In a world where AI-assisted writing and reviewing are prevalent. The very essence of the scientific paper could undergo a profound transformation.
When faced with writer’s block during the composition of a research paper, radiologist Domenico Mastrodicasa turns to ChatGPT, a chatbot renowned for generating articulate responses to almost any query within seconds. “I use it as a sounding board,” says Mastrodicasa. who is affiliated with the University of Washington School of Medicine in Seattle. “I can produce a publication-ready manuscript much faster.”
Mastrodicasa is just one of many researchers who are currently exploring. The capabilities of generative artificial intelligence (AI) tools for crafting both text and code. He opts for the ChatGPT Plus subscription. which is based on the large language model (LLM) GPT-4. And he incorporates it into his workflow a few times each week. Specifically, he finds it particularly valuable for offering more lucid ways to express his ideas. According to a Nature survey, while scientists who regularly utilize LLMs are still a minority. There is an expectation that generative AI tools will increasingly become integral for assisting in the writing of manuscripts, generating peer-review reports, and preparing grant applications.
These applications represent just a fraction of the potential ways in which AI can revolutionize the landscape of scientific communication and publishing. Science publishers are already in the experimental stages of incorporating generative. AI into scientific search tools, as well as using it for editing and swiftly summarizing research papers. Many experts believe that non-native English speakers stand to gain the most from these tools. Moreover, some envision generative AI as a means for scientists to reimagine how they analyze and condense experimental findings. By harnessing the capabilities of LLMs, researchers could delegate a substantial portion of this work to AI. Thus reducing the time spent on writing papers and affording more time for conducting experiments.
“It’s never really the goal of anybody to write papers — it’s to do science,” remarks Michael Eisen, a computational biologist at the University of California, Berkeley, and the editor-in-chief of the journal eLife. He envisions that generative AI tools could potentially revolutionize the very essence of scientific papers.
However, the looming challenge of inaccuracies and falsehoods poses a significant threat to this vision. Large language models (LLMs) are essentially engines design to generate text that adheres to the stylistic patterns of their inputs. And rather than for ensuring factual accuracy. Publishers are concerned that the increased utilization of LLMs might result in a surge of low-quality or error-ridden manuscripts, and potentially an influx of AI-assisted fraudulent content.
Laura Feetham, responsible for overseeing peer review at IOP Publishing in Bristol, UK. Which publishes journals in the physical sciences, comments, “Anything disruptive like this can be quite worrying.”
The potential for a deluge of AI-generated fraudulent content is a major concern. Science publishers and other stakeholders have identified a range of worries regarding the impact of generative AI. The accessibility of generative AI tools could make it easier to create subpar papers. SAnd in the worst-case scenario, compromise the integrity of research, notes Daniel Hook, the CEO of Digital Science, a research-analytics firm in London. Hook states that publishers have valid reasons to be concerned.
There have been instances where researchers have acknowledged using ChatGPT to assist in paper writing. But failed to disclose this fact. They were caught because they inadvertently left behind telltale signs, such as fake references. The software’s default response indicating that it is an AI language model.
In an ideal scenario, publishers would have the means to detect text generated by LLMs. However, AI-based detection tools have thus far struggled to consistently distinguish such text from human-written prose without generating false positives. the scientific paper
While developers of commercial LLMs are actively working on watermarking LLM-generated content to make it identifiable. No company has yet implemented this for text. Additionally, watermarks could potentially be removed. As point out by Sandra Wachter, a legal scholar at the University of Oxford, UK, specializing in the ethical and legal aspects of emerging technologies. She hopes that lawmakers worldwide will mandate disclosure or watermarks for LLMs and potentially make it illegal to remove. These identifying markers.