ChatGPT: The Revolutionary AI That Can Write Scientific Articles

Scientific writing is a challenging and time-consuming task that requires a lot of research, analysis, and critical thinking. Many scientists struggle to publish their papers in reputable journals, and some of them are turning to ChatGPT, an AI program that can write scientific review articles for them.

What is ChatGPT?

ChatGPT is an AI program that uses natural language processing and deep learning to generate text based on a given topic. It can write essays, review articles, and summaries of the existing research on a subject, by sifting through the relevant literature, synthesizing the findings, drawing conclusions, and providing new insights.

It can write review articles faster and more efficiently than humans, and it can also use better grammar and vocabulary. However, it also has some serious flaws, such as inaccurate references, plagiarism, and inappropriate language.

How does ChatGPT compare to human writers?

Melissa Kacena, vice chair of orthopedic surgery at Indiana University School of Medicine, conducted a study to compare the quality of review articles written by ChatGPT, by students, and by a combination of ChatGPT and students. She published her results in Current Osteoporosis Reports.

She divided nine students and ChatGPT into three groups and assigned each group a different topic to write a review article on. She then asked faculty members and students to fact-check and evaluate the articles based on accuracy, readability, and language.

She found that the articles written by ChatGPT were easy to read and better written than the students’. However, up to 70% of the references cited by ChatGPT were inaccurate: they were either merged from different studies or completely made up. It also copied some sentences from other sources without proper citation.

Kacena said that ChatGPT was very convincing with some of the false statements it made and that they were hard to detect without a good knowledge of the data or thorough fact-checking. She also said that ChatGPT used more flowery language that was not suitable for scientific journals unless the students instructed it to write from a graduate-level perspective.

How can it be improved?

Kacena said that ChatGPT is not ready to author scientific journal articles by itself. Still, it can be improved and become a useful tool for researchers with proper programming and training. She said that it was good at suggesting ways to summarize data in figures and graphs and that it learned from the feedback provided by the students.

She suggested that ChatGPT could be programmed to not merge references to treat each scientific journal article as a separate reference and to limit copying consecutive words to avoid plagiarism. She also said that it could benefit from more user input and guidance.

Kacena believes that it has the potential to help researchers with the writing process and to generate new scientific insights. She said that ChatGPT is here to stay and that figuring out how to make it better and how to use it in an ethical conscientious and scientifically sound manner is very important.

What are some other AI programs for scientific writing?

Many AI programs can help researchers with various aspects of scientific writing, including literature reviews, data analysis, grammar checking, citation management, and more. Some examples of these programs include:

• Scite Assistant: This AI-powered research tool assists researchers in finding, reading, and comprehending scientific literature. It employs deep learning to analyze citation statements from millions of papers and classify them as supporting, contradicting, or mentioning the cited work.
• Consensus is an AI search engine that assists researchers in finding relevant research papers. It processes the user’s query using natural language understanding before ranking the results based on semantic similarity, citation impact, and recency.
• Elicit: This AI research assistant assists researchers with brainstorming, research, and writing. It uses natural language generation to generate outlines, summaries, and drafts based on user input and feedback.

These are just a few of the AI tools that can assist researchers with scientific writing. There are numerous other tools available that can provide varying features and benefits.