AI could increasingly deliver clinical consultations directly to patients, potentially easing healthcare workforce shortages, but experts disagree over how far the technology should replace interactions with doctors.
Speaking at The Economist’s Second Annual AI in Health Summit, healthcare leaders suggested that AI could move beyond administrative support to deliver patient-facing services independently, with doctors overseeing multiple AI systems.
“The real value is in the higher-risk areas. That’s the place where we could create the highest value for the system, the highest value for the patients,” said Avi Tsur, MD, chief medical officer of the AI Center at Sheba Medical Center in Ramat Gan, Israel. He spoke during a panel discussion and highlighted the World Health Organization's projected global shortfall of approximately 11 million healthcare workers by 2030.
AI Could Deliver Care Directly to Patients
Tsur argued that patient-facing AI could offer greater benefits than technologies designed primarily to improve clinicians’ efficiency.
“At Sheba, a 2000-bed medical center, our AI-native clinical platform includes ambient scribing, summarization of admission and discharge letters, and even the injection of de-identified context from the electronic health record into ChatGPT for Healthcare,” he said. “Deployment will reach all our beds by the end of 2027, and possibly even earlier.”
Tsur predicted that AI-supported clinical reasoning would become routine across the campus in 2027, paving the way for patient-facing AI in 2028.
“As generative AI advances, 20% to 30% of clinical interactions could be replaced by patient-facing AI operating with conditional autonomy,” he said, suggesting that this approach could directly address workforce shortages.
Tsur also suggested that patient-facing AI could provide more accessible, potentially more empathetic care. In addition, it might eventually deliver better care in 25% to 30% of cases, although this is a prediction rather than an established outcome. “That will really bring the relief to the healthcare system,” he said.
Tsur also challenged the emphasis on low-risk applications, arguing that their benefits were correspondingly limited. “The lowest-hanging fruit is also the lowest value,” he said.
Describing a progression from appointment scheduling and ambient AI documentation through clinical decision support and patient-facing AI to physical AI agents, he argued that the potential benefits increased alongside the risks. “The higher you climb, the value is higher,” he said, emphasizing that higher-risk applications must be introduced safely under clinical investigation.
Will AI Replace Doctors?
Eric Cioè-Peña, MD, vice president of the Center for Global Health at Northwell Health in New York, challenged the suggestion that AI would replace doctors, arguing that global workforce shortages make displacement unlikely. “We don’t have an area where there’s a surplus of healthcare workers,” he said. “The idea of AI replacing physicians, to me, seems absurd. It’s going to enhance physicians. It’s going to enhance our ability to deliver high-quality care.”
He argued that AI could help healthcare systems meet unmet demand rather than displace existing clinicians, while maintaining that human interaction would remain essential to patient care. “The physician and the nurse have a human element to healthcare that the public’s certainly not going to be ready to let go of,” he said.
Luke Allen, MBChB, PhD, codirector of global primary care and future health systems at the University of Oxford, UK, described a more radical transformation, distinguishing between AI supporting individual clinicians and autonomous systems delivering consultations at scale.
He suggested that healthcare systems could move beyond the AI “copilot” model, where technology assists doctors during consultations, to one in which AI conducts consultations independently, with clinicians providing remote oversight.
“We’ve got clinicians out in the back room somewhere, and AI is working at scale, delivering more consultations than can be delivered by one person. So, clinicians become supervisors of semi- or fully autonomous AI agents,” Allen said.
He questioned the implications for doctors’ professional roles, comparing remote supervision of AI-delivered care to military drone operations. “It doesn’t feel like a great thing that you’d want to go into medicine to do,” he said.
Nevertheless, he acknowledged the potential benefits for lower-risk patients. “That’s how you solve the workforce crisis. We don’t have enough people to deliver high-quality care at scale,” he said.
Rethinking Clinical Workflows
Eduardo Freire Rodrigues, MD, chief executive officer and cofounder of UpHill Health, a healthcare technology company based in Lisbon, argued that healthcare systems should redesign clinical pathways around AI rather than simply using it to make existing consultations more efficient.
“We should start with the waiting room. We should focus on where access issues exist and points in the patient journey where the standard of care is no care at all,” he asserted.
Rather than focusing on marginal improvements in consultation efficiency, Rodrigues suggested that AI could help address delays in accessing care. “Instead of us thinking inside the appointment, just to optimize 30 seconds of that appointment, we should think outside of the appointment. How can we re-engineer the workflow to unlock capacity and increase access?”
Safety and Accountability
The prospect of AI delivering consultations independently raises questions about clinical oversight, patient safety, and accountability. Allen warned that clinicians could become the “moral crumple zone,” bearing responsibility for failures involving autonomous AI systems despite having limited control over their decisions.
Rodrigues argued that developers should be held accountable for the safety and performance of AI systems marketed for clinical use. “If a vendor claims that its products can do something ... that vendor must be held responsible if the device is nonfunctional or unsafe. We must assess AI in the same way we assess other medical products,” he said.
Allen also cautioned against assuming that technological progress would necessarily benefit underserved populations. “I’m an optimist about the technology getting better. I’m a cynic about our ability to translate good technology into gains that benefit people, particularly from underserved populations,” he said.
The challenge, panelists suggested, is whether healthcare systems can use AI to expand access safely while preserving the human relationships and professional accountability central to patient care.
Rodrigues, Cioè-Peña, Allen, and Tsur had no further disclosures.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.