Boris Kovatchev, PhD
In the keynote lecture opening the international conference Advanced Technologies and Treatments for Diabetes (ATTD) in Barcelona, Boris Kovatchev, PhD, a School of Medicine professor, adjunct professor at the School of Engineering and Applied Science, and the founding director of the UVA Center for Diabetes Technology, proposed that there is a necessary fundamental transition in diabetes management from metabolic models to data-driven AI.
Artificial intelligence (AI) is often framed as a tool for automation, but in diabetes care, its greatest potential lies in collaboration with the person living with the condition. While AI systems—built on neural networks that learn through continuous data processing and error correction—have driven advances in diagnosis, complication detection, and automated insulin delivery, most applications remain clinician-focused or fully automated, with limited direct engagement from patients. This gap highlights the need for a new paradigm: human–machine co-regulation, in which individuals and AI systems continuously interact, learn from one another, and adapt in real time.
“The question is no longer whether AI can make decisions,” said Kovatchev. “It’s whether AI can work with people to improve those decisions in real time.”
Research at the University of Virginia Center for Diabetes Technology demonstrates how this concept can be operationalized through digital twins—virtual models that replicate an individual’s glucose–insulin dynamics and behavioral patterns. In a six-month clinical trial of Adaptive Biobehavioral Control (ABC), 72 individuals with type 1 diabetes used automated insulin delivery systems enhanced by personalized digital twins, enabling them to explore “what-if” scenarios and simulate treatment decisions before applying them in real life. This human–AI interaction improved outcomes, increasing time in range from 72% to 77% and reducing HbA1c by up to 1.2% among participants with higher baseline levels. These findings suggest that integrating behavioral engagement with algorithmic adaptation can enhance both clinical outcomes and patient understanding. The scalability of this approach is supported by the rapid growth of diabetes data, with more than 1.4 million individuals generating continuous glucose and insulin delivery data every five minutes. Realizing the full potential of human–machine co-regulation will require a fundamental shift from traditional model-based approaches to data-driven AI systems capable of continuous learning and personalization, similar to foundation models emerging across the broader AI landscape.
The long-term vision is an AI-powered digital twin for every person with diabetes, enabling safe experimentation, improved adherence, and more confident self-management. Ultimately, this collaborative framework may lead to fully personalized, closed-loop systems, but its most immediate impact is empowering individuals to better understand and manage their condition through active partnership with intelligent systems.
Filed Under: Research