In a rapidly evolving digital era marked by the proliferation of artificial intelligence technologies, understanding how emerging academic users engage with AI tools is paramount. A recent comprehensive study investigates the intentions of natural science doctoral students in Türkiye to adopt ChatGPT technology for educational purposes. This cutting-edge research advances our knowledge of behavioral interactions with AI by employing sophisticated methodological designs and analytical rigor. It offers crucial insights not only for academia but also for the broader scientific community interested in the societal implementations of AI-driven systems.
The study painstakingly selected its sample through a meticulous three-stage research design aimed at minimizing sampling bias and optimizing representativeness. Initially, the target population was identified as doctoral candidates specializing in mathematics, physics, chemistry, biology, and statistics across 35 leading Turkish universities. These institutions are distinguished by their ranking, falling between 20 and 60 in the national TUBITAK university scale, which denotes institutions renowned for academic excellence and a propensity to attract students with a strong inclination toward technology and artificial intelligence.
Subsequently, a custom-developed Python tool was deployed to extract detailed information about academic staff affiliated with natural sciences departments from Turkey’s central YÖKAKADEMIC database. This innovative approach facilitated direct communication with academics who then voluntarily provided contact information for their doctoral students. This multi-layered data collection method exemplifies the integration of automated digital tools within social science research frameworks, enhancing both efficiency and accuracy.
The final phase introduced an algorithmically generated randomized selection process via Python to ensure unbiased sampling. Here, randomized line numbers from a compiled list of potential participants were matched with sampled individuals, adhering strictly to established random sampling principles. This conscientious effort culminated in an initial sample size of 402 PhD candidates who completed an online survey during May and June 2024. After exclusions due to non-use of AI tools and incomplete responses, the effective sample comprised 361 doctoral students, a figure robust enough to satisfy established minimum sample size requirements for predictive modeling within the study’s analytical framework.
Central to the study’s inquiry was a structured questionnaire developed based on established Technology Acceptance Model (TAM) constructs, adapted to the peculiarity of ChatGPT use in education. The survey assessment included scales measuring social influence, perceived ease of use, perceived usefulness, AI self-efficacy, AI privacy and ethical trust, perceived threat to behavioral stability, and behavioral intention. These constructs have been validated extensively in prior academic research, ensuring the reliability and theoretical soundness of measurement. The questionnaire’s deployment embraced a concise 5-point Likert scale and was calibrated to maintain participant engagement with a completion time of approximately 15 to 20 minutes.
To fortify the content validity of the questionnaire, the researchers solicited expert feedback from four academicians specializing in Artificial Intelligence in Education (AIED). This consultative process resulted in critical refinements regarding linguistic clarity and conceptual precision, significantly enhancing comprehensibility. Following this, a pilot test was conducted involving 15 doctoral candidates in related scientific fields which provided invaluable insights leading to further adjustment of the survey instrument, aligning it closely with the nuanced context of doctoral education and AI integration.
The analytical backbone of the investigation rested on Partial Least Squares Structural Equation Modeling (PLS-SEM), adeptly chosen for its capacity to handle complex models with non-normal data distributions and formative constructs. Given the violation of multivariate normality assumptions detected via Mardia’s test, PLS-SEM offered the optimal solution to unravel the causal pathways affecting behavioral intentions towards ChatGPT adoption. SmartPLS software was utilized for this purpose, enabling simultaneous evaluation of the measurement and structural models with methodological rigor.
What distinguishes this study is its focus on causal inference within a technological acceptance context embedded in higher education. By elucidating how social influence interplays with AI anxiety and self-efficacy regarding ChatGPT usage, the research sheds light on intricate psychological and social dynamics underpinning technology adoption. It transcends simplistic acceptance models by integrating nuanced antecedents such as perceived ethical trust and AI-induced behavioral threat, marking a significant theoretical advancement in TAM-related literature.
Furthermore, the demographic composition of the sample was carefully recorded and analyzed, providing vital context. Though the comprehensive demographic breakdown is detailed elsewhere, critical attention was paid to ensuring gender balance, disciplinary diversity, and variation in technological proficiency. This demographic heterogeneity ensures that results are not confined to a narrow subset of doctoral candidates but rather reflective of the contemporary academic milieu within natural sciences disciplines.
Ethical considerations were rigorously upheld throughout the study. Participation was voluntary with no incentives offered, thus eliminating potential confounding influences due to extrinsic motivators. Informed consent was implied through survey completion, and the study’s conduct complied fully with institutional and international ethical standards governing human subjects research. Transparency and participant autonomy were prioritized, contributing both to the robustness and integrity of study findings.
The implications of these findings are manifold. As AI-powered language models like ChatGPT become ubiquitous in educational settings, understanding the behavioral mechanics shaping acceptance among doctoral candidates is crucial. This research provides educators, policymakers, and technologists with empirical evidence to tailor interventions that foster positive engagement, mitigate AI anxiety, and bolster trust. These elements are vital for successful integration of AI tools into curricula and scholarly workflows, ultimately enhancing academic productivity and innovation.
Moreover, the study’s methodological innovations, particularly the use of Python-based automated data gathering and random sampling in tandem with PLS-SEM, present a replicable model for future research endeavors. This fusion of computational methods with established social science paradigms exemplifies the interdisciplinary synergy needed to confront complex challenges posed by AI adoption in education and beyond.
Importantly, the research also highlights areas of resistance and concern. AI anxiety, often stemming from perceived threats to behavioral stability and privacy, may hinder widespread acceptance if left unaddressed. The interaction effects detailed in the study underscore the delicate balance between enthusiasm driven by ease of use and usefulness perceptions and the reservations instigated by ethical and psychological factors.
Future research directions emerging from this work include longitudinal studies tracking changes in acceptance and anxiety over time as AI technologies evolve and become more embedded in academic practices. Additionally, expanding the scope to include doctoral students from humanities and social sciences, or comparative analyses across different cultural contexts, could enrich understanding of the global landscape of AI use in higher education.
The current investigation stands at the forefront of AI adoption research, effectively combining advanced analytical techniques with a focus on an influential user group, natural science doctoral students, who represent the next generation of researchers and innovators. Its findings resonate strongly with ongoing debates about the role of artificial intelligence in shaping the future of education, research, and knowledge production.
In conclusion, this study offers a timely and rigorous exploration into the behavioral intentions of doctoral students towards ChatGPT use, embedding its analysis within a rich conceptual model supported by robust data and state-of-the-art analytical techniques. It sets a precedent for interdisciplinary research bridging computer science, psychology, education, and social science, and provides foundational insights that will inform both academic and practical efforts to harness AI’s transformative potential in a responsible and inclusive manner.
Subject of Research: Behavioral intentions of natural science doctoral students in Türkiye to use ChatGPT technology for educational purposes, analyzed through the Technology Acceptance Model (TAM) framework integrated with AI anxiety and social influence variables.
Article Title: Artificial intelligence, social influence, and AI anxiety: analyzing the intentions of science doctoral students to use ChatGPT with PLS-SEM.
Article References:
Uludağ, F., Kılıç, E. & Çelik, H. Artificial intelligence, social influence, and AI anxiety: analyzing the intentions of science doctoral students to use ChatGPT with PLS-SEM. Humanit Soc Sci Commun 12, 1308 (2025). https://doi.org/10.1057/s41599-025-05641-x
Image Credits: AI Generated
