AI Research Is Reshaping the Global Push for Sustainable Development, Landmark Analysis Finds

Artificial intelligence has quietly become one of the most powerful engines driving research on the United Nations Sustainable Development Goals, and a new large-scale analysis has now mapped exactly how that transformation happened. In a study published in Discover Sustainability, Reetu Verma of Maharaja Surajmal Institute of Technology and her colleagues combined classical bibliometrics with a modern machine-learning technique called BERTopic to dissect more than three decades of scholarship linking AI to sustainability. Drawing on 3,262 unique scholarly records published between 1989 and 2025, harvested from the Scopus and Web of Science databases after careful deduplication and screening, the team produced one of the most detailed pictures to date of how the worlds of algorithms and global development goals have converged. The results reveal a research field that was once a niche intersection of computer science and environmental policy but has exploded into a central arena of twenty-first-century science.

The numbers tell a striking story of acceleration. While AI and sustainability research existed in scattered pockets well before the turn of the millennium, the analysis found a significant surge in AI-SDG scholarship beginning after 2015, the year the United Nations adopted the 17 Sustainable Development Goals as its flagship global agenda. That surge then accelerated dramatically after 2020, suggesting that the COVID-19 pandemic era, with its heavy reliance on digital technologies, may have further entrenched AI as a tool for tackling development challenges. To quantify this growth, the researchers applied well-established scientometric indicators, including the Relative Growth Rate, which measures how quickly the literature expands year over year, and the Doubling Time, which estimates how long it takes for the body of publications to double in size. They also computed the h-index and g-index, two metrics that capture both the productivity of researchers and the citation impact of their work, and tested the field against Lotka’s Law, a classical model describing how scientific productivity is distributed among authors.

Geography matters enormously in this story. The analysis identified India, the United States, and China as the most productive contributors to AI-SDG research, a finding that reflects both the scale of these countries’ research enterprises and their growing investment in AI capabilities. When the researchers shifted focus from raw output to research impact and citation quality, a slightly different hierarchy emerged: the United States, India, the United Kingdom, and Australia led the pack, indicating that their papers tend to attract the most scholarly attention and influence. This distinction between productivity and impact is a recurring theme in scientometrics, because a country can publish enormous volumes of work while producing relatively few highly cited papers, or vice versa. The fact that India appears at the top of both lists underscores its rapid rise as a hub for AI-for-development research, while the presence of the United Kingdom and Australia among the impact leaders highlights the role of well-funded Western institutions in shaping the field’s direction.

Perhaps the most revealing part of the study is its thematic evolution analysis, which traces how the intellectual DNA of the field has shifted over time. In earlier decades, research clustered around foundational themes: artificial intelligence itself, machine learning, climate change, and decision support systems. These were the building blocks, the raw computational and conceptual tools that researchers were still learning to apply. Over time, however, the thematic landscape migrated toward more integrated and explicitly sustainability-oriented topics, with sustainable development and the SDGs themselves becoming central organizing concepts. The researchers used Biblioshiny, the R-based interface of the Bibliometrix package, and VOSviewer, a popular network-mapping tool, to construct and visualize these conceptual structures. Their thematic linkage maps demonstrate the persistent foundational role of AI and machine learning, alongside the growing centrality of climate change and sustainable development, reflecting what the authors describe as the increasing convergence of AI research with global sustainability priorities.

The methodological innovation at the heart of the paper is its use of BERTopic, a topic-modelling technique that goes beyond the keyword counting of traditional bibliometrics. BERTopic leverages transformer-based language models, the same family of neural architectures that power modern conversational AI, to represent each document abstract as a dense numerical vector that captures semantic meaning rather than just word frequency. It then clusters these vectors to discover latent themes that authors may never have named explicitly in their keywords. Applied to the abstracts of the 3,262 records, BERTopic uncovered eleven interpretable thematic domains, after the researchers excluded one heterogeneous outlier category that failed to cohere into a meaningful topic. This semantic layer of analysis complements citation- and keyword-based approaches by revealing what the literature is actually about at the level of language and meaning, offering a richer and more nuanced map than co-word analysis alone can provide.

