Human-Centered AI: Large Language Models and the Need for Ethical Medical Chatbots (Preprint)

Abstract

BACKGROUND The evolution of medical chatbots underscores the pressing need for human-centered AI to address patient and family concerns more personally and precisely. This paper explores the potential ways to do so, through tracing the historical development of artificial intelligence (AI), charting the progression from neural networks to the prevalent use of Large Language Models (LLMs) like GPT-3 and 4 within the realm of human-centered AI. OBJECTIVE This review critically examines the pivotal role of LLM-based medical chatbots in fostering ethical and humanistic AI in the construction of medical chatbots. Specifically, it investigates the transition from explicit algorithms to the contemporary landscape in which LLMs emulate human speech, language, and behavior. METHODS Beginning with examining the elements of human-centred AI, pathways to build medical chatbots ethically were identified. The loopholes for potential bias in LLMs were highlighted, such as their reliance on extensive datasets and neural networks, which have revolutionized programming methodologies, whereas bias in the datasets also may result in bias in the medical chatbot’s responses. The training methodology employed in developing ChatGPT was also reviewed, finding potential bias because of the bias in the massive training data. RESULTS The study's findings centered on the potential ways to design medical chatbots with human-centered AI. It delves not only into how LLMs, particularly in their proficiency in understanding and generating human language, can significantly shape the future of AI, but also how inherent bias in the training of the LLMs may result in chatbots not for everybody. The focus extends to broader implications for AI and computer science, shedding light on transformative opportunities and ethical challenges inherent in the application of LLMs. CONCLUSIONS This review highlights the transformative potential of LLMs in shaping the next generation of medical chatbots. It underscores the imperative of infusing human-centric values and ethical considerations into AI systems, aligning with the overarching goal of creating human-centered AI. The insights presented include considerations for mitigating bias in ChatGPT and reflections on the future trajectory of LLMs, emphasizing practical applications in the development and enhancement of ethical medical chatbots. CLINICALTRIAL NA