Physical AI reshapes bedside care.
Physical AI is emerging as a new category of technology that could reshape how hospitals deliver care, according to senior leaders from Northwell Health, NVIDIA and the startup Expper who gathered for a recent industry discussion.
What physical AI means for the bedside
Unlike purely software‑based models such as generative AI, physical AI combines perception, reasoning and actuation to interact directly with patients and equipment. The concept envisions robots or smart devices that can move, sense vitals, and make real‑time adjustments without human intervention.
These examples illustrate a shift from AI that merely suggests actions on a screen to AI that can physically carry them out, potentially reducing manual steps and freeing clinicians for higher‑level tasks.
Preparing hospital infrastructure for physical AI
Adopting such technology requires more than buying new robots. Executives highlighted the need for robust data governance, comparable to treating data as a financial asset. Robert Slepin, chief digital officer at SE Health, warned that “without disciplined data pipelines, the AI’s decisions could be unreliable, and patient safety would suffer.”
Related: House Set to Vote on Health Subsidies
Hospitals must also establish clear accountability frameworks. While many leaders accept that model performance alone will not drive adoption, they stress that clinician trust hinges on transparent validation processes.
Training programs are already being rolled out. Sentara Health reported completing more than 60,000 AI‑literacy modules across clinical and operational staff, aiming to build a common foundation for responsible use. The effort combined concise lessons with governance education, ensuring that staff understand both the capabilities and limits of the technology.
In a similar vein, Nebraska Methodist disclosed a $2 million boost from AI‑assisted coding. The system does not replace coders but supplies a “consistent second set of eyes,” flagging potential discrepancies and offering supporting documentation, according to the CFO overseeing the human‑in‑the‑loop workflow.
These initiatives suggest that hospitals are moving beyond experimentation toward systematic integration, though the path remains uneven. Smaller facilities often lack the resources to implement the same level of data rigor, creating a disparity that could widen as physical AI matures.
Comparing the current rollout to earlier waves of electronic health record adoption reveals a familiar pattern: early enthusiasm, followed by a period of practical adjustments, and eventually a steady state where the technology becomes part of routine care. The key difference this time is the tangible interaction with patients, which brings safety and regulatory concerns to the forefront.
Related: Eli Lilly Cuts Drug Discounts for Clinics
Early outcomes and remaining challenges
Early pilots have shown promising efficiency gains. At CHOC Children’s Hospital, clinicians reported that generative AI tools reduced documentation time dramatically and cut after‑hours note‑writing by a third, easing burnout. Meanwhile, Sentara’s AI‑driven workflow tool saved $1.7 million in six months by catching missed radiology follow‑ups and ensuring timely patient outreach.
On the oncology front, UCSF’s integrated AI platform helped oncologists focus on personalized treatment rather than data retrieval, though the initiative highlighted the need for precise data governance to avoid misclassification of cancer subtypes.
However, not all applications have been seamless. A dermatology practice reported that a generic AI scribe frequently inserted incorrect procedure codes, prompting a switch to a specialty‑tailored solution that captures detailed clinical justification automatically. This illustrates the importance of domain‑specific tuning for physical AI systems.
Risk management also remains a priority. Dr. Jay Anders, a technology chief medical officer, warned that even an 80 % accuracy rate means the AI is wrong 20 % of the time, a margin that could be unacceptable in high‑stakes environments. He advocated for layered safeguards, including human review and continuous performance monitoring.
Overall, the move toward physical AI reflects a broader industry trend of embedding intelligence into the care environment. While the technology promises to streamline workflows, improve data capture and reduce clinician workload, its success will depend on disciplined data practices, clear accountability and ongoing validation.