trajectories, characterised by competitive recent output but more moderate accumulated impact indicators.

From a methodological standpoint, it is important to emphasise that the variable of recent productivity strictly reflects output within the analysed period, whereas citations and the h-index are cumulative measures recorded in Scopus over the course of an academic career. Consequently, these results should be interpreted as a combination of recent productivity and historical trajectory, rather than as evidence of a direct causal relationship between publishing more during 2020–2025 and achieving higher levels of bibliometric impact within that same interval.

3.4. University Affiliations

Table 8. Top East Asian universities by research affiliations in artificial intelligence research (Part I: East Asia, 2020–2025).

University	Country	Documents
Chinese Academy of Sciences	China	1251
Tsinghua University	China	762
University of Chinese Academy of Sciences	China	630
Zhejiang University	China	594
Shanghai Jiao Tong University	China	455
National University of Singapore	Singapore	451
Peking University	China	422
Sichuan University	China	414
Nanyang Technological University	Singapore	408
The Hong Kong Polytechnic University	Hong Kong	405
Huazhong University of Science and Technology	China	363
Wuhan University	China	335
Sun Yat-Sen University	China	328
Tongji University	China	327
Fudan University	China	326

and

Table 9. Top Latin American universities by research affiliations in artificial intelligence research (Part II: Latin America, 2020–2025).

University	Country	Documents
Tecnológico de Monterrey	México	239
Universidade de São Paulo	Brazil	223
Universidad Tecnológica del Perú	Perú	103
Universidade Estadual de Campinas	Brazil	95
Universidad Nacional Autónoma de México	México	77

Source: Scopus. Note: In East Asia, 13 of the top 15 universities are Chinese. In Latin America, only the Universidad Tecnológica del Perú is a private institution, with comparatively limited funding and fewer interdisciplinary programs. The Ministry of Education of the People’s Republic of China (1,086 documents) was excluded because it is an administrative affiliation rather than an educational institution. Therefore, the analysis focuses solely on academic institutions.

shows the 15 East Asian and 5 Latin American universities with the most document affiliations on artificial intelligence research publications. It is notable that a few countries—especially China, Mexico, and Brazil in Latin America—account for a disproportionately large share of universities. Further, the table underscores the con-centration of research activity: a handful of universities contribute a large fraction of all publications.

3.5. Most Cited Subject Areas

Table 10. Artificial intelligence research by subject areas.

Subject Areas	Documents
Computer Science	19,215
Engineering	13,233
Mathematics	6203
Medicine	5298
Social Sciences	5166
Physics and Astronomy	3830
Materials Science	3189
Decision Sciences	3062
Biochemistry, Genetics and Molecular Biology	2298
Energy	2249
Business, Management and Accounting	2132
Environmental Science	2003
Earth and Planetary Sciences	1258
Chemistry	1225
Chemical Engineering	1224

shows the top 15 subject areas within artificial intelligence research.

3.6. Most Cited Countries

Table 11. Artificial intelligence research by countries.

Country	Documents
China	26,752
South Korea	3269
Taiwan	1762
Japan	1758
Brazil	1686
Hong Kong	1305
Singapore	1212
Mexico	868
Colombia	560
Peru	545
Ecuador	411
Chile	369
Macao	334
Argentina	183

shows the top 14 countries in artificial intelligence research. China in East Asia and Brazil in Latin America lead, followed by South Korea and Mexico, respectively.

3.7. Regional Disparities

The corpus of AI literature analyzed (35,890 publications) reveals a marked geo-graphic concentration of research output and intellectual production. East Asia accounts for the largest share of publications, with China serving as the primary locus of production and institutional aggregation. This concentration is analytically significant, as it reflects not only differences in research volume but also the presence of consolidated institutional ecosystems that facilitate cumulative visibility, sustained collaboration, and higher citation impact. By contrast, Latin America displays a more diffuse and heterogeneous profile, characterized by a dispersion of output across smaller institutions and a lower average citation impact per document. Rather than indicating lower research relevance, this pattern suggests a structurally distinct mode of knowledge production, in which applied, context-specific research predominates under more constrained institutional and infrastructural conditions. These quantitative disparities co-occur with qualitative differences in thematic emphasis. East Asian outputs are disproportionately represented in foundational and methodologically driven subfields—such as algorithmic development, systems engineering, and advanced machine learning architectures—whereas Latin American contributions are more frequently concentrated in applied domains, such as public health, agriculture, and environmental monitoring. Such thematic bifurcation suggests divergent strategic orientations: one region advancing core technical capabilities likely to influence global standards and core tools, and the other prioritizing application-driven research aligned with local socioeconomic problems. The contrast aligns with prior characterizations of global capacity imbalances: industrialized and strategically invested ecosystems show greater facility for frontier research, whereas resource-constrained systems tend to orient research toward contextualized, applied problems [1,2]. Beyond volume and thematic focus, structural resource factors—access to high-performance computing, concentrated funding streams, and sustained institutional research programs—appear to reinforce East Asia’s lead. Conversely, Latin American institutions face constraints in financing, infrastructure availability, and skilled human capital [2,3]. These resource asymmetries do not merely reduce output; they shape what kinds of research are feasible, the speed of methodological adoption, and the capacity to generate widely cited, reusable technical artifacts.

