Teaching Empathy in the Digital Age

Wearing white coats and virtual reality (VR) headsets, Dr. Rohan Jotwani and Dr. John Rubin, co-directors of Weill Cornell Medicine’s Extended Reality Anesthesiology Immersion Lab (XRAIL), turn on their computer and enter the virtual New York City apartment of A.I. C.A.R.L., a virtually simulated chronic pain patient with sickle cell disease.

In a setting reminiscent of a mid-2000s open-world video game, C.A.R.L. stands firmly and looks straight ahead through the screen, calmly answering questions in crisply ironed jeans and a button-down shirt. When the physicians ask a series of questions about how he’s managing his pain, including how much medication he’s taking, C.A.R.L. responds with nuance: “It’s not about the dosage,” he tells them in his deep, distinctive digitized voice. “It’s about finding a solution that works for both of us.”

C.A.R.L. is a new response to the challenge of preparing medical students and residents for complex, emotionally charged interactions with their future patients. To provide quality care, doctors must be able to communicate clearly with patients to understand their needs and concerns, especially when treating complicated conditions like pain, which can be difficult to evaluate and successfully treat. But opportunities to practice with manikins or standardized patients (paid actors simulating patient scenarios) before meeting their first real-life patients are often quite limited.

Enter C.A.R.L., which stands for Conversational Agent Relief Learning in Pain Management. The brainchild of Dr. Jotwani and Dr. Rubin, both assistant professors of clinical anesthesiology, C.A.R.L. is a large language model (LLM) designed and created by the team at XRAIL, whose goal is to pioneer immersive learning approaches that help physicians and medical residents gain valuable knowledge around appropriate protocols for managing conditions like pain in order to build consistency in each patient encounter.

Ultimately, users’ primary mission is to engage with C.A.R.L. and other diverse virtual patients by communicating in a way that results in optimal health care. Anesthesiology residents at New York-Presbyterian/Weill Cornell are increasingly training with him, and Dr. Jotwani also introduces him to Weill Cornell Medical College students in two elective courses.

Just as physicians cannot predict how a patient encounter will go, C.A.R.L. does not answer from a script, reflecting the infinite variables of each patient case. Another advantage? C.A.R.L. never gets tired, so trainees have an ongoing opportunity to continually hone their skills.

The anesthesiologists — both self-taught in development of extended reality — decided to focus C.A.R.L.’s conversations around pain management because, as Dr. Jotwani explains, “We wanted to find a field where there was a great deal of complexity and there wasn’t always an absolute right or wrong, because I think that tends to be the hardest to train for.” Chronic pain patients, he explains, often face the stigma of addiction in looking for treatment of their condition.

“Treating someone who lives with pain requires such a deep understanding of what it means to be in pain. It’s not always black and white, and because there’s such nuance to it, it’s not enough to have a simulation with a standardized patient once per year. You need a way to continue training in these sorts of complex clinical scenarios,” says Dr. Jotwani, who is also the Nanette Laitman Education Scholar in Entrepreneurship.

Dr. Rubin emphasizes how critical it was to integrate the challenges and biases patients face for a realistic, comprehensive and nuanced virtual simulation experience.

“We see patients every day who have felt marginalized by their physicians and are not sure if they can even trust us,” he says. “There’s frustration with that, but if we’re going to prevent that from happening in the future, we must teach medical students, residents and fellows how to interact with patients.”

Dr. Alessandra Riccio, a third-year anethesiology resident, says C.A.R.L. allows her to “understand how to improve in a safe environment.”

Exploring the Potential of Immersive Learning 

The advances in natural language processing that have enabled creations like C.A.R.L. to provide a realistic patient experience have been under development for decades. For Dr. Albert “Skip” Rizzo, director of medical virtual reality at the USC Institute for Creative Technologies Research in Playa Vista, Calif., the potential of these innovations is what he imagined the field of immersive learning could become in medicine — a tool that can uniquely enhance the quality of care delivered by physicians.

Already, recent studies have shown the significant impact of immersive experiential learning using virtual human patient simulations, with findings published in JAMA demonstrating improved post-training outcomes in motivational interviewing — a counseling technique highly dependent on empathy to encourage patient lifestyle changes.

Powered by AI and VR technology, the design and methodologies of C.A.R.L. largely mirror those of a video game where patient simulations are individualized based on the scenarios and personalities given to each AI conversational agent (AICA).

But unlike a video game, the responses from C.A.R.L. and other AICAs are not scripted. Dr. Jotwani and Dr. Rubin utilize a four-step process to create a sense of realism and spontaneity in how each AICA responds, starting with the development of the LLM, which leverages an amalgamation of peer-reviewed data in clinical research and medical textbooks, real patient stories, and even some experiences that the two anesthesiologists have observed in clinical practice.

Dr. Jotwani and Dr. Rubin begin building the personalities of these virtual patients using a process called narrative hybridization, in which they think of a clinical scenario and shape the AICA around that scenario. They embed certain details into the design, including where the AICA was born, their family dynamics, professional work and hobbies to differentiate how each AICA responds. Dr. Rubin notes that C.A.R.L., the team’s most developed AICA, was able to provide input — using its core training data — on details of his virtual appearance, including his clothes and living space.

The last step in creating an AICA is known as in-clinic expert engineering. “It’s kind of like where an actor and director work together,” Dr. Jotwani says. “We have conversations with the AICA, where we can stop the model and then retrain them if a response is not the way we want it to be.”

