AI Research in Science Education Explodes, but Biology and Geology Lag Behind

Artificial intelligence has swept through education research at a pace that few academic fields have ever matched, and a new large-scale mapping of the literature now puts hard numbers on that transformation. A bibliometric analysis published in Discover Education examined 779 peer-reviewed publications indexed in Scopus between 2015 and 2025, focusing specifically on how artificial intelligence is being applied to the teaching of life and earth sciences, including biology and geology. The results reveal a field that has grown from a single publication in 2015 to 378 in 2025, with a dramatic inflection point beginning in 2023 that coincides with the arrival of generative AI tools such as ChatGPT.

The study, led by Smail Kannaoui of Ibn Tofail University in Morocco together with colleagues at the Regional Center for Teaching and Training Professions in Rabat, followed the PRISMA 2020 reporting standards and used the network visualization software VOSviewer to construct keyword co-occurrence networks, country collaboration maps, and journal co-citation graphs. The researchers began with a raw Scopus search that returned 10,259 records, then applied automatic filters for publication year, document type, source type, language, open access status, and subject area, reducing the set to 779 records. Every remaining record was screened by title and abstract to confirm genuine relevance to biology, geology, environmental science, or life and earth sciences education. The decision to restrict the corpus to open-access publications was deliberate: it ensured that all analyzed full texts were freely verifiable by any reader, though the authors acknowledge that this may have excluded relevant subscription-based research.

The publication curve tells a striking story. After a slow start, annual output reached 40 papers before nearly doubling to 75 in 2023. Then came explosive growth: 207 publications in 2024 and 378 in 2025, the two most productive years in the entire dataset. The authors are careful to note that a bibliometric design cannot establish a causal link, but the timing of the acceleration, coinciding with the widespread emergence of generative AI tools such as ChatGPT, is consistent with a broader reorientation of research priorities toward generative AI across the discipline. They also flag a cautionary note: growth this rapid raises questions about sustainability and about the risk of a trend effect that could influence the quality and depth of contributions.

Geographically, the field is dominated by the United States, which produced 168 documents, roughly 28.2 percent of the cumulative output of the top ten countries. The United Kingdom follows with 64 papers and China with 56. Spanish-speaking and German-speaking countries contributed 48 and 43 documents respectively, while Turkey and Australia published 38 and 37. India, Saudi Arabia, and Malaysia round out the top ten with 31, 26, and 25 publications, a significant presence of South Asian and Middle Eastern institutions in the research domain. The co-authorship network places the United States at an extremely central position with strong connections to many other countries, and Germany occupies a clear intermediary role within Europe. The clustering pattern reveals distinct geographic groupings, including a Middle Eastern-Asian cluster of Saudi Arabia, Turkey, Jordan, and South Korea; a European-Latin American cluster spanning Spain, Italy, Brazil, Mexico, and Colombia; and a Southeast Asian cluster of Malaysia and Indonesia.

Funding data reinforce the picture of American leadership. The National Science Foundation tops the list of sponsors with 16 funded publications, followed by the National Natural Science Foundation of China with 10. Three organizations funded eight articles each: Germany’s Federal Ministry of Education and Research, Taiwan’s Ministry of Science and Technology, and the National Institutes of Health. The United Kingdom Research and Innovation funded seven papers, while the European Commission and Tecnológico de Estudios Superiores each accounted for six. At the institutional level, Tecnológico de Monterrey in Mexico leads with 10 publications, followed by the University of Johannesburg, the University of Hong Kong, the University of Oulu, and the University of Georgia with nine each, and Stanford University, Carnegie Mellon University, the Chinese University of Hong Kong, Sultan Qaboos University, and University College London with eight apiece.

The disciplinary distribution shows a field that is multidisciplinary but uneven. Social Sciences account for the largest share of publications at 45.9 percent, reflecting the educational and societal dimensions of the research, while Computer Science contributes 15.5 percent, representing its technological side. Together the two fields make up 61.4 percent of the corpus. Environmental Science follows at 5.0 percent, then Psychology at 4.7 percent, Engineering at 4.2 percent, and Medicine at 4.0 percent. At the journal level, Computers and Education: Artificial Intelligence, published by Elsevier, stands out for influence, contributing 22 articles to the corpus in 2025 alone, while Sustainability, published by MDPI, ranks first among the top sources in overall productivity and citation totals across its entire multidisciplinary output. The authors caution that this editorial concentration on a few dissemination channels could signal an emerging structuring of the field with a narrow base.

