|
Getting your Trinity Audio player ready...
|
What’s the last conversation you had with ChatGPT? Was it related to your academic work? Or thesis submission? Or perhaps research paper preparation? Or did you simply ask ChatGPT to suggest some good restaurants for a date night!?
Whatever the conversation was, ChatGPT has weaved its way into our daily life and redefined how chatbots are used. Many users consider it more than a tool. Traveling in a foreign country and want something translated? ChatGPT. Need a quick summary of a 7-season series before watching the next season? ChatGPT. Want to analyze your experimental data and interpret the results in a tabular format? ChatG…No, wait!
You’re an academic researcher. And sharing your sensitive, unpublished data with an AI tool must make you think twice! Because regardless of its immense benefits, ChatGPT has limitations when dealing with academic research. Yet, there are ways to address them. We’ve previously seen well-suited alternatives for ChatGPT with respect to academic editing. In this article, we’ll go beyond editing and learn how best to address the drawbacks of ChatGPT in academic research.
Limitation 1: Unreliable Answers
Limitation 2: Fabricated Citations
Limitation 3: Data Security Risk
Limitation 4: Improper Formatting
Limitation 5: Lack of Contextual Understanding
Limitation 1: Unreliable Answers
ChatGPT, to serve its users best, undergoes training through multiple sources of information. According to OpenAI’s official website [1], one of the sources of information is what’s publicly available on the Internet, and the other two include data shared by Open AI’s third-party partners and information generated by ChatGPT users and human trainers. Sounds like a healthy mix. So why is the reliability of the tool still questioned?
Scientific researchers are looking for more than just basic answers. Their complex research questions warrant detailed analyses and strong critical thinking before arriving at a conclusion. And ChatGPT is not entirely suited for this. Moreover, ChatGPT tends to add fluff to its outputs, which, at a surface-level might appear correct, but a deep dive would reveal insufficient research. There’s also a possibility of AI outputs being biased because of training models.
Solution: Any output generated by ChatGPT must be thoroughly vetted by humans. And we’re not referring to when ChatGPT is used for writing and editing research papers. Even if you are simply brainstorming ideas to identify the correct research problem, perform a critical analysis of the ideas suggested by ChatGPT before finalizing your research problem. Is the problem relevant? Is there a genuine need for a solution to the problem? Is it feasible to contribute scientifically to further existing research on the problem? Ask the right questions rather than rely on a tool!
Limitation 2: Fabricated Citations
There have been instances of ChatGPT citations being determined to be non-existent in several legal cases [2]. GPT-5, however, appears to have reduced the risk of AI hallucinations [3]. This implies that large language models (LLMs) can now predict more accurately. Yet this does not prevent generation of completely invented information, which ChatGPT could still do in order to supply the user with a satisfying output!
Solution: Avoid accepting ChatGPT’s output at face value. Always verify the citations in authentic sources like Scopus, Web of Science, DOAJ, Google Scholar, PubMed, etc. before listing the works in your research paper.
Limitation 3: Data Security Risk
Sharing data with AI tools is risky. Considering the sensitivity of unpublished data, directly posting your experimental data on ChatGPT to analyze or interpret them could pose a security issue. Although advanced training models do declare whether your data remains protected and paid versions of ChatGPT allows users to opt out of sharing the data for training purposes, you must be watchful of what data is shared and whether the correct settings are opted on the tool before using it.
Solution: Prefer tools that are designed for academics over general-use AI tools. With writing assistants like Paperpal, you can be assured that your information stays confidential and is not used for training AI models.
Limitation 4: Improper Formatting
ChatGPT can create tables, convert text into tabular data, and even generate images based on prompts. But formatting an entire research paper? That task still requires manual effort!
Journals typically outline their formatting requirements that must be satisfied for initial submissions. Each guideline must be understood, journal template (if provided) should be applied, page layout must be set, and other styling requirements should be reviewed and implemented accurately. ChatGPT cannot yet do this for authors.
Solution: Set aside time to check the accuracy of formatting of the manuscript. If it becomes too much of a hassle, choose manuscript formatting services, where experts ensure that all journal-specific guidelines are flawlessly applied to expedite your publication process.
Limitation 5: Lack of Contextual Understanding
ChatGPT is excellent for fixing grammar errors, drafting sentences, enhancing language, and tightening sentences. Yet, its corrections lack the understanding of research context. Here are a few examples of expert editor comments that you cannot always expect from ChatGPT:
“Generally, academic journals prefer the term ‘significantly’ to be used when reporting statistically significant differences (ideally with a P value). Since that is not the case here, I have replaced ‘significantly’ with ‘substantially’ in this context.”
“Consider returning to these metaphors in the Conclusion section to provide a neat structural ‘bookend’ as well as reinforce your conceptualization and address of this puzzle.”
“I am afraid your intended meaning here is unclear. ‘Authorities’ is a broad term that encompasses a very wide spectrum of bodies with authorities in all aspects of society and life (e.g., police, courts, governments, ethics boards, expert panels that create medical consensus regarding a topic, or even an individual who specializes in a topic who have produced enough work to be perceived as an authority on said topic). Thus, I recommend stating what authority you are referring to here. Or you could provide me with an explanation and I will happily work with you to ensure that your intended meaning is conveyed.”
Such context-based feedback, comments, and suggestions can only come from human editors and not an AI tool!
Solution: Consider using expert services to ensure that a human evaluates the submission readiness of your paper. Subject matter expert editors, who not only know the language but also thoroughly grasp the context of your study, can provide you with insightful comments and edits to make your manuscript journal ready.
References
1. How ChatGPT and our foundation models are developed https://help.openai.com/en/articles/7842364-how-chatgpt-and-our-foundation-models-are-developed
2. NY lawyer fined $5,000 for ChatGPT fake citations: Lessons to learn https://spellbook.com/learn/lawyer-who-used-chatgpt
3. Marked reduction in hallucination rates with GPT-5: A positive development for medical and scientific writing https://pmc.ncbi.nlm.nih.gov/articles/PMC12701941/
