How AI and Digital Twins Are Enabling Human–Machine Co-Regulation in Diabetes Care
Dr. Boris Kovatchev
March 30, 2026 – In the keynote lecture opening the international conference Advanced Technologies and Treatments for Diabetes (ATTD) in Barcelona, Spain, Dr. Boris Kovatchev, proposed that there is a necessary fundamental transition in diabetes management from metabolic models to data-driven AI.
Artificial intelligence (AI) is often described in sweeping, futuristic terms as machines that think, learn, and even replace human decision-making. But in diabetes care, the most promising role for AI may be far more collaborative: not replacing humans but working alongside them.
“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.”
At its core, AI is built on systems like neural networks, which are models that continuously learn by processing data, making predictions, and correcting errors. These systems power everything from image recognition to language processing. In healthcare, they are increasingly being applied to diagnose diseases, predict outcomes, and guide treatment decisions.
Since 2018, nearly 5,000 research papers have explored applications of AI in diabetes. Much of this work has focused on diagnosis, detecting complications like retinopathy, optimizing decision support, and enhancing automated insulin delivery (AID) systems. Yet most of these applications are designed for clinicians, not for meaningful interaction with the person living with diabetes.
“We’ve built powerful tools for clinicians,” Kovatchev explained. “But we haven’t fully explored how AI can engage the person with diabetes as an active partner in their care.”
This gap has led to a new concept: human–machine co-regulation.
“Diabetes management is not a one-time decision, it’s a continuous process,” said Kovatchev. “That makes it the perfect setting for human–machine collaboration.”
Rather than replacing human decision-making, co-regulation creates a feedback loop, where AI and the individual continuously learn from and adapt to one another.
- The machine learns from data and suggests optimized strategies
- The human interacts, experiments, and makes informed decisions
Both continuously adapt to each other. This idea has already been tested in a clinical trial – see Nature Digital Medicine https://rdcu.be/ekVgP.
The Role of Digital Twins
At the University of Virginia Center for Diabetes Technology, researchers are advancing this concept through digital twins, virtual models that replicate an individual’s glucose-insulin dynamics and behavioral patterns. These digital twins allow individuals to test decisions in a risk-free, simulated environment before applying them in real life.
The paper noted above reports a six-month clinical trial known as Adaptive Biobehavioral Control (ABC), involving 72 individuals with type 1 diabetes using automated insulin delivery systems. Participants could interact with their digital twin to simulate treatment decisions, while the system simultaneously adapted recommendations based on each participant’s data.
“A digital twin gives you the ability to ask, ‘What if?’, without consequences,” Kovatchev said. “It turns uncertainty into something you can explore and understand.”
The results were very positive:
- Time in the target glucose range (70-180 mg/dL) improved from 72% to 77%
- Reduced HbA1c from 6.8% to 6.6%
This approach represents a shift from algorithm-only optimization to algorithm-to-human adaptation, where individuals actively engage with AI to better understand and manage their condition.
Why Data Is the Key
The future of this approach depends on data, and lots of it.
Today, more than 1.4 million people use continuous glucose monitors and automated insulin delivery systems, generating data every five minutes. This constant stream creates the foundation needed to train advanced AI systems and build accurate digital twins at scale.
To make human-machine co-regulation widely available, several steps are essential:
- Foundational models of glucose–insulin–behavior interactions
- Continuous data integration from connected devices
- Build individualized digital twins
- Optimize treatments in simulation before real-world use
- Enable users to interact with and learn from their digital twin
A Necessary Shift in AI Thinking
To fully realize human–machine co-regulation, the field must shift from traditional model-based systems to data-driven AI that can continuously learn and adapt. This shift mirrors broader trends in AI development led by companies like Google and Meta, where large-scale foundation models are transforming how machines understand and interact with the world.
“Static models can only take us so far,” Kovatchev said. “The future belongs to systems that learn continuously, from data, and from the person.”
The Future: AI as a True Partner
The concept of human-machine co-regulation is no longer theoretical. Early studies show it is both feasible and effective. Looking ahead, the goal is clear: an AI-powered digital twin for every person living with diabetes.
Such a system would not only improve clinical outcomes but also empower individuals, helping them understand their condition, test decisions safely, and build confidence in their care. The ultimate outcome may be fully automated, personalized closed-loop care. But the real breakthrough lies in something deeper: A future where humans and intelligent systems don’t compete but collaborate.
Support
The research was presented as the keynote at the 2026 ATTD Conference in Barcelona.
The ABC study was published in Nature Digital Medicine – https://rdcu.be/ekVgP.
The concepts and clinical trials used in this presentation were supported by NIH/NIDDK grants RO1 DK 085623 and RO1 DK 133148.