A recent Journal of the American Academy of Dermatology study has evaluated the risks and benefits of consumer-facing artificial intelligence (AI) in clinical dermatology.
Study: The Chatbots are Coming: Risks and Benefits of Consumer-Facing Artificial Intelligence in Clinical Dermatology. Image Credit: greenbutterfly / Shutterstock
AI Chatbots in Dermatology
AI has advanced tremendously in recent times, bringing it closer to its implementation in dermatology. As an example, image-based classifiers using AI have been deployed to detect skin cancer, and they have often outperformed dermatologists.
A new form of artificial intelligence is AI chatbots. They are available to the public and simulate conversations with human users. When a question is put to the platform, natural language understanding is used by chatbots to determine what the user wants to know. Subsequently, an answer is generated by a neural network trained with large text datasets. In doing so, words are selected based on their probability distribution. Chatbots have the capability to learn and improve over time with repeated interactions. There are many models currently under development, and OpenAI’s ChatGPT is one such program.
Advantages of Chatbots
Translation of medical jargon into patient-friendly language is one key use of Chatbots. Recent research has noted the efficacy of ChatGPT in simplifying radiology reports and that several radiologists have validated the translation to be accurate and complete. During the coronavirus disease 2019 (COVID-19) pandemic, Penn Medicine used a simpler chatbot to answer patient questions when their call center was overwhelmed.
Currently, the wait times for dermatology appointments are quite long, and patients often use online resources for self-diagnosis. It has not been fully established whether such online tools offer better information than existing resources, and there could well be potential harm from incorrect medical advice and misdiagnosis. The authors of the present study tested ChatGPT against different scenarios generated from the list of the 25 most asked dermatological conditions on DermNetNZ.
It was observed that for a “red itchy spot,” rosacea in a light-skinned person, and a black streak in the nail, ChatGPT generated a reasonable differential diagnosis. It also suggested treatments and recommended consulting a doctor for further evaluation. ChatGPT also successfully answered questions on skin cancer risk, wound care, hair care, sunscreen, and acne. These results prove that chatbots could provide reliable information in situations where a patient may not be able to see a physician on short notice.
Limitations of Chatbots
An important limitation while taking advice from the chatbot is that patients need to describe skin morphology very accurately. This is because the chatbot is entirely dependent on user input. For example, when ChatGPT was presented with a growing red bump” on the lip or eyelid of an infant, it diagnosed a cold sore and a stye, respectively. However, the diagnosis of hemangioma was not considered. Hemangioma was suggested by ChatGPT when it was presented with a “growing purple bump” instead.
Another significant limitation is that chatbots are quite limited in their knowledge. They are unaware of developments in the medical field beyond the date of their training set. For instance, ChatGPT incorrectly stated that baricitinib was not FDA-approved for alopecia areata since its training data extended to 2021 only. Racially incomprehensive training data could also lead chatbots to face difficulties responding to patients with different skin types. While diagnosing a patient with a darker skin tone, ChatGPT overlooked rosacea as a plausible diagnosis.
In sum, chatbots have demonstrated promise in answering patient questions, but given their current drawbacks, they should be used with caution. In the future, more research should be conducted to determine how AI chatbots could be effectively integrated into dermatologic practice. Perhaps prior authorization letters and writing patient education materials could also be considered.
- Chen, R., Zhang, Y., Choi, S., Nguyen, D., & Levin, N. A. (2023). The Chatbots are Coming: Risks and Benefits of Consumer-Facing Artificial Intelligence in Clinical Dermatology. Journal of the American Academy of Dermatology. https://doi.org/10.1016/j.jaad.2023.05.088, https://www.sciencedirect.com/science/article/pii/S0190962223011647