Evolution of Future Medical AI Models — From Task-Specific, Disease-Centric to Universal Health
Weizhi Ma, Bin Sheng, Yang Liu, Qian Jing, Xiaoxuan Liu, Jingshan Li, David Ouyang, Haibo Wang, Atanas G. Atanasov, Pearse A. Keane, Wei‐Ying Ma, Yih Chung Tham, Tien Yin Wong
Abstract
Medical artificial intelligence (MAI) has evolved from traditional machine learning to deep learning and from supervised methodologies to unsupervised learning paradigms. Recently, the focus has shifted from task-specific to generalized medical artificial intelligence (GMAI) models. These new artificial intelligence (AI) models and algorithms still need to be translated to clinical use in various settings. This article discusses the foreseeable transition from specialized MAI models toward more universally applicable models. We introduce two concepts as new paradigms: universal medical artificial intelligence (UMAI) and universal health artificial intelligence (UHAI). UMAI models will be distinguished from GMAI by their capability to emulate critical aspects of human intelligence necessary in clinical practice, particularly physician empathy and intuition. UHAI further expands beyond addressing disease states, a domain of UMAI, and covers health maintenance and disease prevention, shifting from relying solely on traditional clinical data to integrating broader nonclinical data to allow for the incorporation of AI into a more holistic understanding of human health and disease origin. Outlined here are key research priorities and future pathways from GMAI to UMAI and subsequently, UHAI, allowing AI to be more integrated, intuitive, and attuned to the needs of patients, physicians, and society.