Artificial Intelligence (AI) and Scientific Publishing

By Ari LaColla

Spring 2024 Intern

Introduction

Since ChatGPT launched in November 2022, Artificial Intelligence (AI) has been a topic of conversation in academic communities. Experts in all fields have been trying to decide if AI use should be allowed and, if so, how it can be detected and disclosed. The scientific publishing community also finds itself asking: what are the potential gains and risks of using AI to assist with writing scientific articles?

Use in Scientific Publishing

Generative text AIs such as ChatGPT are referred to as large language models (LLM) and can aid in writing and suggesting more comprehensive ways of explaining ideas. Using AI has the potential to give scientists more time to do research instead of writing papers. This might be idealistic, though, as there are many concerns as to the accuracy of AI writing.

While LLMs can replicate style, they often do not provide accurate information. If this goes unnoticed by the author of an article or a peer reviewer, misinformation could be accidentally included in a scientific paper. LLMs also may contain biases or copyrighted content if these are present in its system. A study by Cornell University found that two LLMs, ChatGPT and Alpaca, used biased language when provided with the same prompts with swapped genders.

Because of these issues, the use of LLMs is not widespread currently. A study by Nature found that only thirteen percent? of researchers who are not actively studying AI use an LLM in their writing process.

Benefits for Non-Native English Speakers

One of the potential benefits of AI use in scientific publishing is translation. While non-native English speakers might have to rely on a translator or editor before, AI can now assist with grammar and sentence structure. According to a survey of scientific researchers, fifty-five percent? believe the most beneficial use of AI in scientific publishing is translation.

Those with English as a second language have been systemically underrepresented in scientific publishing. A study by the Public Library of Science Biology found that non-native English speakers’ papers were rejected more than twice as often as their peers with English as their first language.

Unfortunately, AI might also hurt non-native English speakers as AI-detection software tends to flag their writing as being written by an AI. This is because those who learned English as a second language use less variety in their linguistics, which incorrectly triggers software meant to stop AI usage.

Publisher’s Reactions

Several scientific publishing companies have created guidelines around the use of LLMs by authors. A study of one hundred? publishers in May 2023 found that seventeen percent? of publishers and seventy percent? of journals have established rules around AI usage. Some publishers have also banned researchers from directly uploading their work into LLMs. This is because their writing may be fed back into the program to create responses, which may violate privacy policies publishers have in place.?

Another potential worry for publishers is that AI usage might lead to more “paper mills.” This term refers to publications that produce papers without research or scientific content. LLMs could make production easier for these kinds of papers, harming real publications.

Publishing Solutions Group

Publishing Solutions Group is dedicated to evolving alongside technology and being aware of changes being made in various publishing industries. We are committed to inclusivity and offer content development services to help to educate everyone. Our team stays informed, so we can help you in whatever field of publishing you specialize in.

[1]?Wan, Yixin, et al. "Kelly is a Warm Person, Joseph is a Role Model": Gender Biases in LLM-Generated Reference Letters." arXiv preprint arXiv:2310.09219 (2023).

[2]?Van Noorden, R., & Perkel, J. M. (2023, September 27). AI and science: What 1,600 researchers think. Nature News. https://www.nature.com/articles/d41586-023-02980-0?

[3]?Amano, T., Ramírez-Casta?eda, V., Berdejo-Espinola, V., Borokini, I., Chowdhury, S., Golivets, M., González Trujillo, J. D., Monta?o-Centellas, F., Paudel, K., White, R. L., & Veríssimo, D. (2023, July 18). The manifold costs of being a non-native English speaker in science. PLOS Biology. https://journals.plos.org/plosbiology/article?id=10.1371%2Fjournal.pbio.3002184?

[4]?Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023, July 14). GPT detectors are biased against non-native English writers. Science Direct. https://www.sciencedirect.com/science/article/pii/S2666389923001307?via%3Dihub?

[5]?Conroy, G. (2023, October 10). How CHATGPT and other AI tools could disrupt scientific publishing. Nature News. https://www.nature.com/articles/d41586-023-03144-w?