Answers are grounded in this paper and cite their sources.
The American artificial intelligence (AI) researcher and writer, Eliezer Yudkowsky once said ‘by far the greatest danger of artificial intelligence is that people conclude too early that they understand it’. The development of several large language models (LLM) has generated interest in the abilities of AI chatbots like OpenAI’s ‘ChatGPT’ to construct academic manuscripts. The increasingly hostile environment of ‘publish or perish’, and ‘slice or perish’, may encourage the misuse of several AI chatbots being developed by multiple companies. What is a Chatbot? It is computer software that is trained using extensive libraries and, in turn, stimulates and processes written or spoken human communications. A human being can communicate with a machine like they do with other humans. In essence, they are conversational tools that can perform several functions (based on what they are designed for) for humans for example, voice chatbots like Alexa, Siri, and Google Assistant. ChatGPT and several of its sophisticated alternatives are designed to follow human conversational instructions and respond to them in detail. There are several limitations of the present generation chatbots. One, they are not conscious. They can only produce content based on the libraries they were trained upon and hence do not have original new thoughts. Second, they can produce factually incorrect answers which may sound credible. Third, the source of the AI content may be kept secret. Fourth, the information can be old (from the time of development of the AI software) and not current. Fifth, chatbots have the potential to respond to harmful instructions. Sixth, chatbots may facilitate the production of fraudulent papers like paper mills. Overall, there are questions about scientific integrity of the contents that chatbots produce in their present form. For example, when the authors asked ChatGPT how many patients of PANDO have acute dacryocystitis? Its response was a general text stating that if such numbers exist, ‘it will extract that information’. When asked ‘please extract that information?’, the response was which information and that there is a need to be specific. The human element of conversation continuity was missing. When asked, if acute dacryocystitis relapse after treatment?, the response again was a generic text that was not entirely factually correct. On further prompting on similar topics, it becomes evident that certain phrases were repeated from the previous answers or slightly recycled and presented. Chatbots and large language models are rapidly evolving, and it is possible that soon enough, many of these limitations will be overcome. ChatGPT is not all bad for science. Over time it can provide healthcare systems with several potential advantages like accelerating innovations, optimizing academic training, improving writing skills, statistical analysis, data analysis, enhancing knowledge database, and helping with radiology reports and discharge summaries. While the potential is enormous, the healthcare is not yet ready for it and several systems including ethics need to be in place before that happens. There are several ethical dilemmas in the context of their use for scientific publications. Can a chatbot like ‘ChatGPT’ be considered an author? Who is responsible for the manuscript, or the portion written by the chatbot? Who is responsible for the source attribution or its accuracy for the content produced by the chatbot? How can an AI-generated text be detected? What additional responsibilities would have to be taken by the journal editors? Each of these dilemmas can now be examined.
No takes yet. Share an insight, caveat, or question.
Ali et al. (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: