Medical AI has come a long way. It started with basic machine learning, then moved to more complex deep learning. At first, it needed a lot of guidance, but now it can learn more independently. Recently, researchers have been trying to make medical AI that can handle many different tasks instead of just one specific job.
The paper talks about how we might move from these general medical AI systems to even more advanced ones. The authors introduce two new ideas:
- Universal Medical AI (UMAI): This would be AI that can do what doctors do, including showing empathy and using intuition – things that are hard for machines to copy.
- Universal Health AI (UHAI): This goes beyond just dealing with diseases. It would also help keep people healthy and prevent illness. It would use information from outside the doctor’s office to understand health in a more complete way.
The authors explain what needs to be researched to make these ideas a reality. They hope that in the future, AI will be better at working with patients, doctors, and society as a whole.
Full text (can be downloaded from the journal website, or requested from the authors through email or ResearchGate): Weizhi, M., Bin, S., Yang, L., Jing, Q., Xiaoxuan, L., Jingshan, L., David, O., Haibo, W., Atanas, G.A., Pearse, A.K., Wei-Ying, M., Yih-Chung, T., Tien Yin, W., 2024. Evolution of Future Medical AI Models — From Task-Specific, Disease-Centric to Universal Health. NEJM AI 1(8), 1-5, doi:10.1056/AIp2400289.
Keywords: Medical artificial intelligence, AI models, deep learning, supervised learning, unsupervised learning, task-specific models, generalized medical artificial intelligence, GMAI, universal medical artificial intelligence, UMAI, universal health artificial intelligence, UHAI, physician empathy, clinical practice, disease prevention, health maintenance, nonclinical data, health data integration, transformer techniques, multimodality data, data alignment, contrastive learning, medical tacit knowledge, heuristic-based approach, decision rules, data imputation, value alignment, nontraditional data sources, AI of Things, wearable devices, health ecosystem, foundational models, LLMs, smart healthcare devices, data collation, messy data integration, domain knowledge, reasoning, imitation learning, multi-agent collaboration, neuroAI techniques, human feedback reinforcement learning, chain-of-thought reasoning, tree-of-thought reasoning, pathophysiological status, sequential actions, multi-specialist collaboration, dynamic agent selection, tacit knowledge injection, AI techniques in healthcare.