I am a pediatric cardiologist and have cared for children with heart disease for the past three decades. In addition, I have an educational background in business and finance as well as healthcare administration and global health – I gained a Masters Degree in Public Health from UCLA and taught Global Health there after I completed the program.
“The future is already here- it is just not evenly distributed.”
William Gibson, Canadian science fiction novelist
Time to recount my thoughts on the future of AI in clinical medicine and healthcare. For ease of digestion, allow me to deconstruct the larger futuristic vision of AI in medicine into three weekly parts: technology, people, and the dyad of machine and human.
There is a trend towards faster pace of development of both the AI technology such as natural language processing (GPT-3 and other transformer language models) as well as other emerging technologies such as quantum computing that will have a strong impact on AI.
First, the area of edge AI and its AI applications in devices will impact on internet of things (IoT) and Internet of everything (IoE) for chronic disease management with some of the care displaced and improved in the home. This migration of AI locally into the periphery is coupled with continued cloud AI capabilities in the cloud but also at the edge (edge cloud).
In addition, healthcare data in and of itself needs to improve in its AI potential, and one possible innovative strategy is to transition from traditional relational databases to a graph and even hypergraph format. While an ordinary graph has an edge that connects to two vertices, a hypergraph has an edge that connects to any number of vertices and can represent complex relationships in biomedicine even better. This transition will potentially enable better yield of information and knowledge from machine and deep learning so that we can reach both precision medicine and population health paradigms that depend on multiple layers of data and information to interrelate.
In terms of technologies that can neutralize the paucity of labeled data limitation, generative adversarial networks (GANs) and few shots learning are methodologies that can be useful to provide more data or make better use of little data, respectively.
Learning in all forms will become increasingly more sophisticated in the future. Transfer learning, deep reinforcement learning, self supervised learning, and predictive learning (unsupervised learning that models the world) will all contribute to the effort in clinical medicine and healthcare.
More consistent and widespread sharing of data in the future can also accelerate federated learning to enable real time actionable intelligence for clinicians that was so lacking during the pandemic. In federated learning, the machine learning models are brought to the data source (rather than the other way around) and thus a central server pools the model results from several sources and generate a global one without accessing any data.
Lastly, AI can be coupled to the extended reality domain to render it more intelligent and interactive for medical education and clinical training; this will then be “intelligent” reality.
Part II of this series on the future of AI in medicine will be explored next week.