Limitations of artificial intelligence (AI)

What researchers should know about using AI

It is undeniable that AI can be used in scientific writing to generate well-written passages, to check grammar and sentence structure in English and to suggest alternative phrases. However, AI needs to be used responsibly and ethically, under the direct supervision of a human expert.

Here are some aspects you should consider when writing your paper:

Since most AI systems rely on the Large Language Model (LLM), they lack the ability to associate relevant academic sources with any specific topic and, therefore, are very likely to fabricate imaginary references;

AI systems cannot verify the accuracy of the information they retrieve and there is a distinct risk that they may generate or report false or misleading content;

Because AI systems are trained with large amounts of data, they will incorporate any bias errors present in the data;

AI systems can manipulate data to support a research hypothesis or conclusion even if the facts are false or do not adequately reflect the original idea;

Depending on the topic and the availability of data on a subject, AI systems may summarize the wrong source or provide inaccurate summaries of specific articles and sometimes invent details and conclusions;

Since AI systems do not truly understand scientific principles, methodology, or experimental nuances, any conclusions offered by AI are likely to be unreliable;

AI-generated content has no author responsibility, a feature that not only conflicts with the standards of scientific integrity but generally violates journal policies and academic ethics;

In the long term, reliance on AI will weaken scientific creativity, conceptual innovation, and hypothesis-driven thinking;

Moreover, AI uses enormous amounts of energy and water in its operation. The International Energy Agency, for example, expects the global requirement for electricity by AI data-centers to exceed 940 TWh by 2030, representing a non-trivial contributor to greenhouse gas emissions. Each of us has a moral and environmental duty to reduce our own carbon footprint.

Although AI can support many aspects of scientific research, the creation of a manuscript worthy of submission to a quality international journal still demands the discernment and accountability of the authors and analytical revision by an experienced human editor.