Abstract
Dear editor
ChatGPT is a large language model artificial intelligence that has revolutionized the field of natural language processing. ChatGPT can generate human-like responses to complex medical questions using language that is easy to understand. The potential use of ChatGPT in clinical has gain more and more attention since its presentation. The authors reviewed a lot and concluded with “potential value of ChatGPT in supporting the decision-making process for LSS diagnosis and treatment”. However, as decision-making in clinical is a complex process based on the results of medical history collection, physical examination, radiological evaluation, intraoperative feedback, and the subjective feeling the patients, we don’t think the ChatGPT can be involved deeply in decision-making process. The only and most valuable ability of ChatGPT is providing information upon request. And this kind of ability has many limitations and even risks in clinical decision-making. First, as ChatGPT has no sensory perceptual system at all, the ability of information providing is based on questions of the surgeon. Thus, if the surgeon makes a wrong judgement and asks a wrong question, the ChatGPT may response a wrong answer upon the clinical situation. The value of ChatGPT totally relies on the user, if the user is right in judgment, the ChatGPT is a sharp weapon, but if the user misjudges, the ChatGPT is a cup of poison. Second, though the ChatGPT is trained on large datasets, it can’t avoid the biases present in the data sources. These biases can inadvertently influence the generated content, including the code snippets or algorithmic suggestions, which can impact the fairness and reliability of the information provided. Although OpenAI strives to make ChatGPT a neutral and impartial tool that does not exhibit bias towards specific viewpoints, and it takes a variety of approaches to realize it. However, in some ways, the artificial setup or algorithm selves are sources of bias. Third, in practice, we can find the answers of ChatGPT are sometimes blurry, sweeping, and lack of operability. For example, if we ask the ChatGPT “When performing endoscopic interlaminar access to the vertebral space, how to safely and effectively rupture the ligamentum flavum? Please emphasize on detailing the surgical procedure.” The answers of ChatGPT are sweeping, and no details of the specific operation procedures is mentioned (https://chat.openai.com/share/8dc0142b-afbb-42d0-8d32-1edbc9694e6f). But if you ask an experienced surgeon, the answers maybe like “Use the endoscopic plasma knife to thin out a small area. Use a blunt probe to penetrate the thinned ligamentum flavum carefully. Then the dural sac can be pushed away by the water pressure. And a safety operation area is thus made. The remained ligamentum flavum can be removed easily by endoscopic forceps piece by piece”. So, taken together, ChatGPT is more like to be an information provider, the role as “decision-maker” in this paper is exaggerated in some way. And with the continuous evolution of ChatGPT, it may a more and more useful helper for surgeons.
