ChatGPT Changed the World—and Medical Writing Too: Friend or Foe?

Everyone from cancer patients to healthcare professionals is struggling to keep pace with rapid advances in artificial intelligence and medical writing. Although this important subject requires expertise, it has the potential to directly affect many different people’s lives.

The Birth of ChatGPT and Its Connection to Medical Writing

Developed by OpenAI, ChatGPT became known worldwide on 30 November 2022. GPT stands for Generative Pre-trained Transformer, an AI chat tool based on large language modeling. It stands out for its ability to generate natural-language responses to user input, a feature that has brought it worldwide recognition.

ChatGPT attracted attention by often giving highly accurate and comprehensive answers to questions in human language. As of January 2023, it had become the fastest-growing consumer software application in history, with more than 100 million users, raising OpenAI’s market value to 29 billion United States dollars. Free-tier users of ChatGPT can access GPT-3.5, while paying subscribers have access to the more advanced GPT-4.

ChatGPT is generally used as an aid in writing scientific research papers. First, a researcher develops an idea or concept and gives this information to ChatGPT. The tool can use it to generate more text, usually forming a draft or preliminary text. The researcher then adds details, makes corrections, reviews the text further and refines it. This can speed up scientific writing and give researchers more opportunity to focus. However, ChatGPT does not guarantee that information is current or accurate, so researchers must also include their own knowledge and findings.

Its impact on popular culture has been particularly strong among people following technology and AI. ChatGPT excites people inspired by technology and eager to use it in their own disciplines. However, others, including professional writers and content creators, may see it as harmful or threatening because its results are often incomplete or incorrect and may infringe others’ intellectual property.

Because ChatGPT cannot independently provide sources, it may also unintentionally borrow from another source or provide misleading or biased answers. This raises concerns in areas such as health advice, writing and programming, particularly when users tend to accept results as “good enough.”

Related article: Bill Gates – This Time, the Age of AI Has Truly Begun with ChatGPT

Medical publishing and medical writing have an important place in human history. The need to record and share health information gave rise to medical texts and publications. The history of medical publishing begins shortly after the invention of writing, with the earliest medical texts dating back to around 3000 BC.

With the announcement of ChatGPT at the end of 2022, AI language models such as ChatGPT became available for writing and editing scientific papers. In this context, users are expected to disclose how such tools were used and not list them as authors.

This article serves as a guide to understanding that process and the influence of such AI tools on medical publishing.

AI language models can help researchers write and edit scientific papers, according to a new statement from the International Committee of Medical Journal Editors (ICMJE). The committee stated that these tools should not be listed as authors and that researchers should disclose how AI-assisted technologies were used.

These new rules are the latest efforts by medical journals to establish policies for using large language models (LLMs) in scientific publishing. These AI-assisted tools can help with writing, data analysis, error detection and much more, but they are also prone to errors. It is also not entirely clear how information is stored and processed in these tools or who has access to it.

Experts also argue that these AI tools could benefit the field by reducing language barriers in scientific publishing and easing some of the monotonous or mechanical tasks involved in writing papers.

Experts agree, however, that the use of AI tools will continue to grow. At this point, there seems to be no avoiding it.

Change in Medical Publishing

OpenAI released ChatGPT in November 2022. In its own words, ChatGPT is “a deep-learning model trained on a large amount of textual data (170 trillion parameters in its latest form) to understand and generate text like a human”. Enter a question or prompt, and it will respond. For example, when asked how the AI tool could be used in scientific publishing, ChatGPT replies:

“ChatGPT can assist scientific publishing by generating ideas, explaining concepts, conducting preliminary literature searches, suggesting corrections and edits, and arranging and organizing scientific papers. However, it should be used as a support tool alongside verification from reliable scientific sources and subject expertise.”

Soon after ChatGPT became available, researchers began using it in their work. For example, Dr. Som Biswas, a radiologist at the University of Tennessee Health Science Center, used it to write 16 scientific papers in just 4 months. Five were published in four different journals. Biswas declined to be interviewed about these papers.

There were also reports of papers listing ChatGPT as an author, prompting criticism. In response, JAMA, Nature and Science published articles in January setting out policies for using ChatGPT and other large language models in scientific authorship. Journal editors at the American cardiology association and American College of Rheumatology also updated their policies to reflect the impact of AI writing tools.

The point of agreement is that AI should not appear on the author list.

“We don’t consider this appropriate, because coauthorship means taking responsibility for analyzing and generating the data in a paper. An AI-driven machine cannot take responsibility,” said Dr. Daniel Solomon, a rheumatologist at Brigham and Women’s Hospital and editor-in-chief of the ACR journal Arthritis & Rheumatology.

Problems with AI

A major concern about AI-assisted writing is that it can generate text that is unsupported by data or untrue.

In another instance, ChatGPT falsely alleged that a prominent law professor had committed sexual assault, even though no Washington Post article documenting the alleged event exists.

We live in a society deeply concerned about fake news. Such errors could certainly worsen the problem within the scientific community, which is concerning because science guides public policy.

Another problem is the lack of transparency about how large language models such as ChatGPT process and store the data used to carry out queries.

OpenAI addressed some privacy concerns by allowing users to turn off their GPT chat history so conversations could not be used to train or improve the company’s models. However, the terms of service remain quite unclear.

The real concern is that large language models (LLMs) are highly sophisticated, and many people do not fully understand the mechanism behind such a machine.

This means AI tools such as ChatGPT cannot replace an author: the technology is only an aid. However, experts wonder what will happen once AI itself begins using tools.

Another concern is that everyone may start using the same sentences, risking a loss of linguistic richness and variety.

More work is needed to see how AI’s role in medical publishing will develop. The ICMJE is considering creating a working group to provide general guidance on AI use in medical publishing.

For now, the relationship between AI and medical publishing is a matter of balance, rather than friendship or enmity —AI can support medical research but can also create significant problems with information access and accuracy.