Artificial Intelligence in Science: A Helpful Assistant or a Misleading Tool?
Artificial intelligence (AI) tools are rapidly becoming embedded in everyday academic practice, transforming not only how scientific information is searched and analyzed, but also the very logic of research work. Speed, scale, and technological convenience are becoming the new standard. At the same time, there is a growing need to address fundamental questions: to what extent can AI-generated results be trusted, how can their transparency be ensured, and where do the limits of technology end and the researcher’s responsibility begin? These questions are explored through empirical evidence in the study “Assessment of the Applicability of Artificial Intelligence Tools for Scientific Literature Search and Analysis”, conducted within the Student Research Semester programme funded by the Research Council of Lithuania (RCL).
The study was carried out by Vilnius University (VU) student Vanesa Zaicaitė, who examined the use of AI tools across different stages of scientific literature search and analysis. The findings reveal a dual reality: on the one hand, AI tools significantly accelerate processes; on the other, they introduce new challenges related to the accuracy and reliability of results.
Speed and Accuracy: An Unresolved Tension
“Although AI tools enable faster identification, selection, and analysis of large volumes of scientific publications, their application does not always ensure sufficient accuracy of results,” says the research supervisor, VU Professor Dr Rasa Subačienė. According to her, AI-generated outputs are highly sensitive to how queries are formulated: “Using broader keywords may result in a considerable number of irrelevant publications. This shows that automation does not change the fundamental principle—scientific search processes remain dependent on the researcher’s competencies and decisions.”
The study confirms an ongoing shift from traditional, structured, and clearly defined search methods toward more automated solutions. However, this transition is not absolute. “This process cannot be considered fully autonomous and requires continuous user oversight of AI tools,” emphasises R. Subačienė.
Issues of Reliability and Transparency
The findings indicate that traditional literature search methods still provide greater accuracy and reproducibility, as they rely on clearly defined criteria and reliable databases. Meanwhile, as V. Zaicaitė notes, the functioning of AI tools often lacks full transparency—it is not always clear how results are generated or what filtering principles are applied. As a result, different tools or their versions may produce inconsistent outputs, complicating the comparability of research.
Practical Experience: Efficiency at an Additional Cost
“One of the biggest surprises was how effectively AI tools accelerated the initial search for scientific literature and helped quickly identify potentially relevant articles,” says V. Zaicaitė. However, she points out that this efficiency is not absolute: “The publications they provided did not always precisely match the research topic, which required additional evaluation during the article selection process.”
The study also recorded cases where AI-generated results raised doubts. “There were moments when it was necessary to consciously question the outputs provided by AI,” the student admits. In such cases, decisions were made based on additional analysis—evaluating article titles, abstracts, and their relevance to the research problem.
The Changing Role of the Researcher
These findings highlight a broader trend: the role of the researcher and the required competencies are evolving. The importance of critically assessing AI-generated information, formulating precise queries, and responsibly interpreting results is increasing. As noted by R. Subačienė, responsibility for AI-generated content, its validation, and its use ultimately lies with the researcher employing these tools.
In assessing the applicability of AI tools for different user groups, the study shows that they are beneficial for both novice and advanced researchers, though in different ways. “For beginners, they help to quickly gain an overview of a topic, while for advanced researchers, they help optimise research processes,” says V. Zaicaitė, while emphasising that effective use is inseparable from a solid understanding of the research process.
Future Direction: Complement, Not Replacement
The study leads to a balanced yet clear conclusion: in the near future, AI tools will not replace traditional methods of scientific literature search and analysis, but they will become an increasingly important component of them. Technological progress will enhance efficiency while also raising demands for transparency, reliability, and academic responsibility. It is precisely within this tension between innovation and control that a new reality of scientific research is emerging.
Information by the Research Council of Lithuania