AC LITERATURE

Dr. Anthony Chang

“The scaling law continues. The internet is producing data twice the year before. The next two years humanity will produce more data than the entire history of mankind. In the future, every factory will have a digital twin.” 

Jensen Huang, CEO of Nvidia at Consumer Electronics Show, 2025 

 

We discussed the dimension of people in AI in healthcare for 2025 and coming years last week, this week we will delve into some of my thoughts about another dimension in AI adoption and implementation: the technology itself: 

The AI Ensemble of Tools

The growing assemblage of multimodal and generative AI tools has evolved into an impressive AI in health ensemble of tools (ensemble defines as a collection that is more cohesive with higher function and with more sophistication and synergy vs an assemblage), but we still need to have impact. Medical images are now interpreted frequently with deep learning tools (convolutional neural networks) and even moving images such as echocardiograms can be analyzed with deep learning. With large language models such as ChatGPT o1 pro and Open Evidence, I am able to see patients with not only a seasoned clinician knowledge, but an impressive panoply of AI tools to help guide me in diagnosis and treatment of patients during clinic. Recent Nobel Prizes are given to AI pioneers who are exploring ways to design proteins, drugs, and vaccines with aim to improve outcomes in diseases. The eventual precision medicine strategy will need to involve digital twins being coupled to deep reinforcement learning.  

 

Data and Information Infrastructure

A major obstacle to more AI adoption and effectiveness, however, is rooted in data and information technology infrastructure. It is not uncommon for a health system not to have a robust data governance structure not an organized approach to artificial intelligence. Most health systems also lack expertise in enabling an IT infrastructure for AI projects (or applying AI for improved IT functionality). In addition, many hospitals and health systems (and most clinics and group practices) do not have personnel who understand and optimize the technology of AI and its support structures (data pipelines and databases as well as IT infrastructure).  

 

AI and Workflow

Perhaps a very important aspect of AI in healthcare is to enable this resource in the entire workflow (both clinical and non-clinical dimensions). For example, AI can be used in radiology or cardiology not only in image interpretation (perhaps the most obvious AI application), but also the entire workflow: patient selection, procedure authorization, report configuration, billing procedure, image interpretation, and information dissemination. In the future, the shortage of certain image acquisition technicians can be partly reconciled by AI-enabled robotic tools that can replace human operators who are highly skilled (and therefore at times expensive and unavailable).