Integrating an artificial intelligence chatbot in scientific communication: Dos and don’ts

Abstract

The art of narrative and storytelling in scientific prose is now considered as an integral part of any well-rounded scientific communication process. Hence, when the AI-driven ChatGPT facility was made available online many scientists wanted to assess this new marvel. Over the course of several months, we explored its functionalities and opportunities in the field of scientific publication and collected both positive and negative experiences.

Among the latter, the most compelling example was asking ChatGPT for scientific references. In this short communication we describe dialogs in which we sought to steer the AI towards fact-based assertions while it continued to produce false citations of papers, abstracts, communication to non-existing congresses, etc.

Through exchanges on the forum of the open AI platform we realized that other scientists have encountered the same problem, even when searching for patents and industrial property rights.

Our recommendation is that the AI of ChatGPT would be instructed on the role and uniqueness of scientific paper citations. Moreover, relevant agencies should plan for creating and adopting policies to rule out inappropriate behaviors by AI in fields in which errors could cost (at least) loss of trust in science by the general population.