IS CHAT-GPT A SILVER BULLET FOR SCIENTIFIC MANUSCRIPT WRITING?

Abstract

Open AI, an artificial intelligence laboratory of San Francisco California released their new AI model called ChatGPT in November, 2022.1 Chatbots are computer programs which are designed to carry out conversa­tion with humans through input via text or voice. GPT (Generative Pre-Trained Transformer) is a machine model that utilizes supervised learning techniques for understanding and generating human like language.2 It is an artificial intelligence Natural Language Process­ing (NLP) module which uses algorithms for analyzing and interpreting human language in the form of text or speech to extract meaningful and useful information based on human input. However, ChatGPT can extract data and phrases from over 150 billion parameters stored on the internet and can carry out long dialogues with its users.4 Chatbots using GPT can respond to questions very similarly as to a human's own response thus perfectly simulating human conversation. They can predict the next word in a sequence by detecting the context of previous words.3

ChatGPT has revolutionized scientific manuscript writing. It can assist in literature search and helps in gathering consistent and accurate data at a faster speed and with more ease. It also helps in reviewing and editing. In addition, ChatGPT can be used as a tool to assist in manuscript writing by helping authors draft a manuscript as per their (Journal’s) requirements A re­searcher’s time and efforts are significantly saved with ChatGPT, hence allowing them to concentrate more on data analysis and its interpretation.4

Ever since ChatGPT made its debut it has taken the scholarly world by storm, gaining authorship positions in four different published and pre-print articles.5-8 On the other hand, the place ChatGPT holds in author­ship criteria is currently being hotly debated among researchers, editors and publishers.9 According to the recommended authorship criteria, since ChatGPT can­not take responsibility for, or ensure the accountabil­ity and integrity of research, it is unlikely to fulfill the authorship criteria. This is endorsed by well reputed  journals like “Nature” and “JAMA”, making necessary changes in their policies of publication, by advising au­thors to mention AI in Methods or acknowledgement, if used.10,11 Likewise, the representative body of inter­national medical editors, World Association of Medical Editors (WAME) has banned inclusion of ChatGPT in the authors list of manuscripts.12

ChatGPT cannot replace human medical writers be­cause it lacks the expertise and level of understanding that humans possess. It can generate text that is con­vincing or “Sounds Plausible” but can possibly be non­sensical or incorrect.1 This is a common phenomenon exhibited by language models and has been termed as “Hallucination”. ChatGPT, at this moment, cannot pro­vide citations or references. It may also overuse phras­es. Moreover, errors and biases in the text generated by ChatGPT cannot be overruled. Articles generated with ChatGPT are not free of plagiarism and need to be corrected. As ChatGPT relies on previously stored data, text repetition can be a possible consequence, which may result in a lack of innovation, creativity and origi­nality. Another shortcoming of ChatGPT is that it cannot distinguish between fact, fiction and unreliable infor­mation. As a result, the potential misuse of ChatGPT cannot be overlooked as ChatGPT can be tricked and misused.13

In a nutshell, ChatGPT is an AI machine that can serve as a friend and guide to its user. It is a powerful authors tool that helps writers in accomplishing many tasks but despite its usefulness in medical manuscript writing, ChatGPT is not a Silver Bullet and can commit errors.14 The limitations and risks of ChatGPT should be given due consideration while adopting it for the purpose of medical manuscript writing.15 While on one hand ChatGPT can serve as a useful authors tool for writing, on the other hand the integrity of ChatGPT generated manuscripts are a challenge for editors as it can deceive inexperienced peer reviewers by generat­ing fabricated articles.16 ChatGPT has been termed as a “double-edged sword” and the scientific community has been advised to proceed cautiously with regards to it.17 The limitations of the ethical uses of ChatGPT in medical manuscript writ­ing have yet to be determined though. Dow­ling beautifully narrated the role of ChatGPT in research as “It ain’t what you do, it’s the extent that you do it, and that’s what gets (ethically acceptable) results.”18