For the past several years, hospitals have focused on deploying generative AI for documentation, summarization, and coding. Now, healthcare leaders from Northwell Health, Nvidia, and Expper Technologies say the industry's next major shift will move AI beyond the computer screen into physical three-dimensional space - systems that can perceive, reason, and act in the real world.
Physical AI combines robotics, sensors, computer vision, and autonomous systems capable of making decisions and taking actions in a hospital environment. Unlike a chatbot that generates a response, these systems could transport medications, monitor patients, assist surgeons, guide rehabilitation, and coordinate logistics while adapting to changing conditions around them.
But this evolution raises different questions than generative AI. Hospitals will need to address physical safety, human oversight, governance, cybersecurity, and liability when intelligent machines interact directly with patients and clinicians. All three experts interviewed agreed that widespread deployment remains years away, but the underlying technologies - multimodal AI models, simulation, edge computing, and robotics - are advancing rapidly.
From information to action
"For me, physical AI begins when software is no longer only generating an answer - it is generating an action in a shared physical environment," said Davit Martirosyan, co-founder and head of engineering at Expper Technologies. His company builds Robin, a socially assistive robot deployed in pediatric and senior care settings.
Dr. Filippo Filicori, system chief of surgical innovation at Northwell Health, described physical AI as intelligence "connected to sensors, machines and the physical environment" that can complete "a closed loop of perception, planning and action." He does not see generative and physical AI as competing. "Generative models can help physical systems understand instructions, generate training scenarios and reason about complex situations. World models add the ability to simulate what might happen next. Physical AI connects those capabilities to action."
David Niewolny, senior director of business development for healthcare and medtech at Nvidia, said hospitals should think of physical AI as the next wave. "Physical AI is not a chatbot with wheels," he said. "It requires models that understand the physical environment, simulation that teaches systems how the world behaves, and edge computing that can respond in real time."
Where hospitals will see value first
All three experts expect early success from narrowly defined, repetitive tasks - not from autonomous robot doctors. Martirosyan predicted the largest opportunity over the next five to ten years will come from "bounded, repetitive tasks" such as transporting supplies, monitoring equipment, supporting patient navigation, and handling routine nonclinical requests.
"Physical AI can reduce this operational burden and give nurses and other professionals more time for work that requires human expertise," he said.
Filicori agreed. "Over the next five years, logistics will probably scale fastest," he said. "Supply delivery, pharmacy workflows, specimen transport, sterile processing, room turnover and inventory management involve repetitive, measurable tasks in relatively structured environments." These applications carry lower clinical risk while offering measurable operational returns, making them attractive starting points for AI for Operations in healthcare.
Niewolny added that healthcare is one of the most important applications for physical AI because demand for services far exceeds available clinical workforce. Physical AI can help close that gap by extending access and letting clinicians focus on patient needs rather than tedious tasks.
Clinical care as the longer-term sweet spot
While logistics may arrive first, all three see deeper clinical impact ahead. Filicori pointed to surgery as a strong candidate because robotic platforms already generate structured data - high-resolution video, kinematic data, device telemetry. But he expects "autonomy by task, not autonomy by procedure." Rather than fully autonomous surgery, carefully bounded assistance will define the approach over the coming decade.
Niewolny noted that imaging assistance, patient monitoring, and procedural support are likely to join logistics as early deployed use cases as the technology matures.
Why this matters for Healthcare
Physical AI promises to shift artificial intelligence from a passive advisor to an active participant in care delivery. For hospitals and health systems, the readiness gap is the key takeaway: organizations that wait until the technology becomes commonplace may find themselves scrambling to build the infrastructure, governance, and operational expertise required for safe deployment. Those that begin now - starting with low-risk logistics tasks - will be better positioned to scale physical AI into clinical applications later. Understanding how to prepare today requires looking at both the operational foundations and the broader AI for Healthcare landscape, where the convergence of robotics, sensors, and multimodal models is accelerating faster than many expect.
Your membership also unlocks:
