ChatGPT in clinical practice – What is the way forward?

In this issue, there is an interesting article by Mittal et al. on 'Use of ChatGPT by physicians to build rehabilitation plan for elderly.'[1] The authors have tried to generate Artificial Intelligence (AI) responses for specific case scenarios. Various complex clinical scenarios are used, and ChatGPT provides tables with a rehabilitation plan. The output depends on the input – hence, a good history, meticulous clinical examination, and summary can only elicit such a response from the AI tool. When the questions raised by authors were copy-pasted in ChatGPT May 24 Version, the answer always started with 'I'm not a doctor, but I can provide you with a general example of a rehabilitation plan for……'. The responses were also slightly different from what the authors reported in the paper. ChatGPT is a language model developed by an organization called OpenAI. Being a language model, it has access to contents only till a time point at which the developers have decided. It doesn't give recent data, and sometimes it declines to give any specific data. For example: Question What is the trend of medical articles which contain the term 'ChatGPT' in the title or abstract. Response Since my training data only goes up until September 2021, I cannot provide real-time information on the specific trend of medical articles containing the term 'ChatGPT' in their titles or abstracts. As of my knowledge cutoff, 'ChatGPT' was primarily associated with OpenAI's language model and its applications in natural language processing tasks, including conversations and dialogue systems. To obtain the most accurate and up-to-date information on the trend of medical articles related to 'ChatGPT,' I recommend conducting a search on medical databases, academic search engines, or specialized platforms that index and provide access to medical literature. These sources can provide you with the latest information on the number and nature of articles related to 'ChatGPT' in the medical field. The most significant advantage of an AI should be its reproducibility (validity) and updated information. But here, the responses vary with versions (and users), and the AI tool declares that it has limited access, being a language model. This inconsistency precludes any recommendation on the utility of these tools in medical practice at the current date. We already have many technologies to refine medical education, patient care, and clinical decision-making. Undoubtedly, an advanced AI tool will be developed to further the benchmark in these areas soon. But we cannot neglect ethical, patient privacy, and professional liability issues before using these tools in clinical practice. Similarly, the role of AI in medical research and publication is another area of contention, but that is outside the scope of this commentary. In the era of evidence-based medicine, we need to explore the utility of these tools in various medical fields, including student training, clinical decision-making, patient support activities, medical research, etc., with good-quality studies. The free (open access) tools, which are easily accessible, can lure anyone to the comfort of delegating important work to AI. There is a definite scope for these companies to generate revenue from these in the future, which will add a burden to the increasing healthcare cost. AI does have the potential to affect job opportunities, particularly desk jobs. The lack of empathy and human value in patient care may impede AI in healthcare settings. ChatGPT and Artificial Intelligence Natural Large Language Models for Accountable Reporting and Use (CANGARU) is an initiative by Editors to bring transparency in usage of AI tools in research and publishing.[2] Similar effort is needed to bring transparency and accountability in usage of AI tools in patient related care. Till we have guidelines or laws to regulate these entities, we should practice extreme caution in using these tools in clinical care.