AI’s Growing Role in Health Care: Dangers and Opportunities

Commentary

A wave of artificial intelligence (AI) innovation has hit the health care system, rocking the boat in significant ways. AI now plays important roles in many areas of medicine, including medical imaging and diagnostics, notetaking and recordkeeping, billing, drug research, and triage and symptom checking.

Two-thirds of American physicians and 85 percent of health care organizations now use AI in their practice. One of the most common applications of AI in health care takes the form of “ambient AI scribes” that listen to patients’ visits and autogenerate transcriptions and notes for electronic health records. Another common place AI can be found is in the lab and image reading room, where AI-assisted diagnostic tools review scans and samples. These systems already match or even exceed human diagnosticians in interpreting and diagnosing data. Health care systems also use AI to handle the logistical background of their practices, including scheduling, billing, and even triaging patients.

With this tidal wave of change comes a strange mixture of optimism and pessimism, hope and fear, opportunities and dangers.

First, some good news. Many doctors and medical professionals who responded to The Epoch Times’ inquiries mentioned the benefit of AI medical scribes. These scribes take an enormous burden off doctors’ shoulders. Now they can focus on being present with patients rather than constantly turning to the computer to type notes. AI slashes doctors’ clerical and bureaucratic workload—removing hours of paperwork—leaving them more free to better perform their real job of caring for patients.

Dr. Cooper Stone, psychiatrist and assistant professor at the University of Pennsylvania Perelman School of Medicine, described for The Epoch Times the relief AI can bring on the paperwork front.

“One of the most onerous parts of practicing medicine is documentation,” he said.

“It’s very challenging to be fully present with a patient when one has to simultaneously take grammatically correct, organized and detailed notes,” he told The Epoch Times. “AI scribes offer a great solution to this issue and allow clinicians to be more fully engaged with their patients.”

However, at the same time, Stone is realistic about the limitations of these robo-scribes. He pointed out that they can often miss important nuances in the way patients speak, which often reveals as much as what patients say.

“This is particularly important in my field of psychiatry, where what is said may not be as important as how it is said,” he said. “For example, the listening AI scribe may document, ‘the patient reports he is not depressed,’ which paints a very different picture from, ‘the patient reports in a soft and defeated tone that he is not depressed.’”

Another of Stone’s concerns relates to privacy. He noted that patients aren’t always aware of the way their information could be commoditized by being used to train AI scribes. The AI scribe records intimate details of doctor–patient interactions, and the guarantees of privacy are not bulletproof.

“Even if reassurances were in place, AI scribes, while convenient, increase the risk of a data breach as more entities have access to patients’ health information,” he warned.

This becomes even more problematic when ambient AI scribes autopopulate consent language without the patient’s knowledge. That very issue lies at the basis of the class action Sharp HealthCare lawsuit currently underway.

Covert Data Collection

AI scribes introduce another player into the system of gathering and storing sensitive medical information. And the more players involved, the greater the risk that data get leaked, sold, or stolen.

Urologic surgeon, writer, and filmmaker Dr. George Ellis told The Epoch Times: “The use of vast, complex datasets—including genomic information, medical records, and real-time biometric data—raises significant concerns about data privacy and the potential for security breaches and thus will require ethical and legal frameworks that are still catching up with technological capabilities.”

In some cases, the implementation of AI systems isn’t just about improving logistical efficiency and mitigating paperwork overload. Sometimes AI gets directly involved with diagnosing and treating injuries and diseases. AI not only transcribes conversations and generates notes, but also highlights information found within those notes, suggesting diagnoses and even treatments. In specialties such as pathology or radiology, AI can more quickly and sometimes accurately identify red flags in scans and medical imagery.

As Dr. Mikel Daniels, podiatrist and chief medical officer at WeTreatFeet Podiatry, explained, “When they are used well, these systems reduce missed findings and help generalists tap into specialty level decision support at the point of care.” This is a potentially life-saving application of the technology.

But although AI can speed up or enhance these processes, it can sometimes muddle them.

“Most current AI is built to reduce uncertainty,” Daniels said. “It will flatten probabilities into recommendations, and present those options in an authoritative tone. This might be a hallucination, or if used incorrectly, pushes care toward a treatment that could be unnecessary.”

Sometimes, the AI is just simply wrong—which can pose a hazard for patients. For example, one study found that 23 percent of AI chatbot responses to questions about sexual health included inaccurate information. Although the use of medical AI chatbots at home isn’t the same as the use of AI by experts in a clinical setting, it demonstrates danger lurking underneath the surface, even when the systems are used by doctors. An excessively admiring view of AI’s capabilities and an overreliance on its outputs could present genuine health risks to patients.

The issue gets aggravated by the inscrutability of some of these systems.

“Many advanced AI models, particularly deep learning systems, are ‘black boxes’ where it is difficult to impossible to understand how they arrive at a specific prediction or decision,” Ellis said. “In high-stakes medical scenarios, clinician oversight is essential to trust and verify recommendations, which is challenging without transparency.”

Maintaining oversight and accountability must be a top priority in the ongoing implementation of medical AI. Medical investor and entrepreneur Joanna Nathan pointed to the need for legislative oversight.

“Policy needs to address clinical accountability, not just product approval. … When AI influences diagnosis or treatment, oversight should extend into how these tools are introduced, trained on, and supervised in clinical practice,” she said.

Some doctors, including Stone, argue that—regardless of oversight—AI systems fall short in clinical settings because they lack nuance, empathy, and human-level adaptability. AI operates on algorithms. But medicine is about so much more than just numbers, flowcharts, and if/then functions.

“Medicine is not a science, it’s an art,” Stone said. “Computers lack clinical experience and patients rarely fall neatly in any one diagnostic category or treatment algorithm. Nothing is ever black and white. There are shades of gray, and we often have to color outside of the lines.”

Programming Humanity Away

Here, Stone raises another key concern in the conversation about AI: What gets lost when systems and machines begin to replace human connection, empathy, experience, and initiative? Daniels pointed out that AI implementation can make health care workers and patients feel “more like data sources than people working on a problem.”

Patient trust and doctor–patient relationships can suffer when systems take precedence over sympathy. There’s something distinctly healing about the human touch, the human smile, and the human voice that says, “I see you, I hear you, I understand you, and I want to help you.” AI lacks all of that.

On the health care workers’ side, overreliance on AI can decrease their pride and fulfillment in their work. Some skilled professionals feel pushed to the sidelines by automated processes.

“Other work on automation suggests that as tasks are automated, workers often report less stress but also worse health and lower job satisfaction,” Daniels said. “People feel more monitored, less autonomous, and less central to the outcome.”

Maybe a middle ground can be staked out. Maybe AI can be used to save time and reduce the bureaucratic burden. It may allow doctors to do what they’re meant to do: connect with and heal patients.

“It is unlikely that AI will replace doctors, nurses and technicians,” Ellis predicted. “However, it can provide the tools to optimize diagnostic and treatment management and make patient care more efficient and allow the human providers to create a patient-focused environment.”

Still, the temptation to cede more ground to the machines remains. Will we find a truly healthy balance in the medical applications of AI, or will it strip away more of the humane element from a medical system that already suffers from depersonalization?

Dr. AI may be in, but the jury is still out.

Views expressed in this article are the opinions of the author and do not necessarily reflect the views of The Epoch Times.