The Boring Revolution: Why Healthcare AI Must Earn Trust Before It Changes Medicine
Author:
In 2015, as the FDA’s first Chief Health Informatics Officer, I helped build precisionFDA, a secure platform where 5,000 researchers validated genomic models. The real lesson wasn’t about code or scale – it was about trust. In healthcare, the most transformative technologies don’t feel radical. They feel boringly reliable.
Today at GE HealthCare, we see AI's evolution unfolding across three horizons, each requiring different strategies, metrics, and trust models. In an age where conferences showcase a plethora of innovative AI demos and where vendors promise significant change, I've learned to focus on what actually ships, scales, and saves lives.
Short-Term: Scale What Works
I'd argue the immediate opportunity isn't invention but implementation. Healthcare organizations need to move from AI pilots to production, and that requires three foundational capabilities.
First comes AI sovereignty, ownership of the full AI lifecycle, from training to inference. Without it, healthcare providers risk black boxes and vendor lock-in. With it, they keep control. To give just one example, during my time at GE HealthCare, I've worked with health systems deploying pneumothorax detection directly securely on critical care devices. These health systems maintained complete institutional control, helped reduce reliance on black boxes, and mitigated vendor lock-in risks.
Second, scaling requires operational intelligence that tackles what seems mundane: predicting bed availability, optimizing staff schedules, tracking supply levels, and streamlining documentation. At Queens Health System, for example, our Command Center generated $20 million in first-year savings through optimized patient flow and increased transfers. At Children's Mercy, similar technology resulted in an 86% reduction in delays, and 87% reduction in canceled surgeries. In obstetric ultrasound, solutions like our Voluson SonoLyst reduce keystrokes by 65%. When you multiply these efficiencies across departments and facilities, operational intelligence becomes the foundation for sustainable scale.
Third, it's time to scale proven models. For example, remote scanning solutions like nCommand Lite, which enables radiologists to guide examinations from a distance, have facilitated over 64,000 examinations with a 74% reduction in rescans. By allowing expert sonographers to remotely assist less experienced operators or work night shifts from home, these systems extend expertise where it's needed most. These aren’t experiments anymore. They’re workflow-integrated, with documented results … delivering measurable impact today.
Medium-Term: From Diagnostics to Intervention
The next wave seeks to transform imaging from taking medical images to guiding action. We're already seeing glimpses of this future: breast ultrasound systems like Invenia ABUS achieving 93% sensitivity for lesion detection while reducing reading time by 33% with QVCADTM. In radiation therapy, certain institutions have utilized planning solutions such as iRT and InstaPlan to reduce treatment preparation from days to minutes. But these examples only hint at the broader transformation ahead.
We think the real shift will be screening at population scale. In pilot programs using portable ultrasound devices like Vscan Air with Caption AI guidance, over 50% of patients showed abnormalities consistent with Stage B heart failure. Early detection at this scale could significantly influence how we approach disease, moving from reactive treatment to proactive prevention.
The challenge is stark: many clinical risk models use tens of data points, while a single patient record holds data points that number in the hundreds of thousands. Why this massive gap? The technical barriers of interoperability and standardization are real, but the deeper challenge is trust. Clinicians rightfully demand transparency. A model analyzing 300,000 data points might be statistically stronger, but if it can't explain its reasoning in clinical terms that connect to established medical knowledge, it remains clinically useless.
Long-Term: Autonomous Intelligence
Over the longer term, I see physical AI bring autonomous capabilities to clinical settings. Picture X-ray systems that position themselves, optimize protocols, and ensure quality without human intervention, extending advanced imaging to underserved areas where technologists are scarce. This technology could help extend access to billions of people worldwide who currently have little or no access to quality medical care.
Tomorrow’s hospitals won’t rely on single point solutions. Instead, multi-agent systems will act like digital shift supervisors, coordinating scheduling, documentation, supply chain, and patient flow, all with human oversight and clear audit trails. Streaming device intelligence will replace our current episodic snapshots with continuous monitoring. Every scanner, monitor, and bedside device becomes an intelligent endpoint, creating living digital twins that evolve with the patient.
The future belongs to these converging capabilities that will fundamentally redefine care delivery.
The Trust Imperative
After decades in healthcare, from the cath lab to the FDA to big tech, I've learned that transformation happens not through revolution but through methodical trust-building. Every interventional cardiologist knows you measure success not in the elegance of your technique but in the patient who walks out healthier than they walked in.
The path forward requires embracing an uncomfortable truth: we'll know healthcare AI has truly succeeded when it becomes boring and routine. When clinician adoption becomes second nature because AI is seamlessly woven into their daily practice, and when patients expect that they will have the benefit of AI-enhanced care across their entire patient journey, that’s when achieved real transformation.
The solutions most likely to succeed won’t be the showiest demos but the quietest systems, the ones that free up a nurse’s time, cut a radiologist’s clicks, and make workflows invisible. In medicine, trust scales slower than technology but lasts longer. That’s why the future of AI isn’t revolutionary. It’s boring. And that’s exactly what will make it matter. Healthcare's AI transformation won't be won by the flashiest demos or the boldest promises. It will be won by those who understand that in medicine, trust scales slower than technology but lasts longer. Make it boring. Make it dependable. Make it matter.
1. https://www.gehccommandcenter.com/2025-outcomes-source-data
2. GE HealthCare Voluson internal claims document JB20479XX / DO C2727504.
3. Results listed here are of these specific customers and may not be typical. GE HealthCare cannot guarantee these or similar results. Comparison over the same time period for exams performed remotely and following the standard onsite workflow. The DASA site solution configuration includes some features that are not available in the US, namely the ability to remotely move the patient table and remotely communicate with the patient. Results may vary. The statements by GE HealthCare customers are based on their opinions and results achieved in their unique setting. Since there is no “typical” facility and many variables exist (i.e., size, case mix, etc.) there can be no guarantee that other customers will achieve the same results.
Keep exploring for FREE!
Create a free account or log in to unlock content, event past recordings and more!