The Human Touch in the Age of AI:
Rethinking Healthcare's Technological Future

Christian Rose, MD, director of the "Missingness in Action" conference on absent and missing data and the future of AI in health care, discusses the possibilities and pitfalls in AI automation and eavesdropping.

You’ve been sounding the alarm about data inclusion and health equity for several years. Your current research pivots from diagnostics to care delivery. Why the shift and how do you see technology helping or hurting the physician experience?

The last major technological innovation in our field —electronic health records (EHRs)—was widely anticipated to make writing notes easier, prevent errors, and improve health care delivery. In reality, it has had one of the most detrimental impacts on the practice of medicine. A significant portion of physician burnout stems from our computer work and a resulting sense of disconnection from patients as we stare at computer screens. Additionally, when physicians focus on their computers, patients often feel unheard, worsening dissociation, and weakening the therapeutic relationship.

Our research, which is funded by the American Medical Association, aims to understand the importance of human factors and their environment in health care delivery. This includes assessing emergency physician computer use and identifying tasks to automate. Operationalizing high-friction parts of emergency medicine—such as ordering, writing notes, retrieving data, and summarizing information—may allow us to be more present for patient interactions. Narrative Medicine has shown this approach can improve the data gathered at the point of care and impact decision-making. 

What is Narrative Medicine and how can ambient AI improve patient care?

In addition to EHR roadblocks, part of the challenge is how we conduct conversations with patients. In emergency medicine, we sometimes need to gather information quickly and follow heuristics to balance competing interests like time and throughput.

The book Weapons of Math Destruction explains that when we focus on criteria we assume are strongly linked to outcomes, we often neglect other important areas. This narrow focus creates a feedback loop that prevents us from considering new information in an algorithmic framework. The focus becomes increasingly detailed, often missing the broader context, which undermines the communication that is essential for building rapport and eliciting relevant information.

Narrative Medicine was founded by Rita Charon at Columbia University, who was one of my mentors. The idea is that when patients are encouraged to share their narratives, we gain a deeper understanding of issues like abdominal pain or headaches.

The big question is: Can machine learning actualize the potential of Narrative Medicine?

I’m collaborating with [Stanford emergency medicine physicians] Brian Suffoletto, Carl Preiksaitis, and David Kim to explore how we can record physician-patient conversations. Using ambient listening technology, we hope to extract data on social determinants of health, physician orders, and language use, as well as engagement.

As opposed to other uses of AI in clinical practice, this technology isn’t designed to make predictions but rather to enhance data collection. Our approach focuses on passive listening and data extraction rather than requiring physicians to input everything into the EHR. By recording conversations, we hope to identify important factors, like housing stability, access to food, or other patient-centered data, that are often missed when physicians narrow the encounter to a medical complaint and manually input data.

You might ask “why not simply use available software to populate a note?” Well, these technologies are often trained on highly structured notes and aim to mimic current models, which limits their ability to capture the full context of conversations.

The same recordings could also analyze the emotional quality of exchanges and measure engagement levels, providing valuable insights into physician-patient interactions. These interpersonal relationships and context of care are to me where the science behind the art of medicine lies.

What other relationships do you consider? How can a “Moneyball” approach lead to higher-quality care?

I spend a lot of time thinking not only about how the patient-physician pair can be understood, but also how the physician and the rest of the care team interact, whether in the ED or across someone’s care network. I’m particularly excited about a project involving "medical sabermetrics" – using contextual factors like busyness,  interruptions, and team dynamics to create new quality metrics that more accurately represent what high-quality care looks like.

This concept draws inspiration from sports. Historically, baseball metrics focused on basic stats like hits and runs, which didn’t fully capture a player’s effectiveness. The introduction of advanced metrics, made famous by the "Moneyball" approach, revolutionized how we assess player performance by emphasizing situational factors, like the significance of a hit with runners in scoring position compared to a hit with no one on base or the odds of success when one player plays versus another.

Similarly, in health care, we must move beyond simplistic measures, such as the number of medical errors or CT scans, which exist in isolation. These metrics need context to be truly meaningful, and they need to be comparable across care settings. We all recognize that bad outcomes are not solely due to misdiagnosis. Rather, they are often due to issues happening within a broader context of care. We can investigate what the data reveals about human characteristics and teamwork. Finding ways to leverage technology could help us enhance our understanding of how we work as a cohesive unit, making us more effective overall.

Can technology solve the supply and demand inequity in health care?

While technology can replace certain functions, it can also enhance access and therefore increase demand. The introduction of ATMs didn’t eliminate bank tellers; instead, it increased access to banking services. Similarly, in medicine, as we implement AI and other technologies, we may see an increase in demand for health care services. This phenomenon, known as induced demand, highlights a growing mismatch between supply and demand in health care.

Technology may streamline processes, but there will still be a need for human interaction—especially when it comes to discussing medication or making critical decisions. We may find ourselves with more patients than ever and the same number of health care providers, which poses a significant challenge.