The eleven domains read like a catalogue of the most pressing challenges on the global development agenda. Digital transformation and green innovation emerged as a dominant theme, capturing research on how digital technologies enable environmentally friendly business models and industrial processes. AI-enabled education formed another major cluster, reflecting the deployment of intelligent tutoring systems and adaptive learning platforms in pursuit of quality education for all. Digital health appeared as a distinct domain, encompassing machine-learning diagnostics, telemedicine, and health-system optimization. Smart agriculture, renewable energy systems, and clean water and sanitation each constituted their own thematic territories, underscoring how deeply AI has penetrated the practical machinery of food security, energy transition, and water management. Remote sensing, natural-language-processing-based decision support, sustainable construction, tourism, and sustainable chemistry and pharmaceutical applications rounded out the map, demonstrating that AI’s sustainability footprint extends from satellite imagery of forests to the molecular design of greener drugs.

What makes this convergence scientifically significant is the way these domains interlock. Remote sensing powered by deep learning feeds climate models and deforestation monitoring; natural language processing distils policy documents and scientific literature into actionable decision support; smart agriculture systems draw on weather prediction, soil sensors, and computer vision to raise yields while cutting water and chemical use. The BERTopic analysis suggests that these applications are no longer isolated experiments but have matured into recognizable, self-sustaining research communities with their own vocabularies, methods, and citation networks. The temporal evolution of the topics also hints at where the field is heading: as foundational AI techniques become commoditized, the frontier is shifting toward integration, governance, and measurable impact on real sustainability outcomes rather than proof-of-concept demonstrations.

The study is not without its limitations, and the authors are transparent about the scope of what bibliometrics can show. A scientometric map reveals patterns of publication, collaboration, and citation, but it cannot measure whether AI applications actually improve lives or reduce emissions on the ground. The dataset, while large, is restricted to two commercial databases and to English-language scholarship indexed by Scopus and Web of Science, which may underrepresent work from the Global South published in regional outlets. The BERTopic clusters, though interpretable, depend on modelling choices such as embedding model and clustering parameters, and the exclusion of one outlier category illustrates the judgment calls inherent in unsupervised topic discovery. The authors also disclose that parts of the manuscript text were refined using ChatGPT to enhance readability, while all analyses, interpretations, and conclusions were designed, verified, and approved by the researchers themselves, a disclosure that has become increasingly common and important in the era of generative AI.

For policymakers, funders, and researchers, the map that Verma and her colleagues have drawn offers both a scorecard and a compass. It confirms that the AI-SDG literature has moved from the margins to the mainstream of sustainability science, with India, the United States, and China setting the pace of production and the United States, United Kingdom, and Australia driving citation impact. It shows that the field’s centre of gravity has shifted from generic AI techniques toward concrete development domains such as health, education, agriculture, energy, and water. And by exposing the latent semantic structure of thousands of abstracts, it identifies where research energy is concentrated and, by implication, where the gaps lie, from underexplored SDG domains to the need for stronger evidence of real-world impact. As the 2030 deadline for the Sustainable Development Goals approaches, studies like this one provide the cartography that the next generation of AI-for-good research will need to navigate by.

Subject of Research: Scientometric and topic-modelling analysis of artificial intelligence research on the Sustainable Development Goals

Article Title: An integrated scientometric and BERTopic-based analysis on examining the role of artificial intelligence in achieving Sustainable Development Goals (SDGs)

Article References: Verma, R., Kumar, A., Pabreja, K., & Kumar, S. (2026). An integrated scientometric and BERTopic-based analysis on examining the role of artificial intelligence in achieving Sustainable Development Goals (SDGs). Discover Sustainability. https://doi.org/10.1007/s43621-026-04841-y

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04841-y

Keywords: artificial intelligence, Sustainable Development Goals, scientometrics, BERTopic, bibliometrics, machine learning, climate change, sustainable development, digital health, smart agriculture, renewable energy, research trends