Artificial intelligence (AI) chatbots for self-diagnosis are on the rise, according to Benjamin Tolchin, a neurologist and ethicist at Yale University. Patients are turning to AI chatbots like OpenAI’s ChatGPT, Microsoft’s Bing, and Google’s Med-PaLM to seek answers to their health-related questions.
These large language models (LLMs) are trained on vast amounts of text from the internet and show good accuracy compared to traditional Google searches. Researchers and medical professionals are optimistic about the potential of chatbots to assist in answering medical questions and diagnosing diseases, especially considering the shortage of healthcare workers.
Tolchin acknowledges the impressive performance of these chatbots, with patients reporting reasonable and helpful responses. However, concerns exist regarding the accuracy of the information provided by chatbots, privacy issues, and biases inherent in the text used to train the algorithms, including racial and gender biases.
Tolchin also questions how people interpret and act on the information received, highlighting potential risks and harm associated with relying solely on chatbot diagnoses.
As per Scientific American, the field of medicine has increasingly moved online in recent years, accelerated further by the COVID-19 pandemic. Digital communication between patients and physicians has surged, and many healthcare systems already employ simpler chatbots for tasks like appointment scheduling and general health information. The more advanced LLM chatbots have the potential to take doctor-AI collaboration and diagnosis to new heights.
A study conducted by epidemiologist Andrew Beam and his colleagues at Harvard University found that when they presented patient symptom descriptions to OpenAI’s GPT-3, the LLM accurately identified the correct diagnosis within the top three possibilities 88 percent of the time. This outperformed online symptom checkers, which achieved a correct diagnosis within the top three possibilities only 51 percent of the time. Even physicians without the aid of AI achieved a correct diagnosis 96 percent of the time in the study.
The ease of use and conversational interface of chatbots make them more accessible than traditional symptom checkers. Patients can describe their symptoms in their own words, and the chatbots can ask follow-up questions, emulating a doctor-patient interaction. However, it is worth noting that the accuracy of chatbot diagnoses may vary depending on the quality and completeness of the patient’s symptom descriptions.
While the potential of AI chatbots in medical diagnosis is promising, further research, peer review, and careful consideration of ethical implications are necessary before widespread adoption can be recommended. Experts anticipate that a significant medical center will soon announce a collaboration involving an AI chatbot for diagnosing diseases.
However, this raises important considerations regarding potential charges for the service, data privacy protection, and accountability if someone experiences harm due to the advice provided by the chatbot. Training healthcare providers to effectively navigate the three-way interaction between AI, doctors, and patients will also be crucial for successful implementation.
While the implementation of AI chatbots in healthcare is expected to progress gradually, researchers emphasize the importance of a cautious approach, potentially limited to clinical research initiatives. The aim is to address any issues or challenges that arise while developers and medical experts work together to refine the technology.
Benjamin Tolchin, the neurologist, and ethicist from Yale University, finds it encouraging that during his tests, the chatbot consistently recommends seeking an evaluation by a physician, indicating a recognition of the need for human expertise in healthcare decision-making.
