Artificial Intelligence in Science

por Camille Leal, Ph.D. (BALSA Science & Innovation)

Despite the growing impact and rapid advancement of artificial intelligence (AI) in recent years, AI has been developed and used in science since the 1980s, with the emergence of the first machine learning models. However, it was only in 2023, with the popularization of large language models (LLMs), that concerns about the impact of AI on science gained space in scientific and non-scientific journals and media outlets. As with the launch of the personal computer and the internet, artificial intelligence divides opinions. While many see AI as a major technological advancement that can positively change the way we live and work, others interpret it as a setback in intellectual development or even as a threat to their jobs.

A recent study published in Nature found that AI users published three times more papers, received five times more citations, reached leadership positions one and a half years earlier, and that researchers who used AI in the early stages of their careers were less likely to leave academia than those who did not use AI (Hao et al., 2026). On the other hand, the authors also identified a reduction in the diversity of topics, geographic areas, and citations in these papers. This may lead to an impoverishment of science, as only a small number of popular papers may be cited, masking the variety of questions addressed in other, smaller studies. What truly matters, however, is that AI is not a passing trend; it is here to stay and is increasingly being integrated into people’s daily lives. That said, AI still does not have the potential to replace scientists. However, it is very likely that scientists who do not make use of AI in their daily work will be more easily replaced by those who do. Therefore, we need to adapt to this technological advancement and learn how to use it efficiently and ethically.

Some relevant points to consider when using AI in research are:

1. Intellectual Responsibility — do not delegate scientific thinking

AI should not take on an authorship role because it cannot assume responsibility for errors, biases, plagiarism, hallucinations, or incorrect interpretations.

2. Qualified Human Supervision — never blindly trust statements or citations

AI systems often hallucinate and fabricate content, citations, and images. For this reason, it is necessary to go beyond AI and verify the citations and information it provides.

3. Protect sensitive and unpublished data

AI systems often use the data entered into them for their own learning. Therefore, it is not safe to input sensitive data, such as patient information, or data that have not yet been published.

4. Consider potential biases

AI may analyze information in a biased way based on its training. Therefore, always keep in mind that the information provided may not represent a broader context.

5. Disclosure of use

Whenever AI is used, make sure this is disclosed and specify how it was used. Always check the AI use policies of your institution, scientific journals, and the funding agencies supporting your research.

AI does not replace scientific thinking. But it greatly reduces wasted time!

The key is to use these tools to accelerate screening, synthesis, and organization, while reserving your brain for what only you can do: ask good questions, design and execute experiments, interpret results, and notice what no one else has noticed yet.

BALSA is sponsored by Consensus, an AI-powered academic search engine that helps you find, synthesize, and organize evidence from more than 200 million research articles, allowing you to search faster and at a larger scale. As part of this partnership, all BALSA members have free access to the premium version of the platform, expanding their research possibilities and making the process of finding scientific evidence faster, more practical, and more efficient.

Reference:

Hao, Q., Xu, F., Li, Y. et al. Artificial intelligence tools expand scientists’ impact but contract science’s focus. Nature 649, 1237–1243 (2026). https://doi.org/10.1038/s41586-025-09922-y