Promising but proceed with caution, according to new study

Recent Journal of the American Academy of Dermatology A study evaluated the risks and benefits of consumer artificial intelligence (AI) in clinical dermatology.

Research: Chatbots are Coming: Risks and Benefits of Consumer-Facing Artificial Intelligence in Clinical Dermatology. Image credit: greenbutterfly / Shutterstock

AI chatbots in dermatology

AI has made remarkable progress recently and is close to being introduced into dermatology. As an example, AI-powered image-based classifiers have been deployed in skin cancer detection, often outperforming dermatologists.

A new form of artificial intelligence is the AI ​​chatbot. These are open to the public and simulate a conversation with a human user. When a question is asked to the platform, the chatbot uses natural language understanding to determine what the user wants to know. Answers are then generated by a neural network trained on a large text dataset. Words are then selected based on probability distributions. Chatbots have the ability to learn from repeated interactions and improve over time. There are many models currently in development and his ChatGPT at OpenAI is one of them.

Advantages of chatbots

Translating medical jargon into patient-friendly language is one of the key uses of chatbots. Recent studies have highlighted the effectiveness of his ChatGPT in simplifying radiology reports, with several radiologists verifying that translations are accurate and complete. During the 2019 coronavirus disease (COVID-19) pandemic, Penn Her Medicine used simpler chatbots to answer patient questions when call centers were overcrowded.

Wait times for dermatology appointments are currently very long and patients often use online resources for self-diagnosis. Whether such online tools provide better information than existing resources is not fully established, and it is entirely possible that incorrect medical advice or misdiagnosis could result in potential harm. The authors of this study tested his ChatGPT against various scenarios generated from a list of the 25 most frequently asked skin diseases on DermNetNZ.

We observed that ChatGPT produced a reasonable differential diagnosis for ‘red itchy spots’, rosacea in fair-skinned individuals, and black lines on nails. They also suggested treatment and recommended seeing a doctor for further evaluation. ChatGPT also answered questions about skin cancer risk, wound care, hair care, sunscreen, and acne. These results prove that chatbots can provide reliable information even in situations where patients may not be able to see a doctor on short notice.

Chatbot limits

An important limitation in getting advice from chatbots is that patients must describe their skin morphology very precisely. This is because chatbots rely entirely on user input. For example, ChatGPT diagnosed herpes labialis and stye when an infant presented with a ‘growing red bump’ on the lip or eyelid, respectively. However, the diagnosis of hemangioma was not considered. It was suggested in ChatGPT when hemangiomas were presented as ‘growing purple bumps’.

Another significant limitation is the very limited knowledge of chatbots. They do not know about developments in the medical field after the date of their training. For example, ChatGPT incorrectly stated that baricitinib is not FDA-approved for alopecia areata because training data for baricitinib is limited through 2021. Racially incomprehensible training data can also make it difficult for chatbots to accommodate patients with different skin types. ChatGPT overlooked rosacea as a plausible diagnosis when diagnosing dark-skinned patients.

Conclusion

In summary, chatbots have proven promising in answering patient questions, but should be used with caution given current shortcomings. In the future, further research needs to be done to determine how AI chatbots can be effectively integrated into dermatology practice. Possibly, the development of prior authorization forms and patient education materials could also be considered.