Editorial Comment: Limitations of ChatGPT for Patient Education Regarding Frequently Searched Questions About Benign Prostatic Hyperplasia

Abstract

Language-based artificial intelligence (AI) platforms, such as GPT-4, published by OpenAI are becoming increasingly popular in everyday use. There is a trend of patients replacing clinicians or traditional search engines with AI programs for questions about their medical care. The crux of the controversy regarding the use of AI in medicine lies with its role, and degree of authority, in patient care. A clinician entrusted with the care of patients in any medical specialty is expected to have proven a competent standard of care,1 evidence-based medicine informed by cornerstone ethical principles with adequate practical training and experience to a documented minimum standard by examination. Although AI is performing well in some studies, including when challenged by examinations, GPT-4 at its current standard cannot provide consistent, validated, and referenced information to the degree required for accountable clinical practice. This study has reported 70% of GPT-4 responses to be accurate when compared by experts with standard, guideline-based practice.2 In a clinical governance-ruled discipline that is constantly striving for excellence, such performance is substandard. Furthermore, when considering individualized patient care, detailed and specific knowledge is required to discuss treatment options in fully informed shared decision making. For example, when considering benign prostatic hyperplasia treatment such as this study, increased risks are associated with comorbidities or medications, prostatic characteristics (size, median lobe), concurrent problems such as PSA elevation requiring investigation, and availability of local services or procedures. AI has already proven to be a valuable tool in the hands of clinicians, in everything from image analysis in radiology to white-cell differential counts in pathology to big data analysis in epidemiology.3 In the future, it is likely to play a large and increasing role in relieving the strain of an overburdened system—especially in patient communication, documentation, and assessment tasks including risk stratification, prognosis, and outcomes calculation.4 However, its accuracy and accountability with appropriate oversight must first be demonstrated before it can be adopted into these roles.

Ähnliche Arbeiten