Accuracy and Bias in Artificial Intelligence Chatbot Recommendations for Oculoplastic Surgeons

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Informatics

Background:

• Artificial intelligence (AI) chatbots are increasingly used for information retrieval.
• Patients may use AI chatbots to find medical specialists, including oculoplastic surgeons.
• The accuracy and potential biases of AI recommendations in healthcare are not well understood.

Purpose of the Study:

• To evaluate the accuracy and identify biases in oculoplastic surgeon recommendations from three major AI chatbot systems.
• To compare the performance of ChatGPT, Microsoft Bing Balanced, and Google Bard in providing relevant surgical referrals.

Main Methods:

• Three AI chatbot systems (ChatGPT, Microsoft Bing Balanced, Google Bard) were queried.
• Prompts included "oculoplastic surgeon," "doctor who does eyelid lifts," and "oculofacial plastic surgeon."
• Recommendations were analyzed for accuracy, specialist training, and demographic representation across 20 major US cities.

Main Results:

• High inaccuracy rates were observed, with up to 38% of recommended surgeons being nonexistent or not practicing in the specified city.
• 35.9% of recommended "oculoplastic" surgeons lacked specialist training in the field.
• Prompt wording significantly influenced results, with "eyelid lifts" yielding more non-specialist referrals.

Conclusions:

• Current AI chatbots exhibit substantial inaccuracies and biases when recommending oculoplastic surgeons.
• The use of specific prompts can lead to a higher proportion of unqualified or irrelevant referrals.
• Patients seeking oculoplastic surgeons via AI chatbots face risks due to unreliable recommendations, necessitating critical evaluation of AI-generated medical advice.