Artificial intelligence is transforming medicine, but successful adoption requires more than powerful models. As healthcare organizations move from AI pilots to enterprise-wide implementation, governance, safety, and clinician readiness have become essential to realizing AI's full potential.
To explore the challenges of AI adoption, we spoke with Andrew Borkowski, former Chief AI Officer for the VA Sunshine Healthcare Network and a leader in clinical AI governance and education. With CHARGE, Borkowski detailed the governance programs he spearheaded at VA Sunshine, his thoughts on the post AI deployment monitoring gap, and explained why medical school curricula are increasingly archaic.
Q: Andrew, please tell us about yourself, what first led you to become interested in AI, and how did your career evolve toward the intersection of AI and medicine?
A: I have always been fascinated by technology, and that curiosity led me to become a pathologist, which allowed me to work with advanced diagnostic instruments and software every day. My colleagues and I even wrote code to improve these systems, giving me an early appreciation for how technology could enhance patient care. A defining moment came in 2012 when AlexNet revolutionized computer vision by winning the ImageNet competition. Inspired by that breakthrough, I began exploring whether deep neural networks could classify cancer on digital pathology slides. My team and I went on to develop AI models for lung, colon, and liver cancers, and the results convinced me that artificial intelligence would fundamentally change medicine. As AI evolved, I realized that creating powerful algorithms was only half the challenge. Hospitals also needed a safe and responsible way to evaluate and use these tools. That led me to help establish the first Hospital AI Committee in the VA Healthcare System and focus on AI governance, research, and education. I believe AI is one of the most transformative technologies in healthcare. Used responsibly, it will help us deliver more accurate diagnoses, better treatments, and higher quality care for our patients.
Q: During your role as Chief AI Officer at VA Sunshine, what AI governance challenges did you encounter within the VA system, and what work did you lead in response?
A: When I became the Chief AI Officer for the VA Sunshine Healthcare Network, I found that many departments had independently adopted AI tools without any central oversight. Working with the VA National AI Institute, we conducted a system wide inventory of every AI application in use, creating a comprehensive catalog to understand what technologies were being used across the network. We then developed clear and practical policies to guide the safe use of AI and shared them with local AI governance committees. In addition, we created policies for the implementation, evaluation, and ongoing monitoring of medical AI systems. To help healthcare organizations build effective governance programs, we developed a blueprint for establishing a hospital AI committee. We also published six principles of Trustworthy AI, which closely align with the Elements of Responsible Use of AI in Healthcare that were later published by the Coalition for Health AI.
Q: How should healthcare organizations evaluate AI tools before clinical deployment to ensure they are safe and effective for real-world use?
A: To safely use artificial intelligence in patient care, healthcare organizations need a clear plan led by clinical experts and supported by a dedicated oversight team, like for example local AI Committee. Before an AI tool is introduced, it should be tested using the organization's own patient data to make sure it is accurate and works fairly for people from all backgrounds. It is also important to make sure the technology fits naturally into everyday clinical workflows so it helps doctors and nurses instead of creating extra work or unnecessary alerts. Once the AI system is in use, its performance should be monitored regularly to identify any decline in accuracy over time. Throughout the process, healthcare professionals should always make the final clinical decisions, using AI as a tool to support, not replace, their medical judgment.
Q: Where do you currently identify the most significant governance gaps in healthcare AI?
A: One of the biggest challenges in healthcare AI is what happens after an AI system has been approved and put into everyday use. Current regulations focus mainly on testing AI before it is deployed, but there are no widely accepted systems for continuously monitoring how it performs over time. As a result, many AI models are left unchanged even as patient populations, clinical practices, and hospital technology evolve. This can lead to a problem called model drift, where an AI system gradually becomes less accurate without anyone realizing it. At the same time, there is no consistent national approach to overseeing healthcare AI. Instead, individual states have created their own rules, leading to a patchwork of regulations. Large academic medical centers often have the resources to build AI oversight programs that monitor these systems and adapt to changing requirements. In contrast, many community and rural hospitals lack the staff, funding, and tools needed to evaluate their AI software, leaving them at greater risk of legal issues and potential harm to patients.
Q: Following your transition to Director of AI Curriculum at USF Morsani College of Medicine, what core AI competencies are essential for future physicians, and how should medical education adapt to prepare them?
A: I decided to transition from serving as a Chief AI Officer to becoming the Director of AI Curriculum at the USF Morsani College of Medicine because I believe the next generation of physicians must be ready for a future where AI is part of everyday healthcare. Right now, most medical schools do not adequately prepare students for that reality. Medical education needs to evolve. Every future physician should understand the basics of AI, including how these systems work, what they can and cannot do, and how to recognize when they may be wrong. They should also learn the fundamentals of data science and understand the importance of protecting patient privacy, reducing bias, and using AI in an ethical and responsible way. Learning about AI should not be limited to lectures. Medical students and residents should have opportunities to use AI tools in realistic clinical simulations, learn how to evaluate new technologies, and demonstrate that they can use them safely in patient care. My goal is to help train physicians who see AI as a powerful tool that supports, rather than replaces, their clinical judgment.
Q: Five years from now, how do you anticipate healthcare AI will evolve, and what technological advances and governance frameworks will most shape its development and adoption?
A: Five years from now, artificial intelligence will likely become a natural part of healthcare, working quietly in the background to support both clinicians and patients. AI will help reduce administrative work by automatically documenting patient visits, giving doctors more time to focus on patient care. It may also use digital twins, which are virtual models of individual patients, to predict how someone will respond to a treatment before it is prescribed. AI will play an important role in home care by continuously monitoring patients and sending early alerts when medical attention is needed, helping people stay healthier and avoid unnecessary hospital visits. At the same time, more advanced AI systems will combine medical images, clinical records, and genetic information to detect diseases earlier and more accurately than is possible today. To make these advances safe and widely accepted, healthcare organizations will need strong governance with clear standards for privacy, fairness, transparency, and ongoing oversight. I am optimistic that, with the right safeguards in place, AI will make healthcare better, safer, and more accessible for everyone.