Individual researchers also show a concentration of influence. Dragan Gašević of Monash University in Australia ranks as the most prolific author with 700 articles, 24,733 citations, and an h-index of 77. Kenneth R. Koedinger follows with 412 publications and an h-index of 70, and Sanna Järvelä of the University of Oulu has 240 publications, 13,464 citations, and an h-index of 62. The authors interpret these figures as evidence that a core group of highly experienced scholars anchors the field, even as newer contributors from emerging academic contexts, particularly developing countries, join the conversation. They note, however, that citation-based indicators are time-dependent and may favor older publications and established researchers, so comparisons should be read as indicative rather than definitive.

Perhaps the most consequential finding lies in the thematic structure. Keyword co-occurrence analysis with a minimum threshold of five occurrences produced five clusters. The largest, containing 428 keywords or 54.9 percent, centers on AI tools and generative technology, including machine learning and large language models. The second, with 133 keywords, focuses on teaching, learning, and higher education. But the third cluster, dedicated to science education and biology and geology specifically, contains only 40 keywords, or 5.1 percent, and sits isolated from the dominant AI cluster. This separation, the authors argue, demonstrates that there has been limited integration of generative AI into science teaching, and it constitutes the study’s core original contribution: a systematic disciplinary gap. Life and earth sciences differ epistemologically from general science subjects because they rest on observational methods, field study, practical experiments, ecology, and geologic time scales, all of which pose unique teaching challenges that demand discipline-specific AI solutions rather than generic ones.

The temporal overlay analysis sharpens this narrative into a five-stage trajectory. Early keywords such as machine learning, deep learning, and e-learning cluster around 2023.4 in average publication year, while newer terms including generative AI, large language model, AI literacy, academic integrity, and ChatGPT gather around 2024.7 to 2024.9, showing how completely the research agenda has pivoted since ChatGPT appeared in late 2022. Bibliographic coupling confirms the same arc: a foundational phase from 2015 to 2020, a growth phase of machine learning and e-learning from 2021 to 2022, a transition to generative AI in 2023, a dense ChatGPT-era cluster in 2024 marked by rapid idea exchange among research groups, and an emerging 2025 cluster devoted to AI literacy and ethics. That final cluster, smallest but strategically vital, signals a shift from asking how AI can be implemented to asking when and under what circumstances it should be, a question that carries special weight in the life sciences, where truthfulness, verifiability, and scrutiny of evidence are foundational epistemological values that careless use of generative AI could undermine.

The authors draw three principal implications from their mapping. First, discipline-specific empirical studies are needed, including longitudinal research on how AI-based instruction affects learning outcomes in biology, geology, and environmental sciences, moving beyond the prevailing higher-education orientation of the literature. Second, life and earth sciences teachers need training that goes beyond operational competence to include evaluating AI-generated content against scientific evidence standards, since prior work has shown that teacher acceptance of AI depends on self-efficacy, anxiety, and perceived usefulness, and that adoption varies markedly across scientific fields. Third, the lack of collaboration between AI experts, science educators, and cognitive psychologists represents a structural deficit that funders and journal editors should consciously address. The study has limitations worth noting: it draws on a single database, Scopus, and restricts itself to open-access and English-language materials, which may narrow its geographic and linguistic scope and exclude valuable work in Arabic, French, or Spanish. Still, as a first systematic map of AI in life and earth sciences education, the analysis makes one thing clear: a rapidly expanding, generative-AI-centered research field is racing ahead, while the disciplines that study life itself, and the deep time of the Earth, are still waiting for their turn.

Subject of Research: Bibliometric analysis of artificial intelligence research trends in life and earth sciences education from 2015 to 2025

Article Title: Artificial intelligence in science education with a focus on life and earth sciences a bibliometric analysis of global research trends from 2015 to 2025

Article References: Kannaoui, S., Elwahab, F., El Houda Derkaoui, N., Brhadda, N., Loukili, A., & Ziri, R. (2026). Artificial intelligence in science education with a focus on life and earth sciences a bibliometric analysis of global research trends from 2015 to 2025. Discover Education, 5(1), Article 1002. https://doi.org/10.1007/s44217-026-02208-z

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02208-z

Keywords: artificial intelligence, science education, bibliometrics, biology education, geology education, generative AI, ChatGPT, large language models, AI literacy, Scopus, VOSviewer, research trends