A recent study found that ChatGPT, AI chatbot, provided inappropriate and potentially harmful medical advice regarding vitreoretinal disease. The study, conducted by ophthalmologists at Tufts Medical Center in Boston, revealed that ChatGPT accurately answered only eight out of 52 questions (15%) about retinal health. However, when the questions were resubmitted two weeks later, all 52 responses changed, with 26 of them significantly altering the accuracy, some for better and some for worst. The researchers expressed grave concerns about the potential harm caused by inaccuracies in medical knowledge provided by AI chatbots.
In recent months, there has been a heated debate surrounding the optimal use of AI technology, and how to prevent it from overshadowing human capabilities. In the medical field, clinicians have been testing AI chatbots to assess their effectiveness in responding to healthcare-related questions. One of the challenges faced by ChatGPT is the inaccuracy rate, which might be attributed to the limited availability of online resources focusing on retinal health compared to other medical specialties.
The study revealed specific instances where ChatGPT provided factually inaccurate responses. For example, when asked about treatment options for central serous chorioretinopathy, a condition strongly affected by corticosteroid use, the chatbot advised using corticosteroids, which would worsen the condition! Additionally, the OpenAI chatbot incorrectly mentioned injection therapy and laser therapy as treatments for epiretinal membrane. These inaccuracies highlight the need for further improvement and accuracy in AI chatbots’ medical knowledge.
While this study provided valuable insights, its methods are not without criticism. Future studies could be improved by including blind assessments that hide whether advice comes from human clinicians or from ChatGPT. Additionally, experts emphasize the importance of identifying the source of bad information produced by chatbots, especially since most are trained from black-box models. Developers need to be aware of how clinicians and patients ask questions and ensure that chatbots can provide consistent accurate referenced responses.
Overall, this study highlights the limitations and potential hazards associated with AI chatbots’ medical advice. It emphasizes the need for advancements in chatbot knowledge, using evidence-based and source-transparent responses. AI developers should consistently update and improve these chatbots to deliver accurate and reliable medical information while recognizing the importance of human expertise in healthcare.
The whytry.ai article you just read is a brief synopsis; the original article can be found here: Read the Full Article…