3.8. Patterns of Collaboration

Network-level analyses reveal distinct collaboration architectures across the two regions. East Asia exhibits dense intra-regional co-authorship clusters anchored by high-centrality institutions—research centers and state-backed consortia—that function as hubs for cross-institutional and cross-sectoral collaboration. These hubs frequently bridge academia and industry, producing translational research and facilitating rapid translations from methodological advances into scalable implementations. The density and multiplexity of these networks create structural advantages: knowledge circulates quickly, mentorship pipelines form, and human capital accumulates in high-capacity centers. Latin America’s collaboration topology is comparatively sparse and modular. The region shows many small clusters with fewer high-degree hubs, and a higher prevalence of bilateral or localized collaborations (e.g., inter-university ties or university–NGO projects) rather than large consortia. Cross-border collaboration within Latin America exists but tends to be episodic and less institutionally consolidated than in East Asia. Inter-regional links (East Asia/Latin America) are present but limited, commonly mediated by a small number of transnational partnerships—often facilitated by multinational companies, diaspora networks, or targeted international funding programs—rather than by broad, institutionalized channels for sustained cooperation. This network heterogeneity has downstream implications for knowledge diffusion and absorptive capacity. Dense hub-and-spoke networks in East Asia enable rapid up-skilling, mentorship, and resource sharing (e.g., access to computer clusters and large datasets). In contrast, the modular network structure in Latin America constrains scale effects and slows the collective accumulation of frontier expertise. The bibliometric evidence suggests that collaboration patterns, not only raw publication counts, are decisive for the capacity to produce high-impact, methodologically influential AI re-search [1,2].

3.9. Policy Implications

The empirical patterns in output and network structure point directly to policy-relevant levers that could alter regional trajectories. First, the dominance of state-coordinated investment in East Asia underscores the strategic funding, public–private partnerships, and, national roadmaps in concentrated capacity. Second, Latin America’s fragmentation suggests that piecemeal funding and weak institutional linkages limit the emergence of high-capability centers; consequently, targeted policies to consolidate regional centers of excellence may yield outsized returns. Specific policy relevant findings from the results include: (1) the need for sustained infrastructure investments (e.g., national or regional shared HPC resources) to enable participation in computationally intensive subfields; (2) the importance of human capital strategies—including doctoral training, postdoctoral fellowships, and industry secondments—to address the shortage of skilled AI practitioners [2]; and (3) the role of open science and repository practices to increase visibility and citation of regionally produced work, potentially alleviating effects of publication-venue bias. Finally, the distribution of themes in the corpus suggests that policy instruments should be tailored to regional comparative advantage while closing structural gaps. For Latin America, policies that incentivize translating strong applied research into internationally visible technical contributions (for example, through joint methodological training programs, co-supervised PhDs with East Asian partners, and co-funded research infrastructures) could increase the region’s influence on global AI agendas. For East Asia, findings call for reflexive governance that integrates ethical, environmental, and social considerations into the rapid development of foundational capabilities [3,4].

