Generative Artificial Intelligence and Scientific Publications
Year: 2024
Type: article
Abstract: Artificial intelligence (AI), the ability of computer programs to reach human-level cognitive function, is fast encroaching on every sphere of human life. With their abilities to detect patterns in images and to learn from large datasets, AI-based algorithms have been shown to detect lesions in radiological images, histopathology slides, retinal images, and even skin lesions. With machine learning, these algorithms can predict outcomes in patients and can suggest molecules for drug development. A newer form of AI program which can produce text, images, video, or audio from text prompts is referred to as generative AI (GAI). A large language model (LLM) is a type of AI model designed to understand and generate human-like text based on the data it has been trained on. These models are built using deep learning techniques, particularly neural networks with many layers, and are trained on vast amounts of text data from diverse sources. Open AI, a company working on a type of LLM called generative pretrained transformers (GPT) released a chatbot version of GPT, called popularly as ChatGPT and released for public use in November 2022. From the day of release, ChatGPT has been creating waves with more than 100 million users within 2 months of release, which is a record for any technology adoption. ChatGPT can produce human-like text conversation using text prompts or queries. Another product of Open AI – DALLE can create realistic images using text prompts. By now, there are several LLM and GAI models for public use such as Gemini, Bard, Adobe Firefly, Stable Diffusion, Midjourney, and Meta AI. These GAI models can create human-like text conversations and produce videos or images from text prompts. The ability of GAI to produce fluent text and images is also being tried for writing scientific articles as well. Researchers have used GAI to produce authentic-looking scientific articles and even credited ChatGPT as the author of a scientific article which was later removed through a corrigendum.[1] Springer Nature has published an entire academic book with authors-generating content using GAI.[2] This has created a new wave of interest, anxiety, and apprehensions about the impact of GAI on scientific publications. Nature was the first journal to publish an update to their author guidelines and stated AI chatbots such as Chat GPT do not qualify to be listed as authors.[3] Several journals and publishers have followed the same. The guidelines of the International Committee of Medical Journal Editors and Committee on Publication Ethics have been updated to include policies regarding use of AI content in scientific manuscripts and authorship. Many journals have now updated the policy regard to use of GAI in publications as well as authorships. We have also updated the author instructions for the APIK Journal of Internal Medicine to include our policy regarding the usage of AI in the manuscripts. An AI chatbot as an author raises several legal issues involving complex copyright laws and ethical concerns regarding the accountability of an author and regarding the published content. An AI chatbot cannot take responsibility and accountability of the content of the article, almost all journals and publishers do not consider GAI chatbots such as ChatGPT as the author of the article. Even the images created by GAI can involve legal issues involving copyright laws. Applications with GAI capabilities need to be trained on large amounts of data involving hundreds of millions of images, many of them could still be under copyright protection, and the similarity of an image generated by AI suffers from the lack of proper attribution expected in such artistic contents. GAI and AI chatbots raise some serious questions about their usage in scientific articles and journals.[4] The text generated by them is fluent and convincing for the uninitiated but is generic and may have some factual errors or even blatantly false information. This phenomenon referred to as hallucinations in technology parlance is a serious concern for their reliability in areas of health-care/scientific information.[5] ChatGPT and Google Bard/Gemini warn the users with this disclaimer. They may provide references and citations which may not exist, and confirmation of the output information can be challenging. Researchers have tried to use GAI to create abstracts of scientific articles which were later reviewed for plagiarism and by human reviewers. Almost all abstracts passed plagiarism checks, but nearly, 68% of abstracts were detected to be fake by expert human reviewers, and one-third of the abstracts could escape the careful eyes of experts.[6] A false dataset was created with prompts using GAI which could potentially change the results and existing treatment of a medical condition.[7,8] At a time, when journals and publishers are burdened with fake articles and “papermill” articles for the sake of publication, and the new onslaught of AI-generated content can be a serious problem to detect fake articles, which may adversely affect the trustworthiness of the publications. There are ways to detect AI-generated content, but they are not always foolproof. In 2023, more than 10,000 published articles have been retracted for several reasons such as research misconduct, false data, and fabricated results.[9] Concerns are also being expressed about AI-generated images in hematology, histopathology, and images of sequencing or electrophoresis images which can pose a serious threat to academic integrity in scientific publications.[10] According to Retraction Watch, a website which maintains a database of retracted articles, around 4% of retractions are due to concerns in the images. Publishers are now using AI tools such as Imagetwin, ImaCheck, and Proofig to detect image manipulations or image duplication in the prepublications checks.[11] The present wave of interest in GAI capabilities is certainly likely to cause more worries for journals and editors to prevent publications with fake data or fake content with AI-generated content. More than 600 articles and editorials have been published since the public release of ChatGPT on this topic in various journals across the world.[12] GAI like any evolving new technology can have both advantages as well as disadvantages. If used properly, this can be a useful tool for language editing, especially for people from nonnative English-speaking countries. AI chatbots can be used to summarize the contents of scientific articles for quick reading. AI can be used to generate codes and formulas for data analysis, and it is imperative that all AI-generated content necessarily be checked for any wrong information or inaccuracies by an expert. GAI seems to have arrived here to stay for longtime and with the current pace of development can be a technology which can be transformative and even disruptive. It is essential for the scientific community to use this technology in a responsible manner so as not to undermine the basic tenet of “trust” in publications. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.
Author: Shrikanth N. Hegde
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W4402901838