The growing capabilities of AI in demonstrating emotional intelligence highlights the technology’s utility as a learning platform, says Dr. Rizzo, a research professor in the Department of Psychiatry and School of Gerontology at the University of Southern California. “You can now have a piece of software that sounds like a human, emotes like a human, shows empathy like a human, but is not a human,” he says. “AI is not the same as interacting with a real person, but there is true value in practicing for everyday social interaction without worrying about someone judging or evaluating you.”

The potential of practicing in different scenarios with virtual patients of varied ages, ethnicities, genders, sexual orientations and health conditions could prove invaluable in shaping more empathetic medical residents and trainees — enabling them to identify potential biases or gaps in their clinical thinking and gain a broader understanding of diverse patient experiences. “This all comes down to giving people opportunities for experiential learning that can be systematically delivered across different types of contexts and characters, but in a safe place to learn from that experience,” says Dr. Rizzo.

This accessible “safe place” to hone soft skills in communication and empathy is key in a field like medicine where speaking with patients is the gold standard for learning social dynamics, notes Dr. Mark Zhang, instructor of medicine at Harvard Medical School and a palliative care consultant at Dana-Farber Cancer Institute.

As the founder and president of the American Medical Extended Reality Association, an organization dedicated to advancing the field of medical XR, Dr. Zhang says the opportunity of immersive learning lies in its potential to be the “great democratizer” in standardizing and streamlining the medical training process.

“Historically, medical training has been difficult to scale, because even if you have a great standardized patient actor, you could only reach up to maybe 100 encounters over the span of an eight-hour day,” he says. “Now, with AI avatars, you can do near infinite [encounters]. And it’s infinitely configurable, so you can work with varieties of different virtual patients at different levels of skill.”

Created by Dr. Rohan Jotwani (center) and Dr. John Rubin (right), C.A.R.L. is designed to help trainees like Dr. Alessandra Riccio (left) understand diverse patient experiences.

How Immersive Learning is Advancing Medical Training

Immersive learning tools like C.A.R.L. are opening new possibilities for understanding and conveying empathy in medical training, but its visual and interactive capabilities will also be transformative, Dr. Zhang says.

For procedural-based medical fields, like surgery or anesthesiology, there’s significant value in the immersion that VR-based medical training offers, he notes. “Being immersed in a virtual environment like an operating room and learning the steps of these procedures is a powerful training tool,” he says. “Before, you watched a video or read a book to learn, but now VR enables you to take part in these scenarios.”

As part of Weill Cornell Medicine’s resident lecture series, C.A.R.L. and other VR-based simulations are now being made accessible to residents. Dr. Alessandra Riccio, a third-year resident in clinical anesthesia, has participated in several training sessions with C.A.R.L.

“Since it’s immersive, it forces you to be engaged in procedures or conversations. I can go through the steps, catch myself making mistakes and understand how to improve in a safe environment,” says Dr. Riccio.

By echoing real reservations that patients have in a virtually relevant scenario, AICAs are helping trainees understand the psychology behind certain patient responses and behaviors. Dr. Riccio is able to use C.A.R.L. to help lead her to appropriate protocols for managing patients with sickle cell disease, which disproportionately affects Black patients and often is challenged by drug-seeking stigmas and biases. C.A.R.L. is intentionally designed to help trainees overcome bias and provide evidence-based, quality care.

“When it comes to communication, there’s real vulnerability to misinterpretations, both in trying to understand what somebody is saying to you or to having your intentions conveyed,” explains Dr. Kane O. Pryor, executive vice chair for academic affairs and director of clinical research and of education in the Department of Anesthesiology, and associate professor of clinical anesthesiology at Weill Cornell Medicine. “Emotional conversations are very challenging,” says Dr. Pryor. “You can’t improve without practice and experience, and these AI models really help trainees learn which approaches are effective and comfortable for patients.”

What’s Next for Immersive Learning in Medicine

For Dr. Jotwani and Dr. Rubin, C.A.R.L. is just the beginning. Through a grant from the Accreditation Council for Graduate Medical Education, they plan on developing other AICAs to train doctors in helping patients manage pain during labor and delivery and at the end of life.

As a palliative care consultant, Dr. Zhang stresses how important training is to prepare for conversations around serious illness, and how AICAs can help fill this critical need. “The real opportunity that arises with AICAs,” he says, “is that you can adjust the difficulty level by your skill set and gain valuable experience before doing it in real life where stakes are much higher.”

In looking to the future of AICAs, Dr. Zhang and Dr. Rizzo agree that a standardized approach for designing and using the technology will be key to ensure ethics are at the forefront of innovation in medicine, and diverse patient populations are accounted for during training.

“We live in a multicultural world [and we need] specific training to understand the norms of different cultures in the perception and implementation of medicine,” says Dr. Rizzo. “So, it’s critical to ensure that peer-reviewed content that is safe and ethical serves as the foundation of information used when designing these virtual guides, mentors and health care support agents.”

“We as individuals only have our own personal experience, but as physicians, we interact with people who come from an infinite range of cultural and experiential backgrounds,” adds Dr. Pryor. “Sometimes there are challenges in aligning across those differences, and the best way of bridging those potential gaps is through experience and feedback.”

As the capabilities of AI continue to evolve and transform the possibilities of medical training, Dr. Rubin and Dr. Jotwani are optimistic about the potential of AICAs to empower quality care and better patient outcomes.

“As we look for more opportunities to create and incorporate immersive learning models into medical training, our primary goal remains the same: to help make more compassionate physicians,” says Dr. Jotwani. “We really do think that, ironically, spending time with AICAs like C.A.R.L. will help physicians become more empathetic and well-rounded in real-life encounters.”