4. Discussion

The comparative patterns observed in this bibliometric analysis suggest that the accelerating global diffusion of artificial intelligence is not unfolding uniformly but is instead reinforcing pre-existing structural asymmetries between regions. Although AI is rapidly transforming industries worldwide, the evidence from East Asia and Latin America demonstrates that the so-called global AI revolution is mediated by material capabilities, political economy dynamics, and institutional design choices that shape whether nations can position themselves as innovators or remain predominantly technology adopters [1]. In this respect, the descriptive disparities observed in the dataset are not merely quantitative gaps; they represent deeper divergences in innovation trajectories, policy orientations, and absorptive capacities across the two regions. East Asia’s research ecosystem—driven by China, Japan, and South Korea—illustrates how sustained public investment, coordinated industrial policy, and long-term national strategies generate cumulative advantages. These countries have consolidated what Aderibigbe et al. [2] describe as “robust technological infrastructures and well-established innovative ecosystems,” enabling them not only to scale AI research output but also to diversify into high-complexity subfields such as deep learning architectures, autonomous systems, and advanced computer engineering. China’s dominance in our dataset exemplifies how state-led investments create institutional density: large research consortia, significant computational resources, and vertically integrated academic–industry pipelines. The effect is a self-reinforcing cycle in which scientific capacity attracts additional funding, talent, and partnerships, thereby consolidating global influence. By contrast, Latin America’s more modest and fragmented contribution to AI research reveals the cumulative impact of the barriers highlighted in the literature. Limited financial resources, high infrastructure costs, and constrained public science budgets significantly narrow the range of research domains in which the region can compete [2]. In parallel, shortages of highly trained AI professionals and uneven educational capacity reinforce dependence on imported technology and externally produced knowledge, mirroring what the United Nations & International Labour Organization [3] identify as structural features of the digital divide. These constraints do not merely reduce the quantity of publications; they channel Latin American research toward applied, domain-specific work—often in healthcare, agriculture, and environmental sciences—while East Asia advances foundational methodologies and core algorithms. This thematic bifurcation reflects two different innovation models: one aimed at frontier research and global competitiveness, the other at localized problem-solving within resource constraints. The interdisciplinary landscape identified in the analysis further illustrates how these structural differences shape the regions’ long-term developmental prospects. The “multidisciplinary reach” of AI research, visible in the corpus of 35,890 publications, underscores the expanding interface between technical AI development and societal domains. Yet this expansion does not occur evenly. East Asia’s concentration in computer science engineering suggests a pipeline that prioritizes algorithmic innovation, hardware optimization, and system-level integration. Latin America’s thematic strengths—often in health, environmental sustainability, and social applications—align with urgent regional challenges but also highlight missed opportunities for deeper engagement in frontier subfields. This divergence has broader implications: without a stronger foothold in foundational research, Latin American institutions risk being structurally excluded from shaping global technical standards and ethical frameworks that increasingly govern AI deployment. The consequences extend to international development goals. As Hammerschmidt et al. [4] warn, inequalities in AI production “divide those who can and cannot create sustainable outcomes with AI,” hindering progress toward SDG #10 on reducing inequalities. Our findings empirically reinforce this concern: disparities in research output and citation performance parallel disparities in innovation capacity, digital infrastructure, and human capital. Unless addressed, these gaps may crystallize into a long-term technological dependency, in which Latin America remains primarily a consumer and not a co-author—of the next generation of AI innovations. Importantly, the analysis also reveals opportunities for strategic intervention. The persistence of regional silos suggests that cooperation—not only competition—could play a pivotal role in shaping inclusive AI futures. Cross-regional capacity-building initiatives between East Asian state-backed labs and Latin American universities offer a feasible mechanism to enhance knowledge diffusion. Models such as joint doctoral programs, shared computational infrastructure, or co-funded research networks could lower entry barriers and generate mutual benefits. Likewise, overlaying bibliometric trends with national R&D expenditure data could clarify whether Latin America’s lower citation impact due to underinvestment or to publication practices that favor local dissemination channels. Such an approach would deepen understanding of structural constraints and inform targeted policy responses.

5. Conclusions

This bibliometric review maps the contemporary evolution of AI research in East Asia and Latin America, revealing pronounced asymmetries in productivity, visibility, and thematic emphasis. China’s dominance—anchored in highly productive institutions and well-connected author networks—contrasts with Latin America’s fragmented landscape, where Brazil leads but struggles to achieve comparable global influence. These disparities reflect underlying structural inequalities in R&D investment, research infrastructure, access to international networks, publication, and dissemination practices. As such, the findings underscore the urgent need for policymakers in Latin America to treat AI as a strategic priority, supported by sustained funding, stronger institutional ecosystems, policies that reduce barriers to international collaboration, interdisciplinary emerging programs, open-access platforms, and incentives that expand cross-regional collaboration.

East Asian institutions, with strong intra-regional networks and state-backed consortia, produce a disproportionately high share of globally cited AI research, whereas Latin American AI output, though growing, remains less visible and often oriented toward local applications. Latin America’s research output and impact could be bolstered by more investment in foundational AI research, thematic priorities, and international collaboration.

This study is subject to limitations inherent to Scopus-based analyses. First, Scopus is a selective bibliographic database and thus cannot be treated as a full proxy for global knowledge production; coverage varies by language, region, and publication type, which may systematically affect Latin American visibility in particular. Second, bibliometric results depend on the quality of metadata, including affiliation country/territory labels and institutional name variants. Prior evidence shows that non-standard country/territory naming and database interface choices can lead to under retrieval or misattribution. Third, citation-based indicators are sensitive to recency effects (especially for 2024–2025 publications). Papers published in recent years have had less time to accrue citations, which can underrepresent their impact in our analysis. Upcoming studies could deepen the comparative analysis between East Asia and Latin America by incorporating additional databases (e.g., Web of Science) and even pre-print servers like arXiv.

Because our search was limited to publications explicitly mentioning ‘AI research’ or ‘AI innovation,’ it likely underestimates the volume of AI publications, particularly those focused on specific methodologies (e.g., machine learning, deep learning, neural networks, and natural language processing) that do not use that phrasing. We acknowledge this and encourage follow-up studies with expanded keywords to verify and extend our findings.

Further research should map co-authorship networks in greater detail and examine how national funding patterns, training ecosystems, and policy incentives shape research visibility. Cross-regional capacity-building models—such as joint doctoral supervision between East Asian laboratories and Latin American universities—could accelerate knowledge transfer and research visibility. Longitudinal studies are also needed to track how generative AI reshapes citations, patents, and real-world deployments, as well as its potential contributions to SDG-related challenges such as climate resilience, public health, and agriculture.