Since the advent of text-generating artificial intelligence (AI) programs like ChatGPT, the question of when and how to utilize them in writing research papers has sparked considerable debate.
According to Science magazine, over the next few months, around 4,000 researchers from various countries and disciplines will participate in developing a set of behavioral guidelines that can be widely adopted in the academic publishing domain, establishing a unified standard representing consensus within the research community, to replace the currently fragmented guidance.
The initiative, spearheaded by the collaborative organization CANGARU, comprises researchers, publishers such as Elsevier, Springer Nature, Wiley, representatives from journals like eLife, Cell, British Medical Journal, and the Committee on Publication Ethics (COPE).
Giovanni Cacciamani from the University of Southern California, a leader within CANGARU, stated that they aim to release the final guidelines by August, with yearly updates to account for rapid technological advancements. The guidelines will outline scenarios where authors should refrain from using large language models (LLMs) powering chatbots and how they should disclose other uses.
Prior to this effort, several journals including Science, Nature, and organizations such as the International Association of Scientific, Technical, and Medical Publishers (STM) and the European Commission have individually established rules on how scientists should employ generative AI tools in their work, albeit with varying degrees of guidance. The array of guidelines has left researchers feeling perplexed.
"Hence, the need for standardized guidance is both necessary and urgent," remarked Jean-Christophe Bélisle Pipon, a health ethicist at Simon Fraser University in the United States.
Many researchers express concerns about the slow progress in guideline development, noting that the world is constantly evolving, and the pace of generative AI development is accelerating.
CANGARU is currently conducting a systematic review of relevant literature to inform the guidelines. Additionally, a panel consisting of researchers, clinical doctors, computer scientists, engineers, methodologists, and editors will evaluate the guidelines.
