The Promise of Ambient AI Technology in Medical Education: Opportunities and Guardrails

Ambient AI and Clinical Care

Operating in the background, ambient AI technologies, commonly known as AI scribes, capture the patient-provider conversation to identify medically relevant information and structure it into a coherent clinical note. A significant contributor to health provider burnout, the administrative burden of clinical documentation in electronic health records (EHRs) is reduced by deploying these tools in the clinical workflow. Early studies evaluating these tools suggest that ambient AI shortens the time required to sign off and close a patient chart encounter, reduces time spent after hours completing notes, improves provider job satisfaction, potentially improves patient engagement during the clinical encounter, decreases short-term provider burnout, and may better address documentation gaps [1-3].

Despite these benefits, questions about the accuracy or quality of ambient AI remain. More research is needed to address several aspects of this technology, including the safety and health impact of omission and commission errors (eg, not including information, providing incorrect contextual interpretation, or adding incorrect content), its utility in certain clinical environments, its ability to adequately understand and transcribe the speech of patients or providers with differing accents [4], the optimal integration and adoption of the technology into clinical workflows and EHRs, its overall cost-effectiveness, and its long-term impact on the cognitive workload and clinical reasoning of even seasoned clinicians [5,6]. Clinical implementation should follow a thoughtful and guided approach, anchored in implementation science and quality improvement pedagogy [7,8].

Ambient AI and Medical Education

Overview

There is no doubt, however, that as this technology improves and the health system gains more confidence in its use, there is likely to be widespread adoption of ambient AI, akin to moving away from paper charts and adopting EHRs several decades ago. Medical trainees are likely to witness this change and even want to use these tools themselves during their training. It is therefore imperative for medical educators to pay close attention to how these tools are being introduced and used in medical schools, residencies, and fellowship training programs where the youngest members of the physician workforce are often engaged with and more likely to be early adopters of new technologies in their daily personal and professional lives. Critically, there are few data on the potential use and impact of these technologies on formative educational experiences, such as skill development in clinical communication, reasoning, and documentation.

This viewpoint highlights the challenges that may be caused by offloading documentation during training across the medical education continuum when learners develop reasoning, synthesis, and communication skills. We share potential consequences, outline deliberate guardrails, and advocate for intentional investigation into understanding these risks.

Undergraduate Medical Education

We have not found any published studies to date that look at the impact of ambient AI on the early acquisition of clinical documentation and reasoning skills. A potential benefit of automating the note-taking process may be that medical students become more present and engaged during patient and standardized patient encounters, allowing for a deeper focus on developing fundamental skills, such as taking histories, performing physical examinations, and building patient rapport. The reduction in students’ documentation cognitive load may allow for more capacity to observe, process, and learn from the nuances of the patient-provider interaction and the clinical reasoning demonstrated by their preceptors. Early exposure to and training with AI tools could better prepare students for the technologically advanced health care environments in which they will eventually practice. Ambient AI could offer complementary educational opportunities as well. By teaching students to critically engage with the notes generated by ambient AI and comparing these notes to the students’ own independently written notes, faculty have an opportunity to initiate meaningful discussions about communication, documentation structure, important details that guide the differential diagnosis, and clinical reasoning. However, educational theory encompasses the process of learning by doing, which is a key value of clinical documentation and reasoning. It emphasizes active learning, deliberate practice, cognitive apprenticeship, and scaffolding, all in the progression from novice to expert. These components of learning may be negatively affected by removing clinical documentation during the undergraduate medical education process.

Graduate Medical Education

In graduate medical education, small pilot studies of resident physicians across different specialties suggest increased satisfaction and decreased short-term burnout with ambient AI use. The reports vary on the impact that ambient AI scribe tools have on time spent documenting clinical notes both during and outside of work hours [9-11]. In theory, less time on clinical documentation should allow for more time spent on patient care, knowledge acquisition, or practicing new skills. To date, there are no published studies evaluating whether residents or fellows foster more effective relationships with their patients when the burden of clinical documentation is reduced through ambient AI use. Additionally, we have not seen any specific curricula on teaching trainees effective reviewing, editing, and oversight of the output from ambient AI scribes beyond vendor-created technical skills on how to navigate platforms.

Potential Unintended Consequences and Guardrails

The benefits of ambient AI in clinical practice need to be balanced against the possibility of unintended consequences, which require clear and intentional guardrails to prevent harm (Table 1).

Writing a clinical note is an important exercise in organizing thoughts, identifying knowledge gaps, and communicating effectively with other health care professionals. Overreliance on AI-generated notes may cause these skills to atrophy. Unintended consequences include the potential for diminished development of essential clinical reasoning skills or cognitive deskilling in those who have newly learned these skills. The act of manually documenting a patient encounter nudges trainees to actively synthesize the information they have gathered, formulate a differential diagnosis, and develop a treatment plan. Automating this process could lead to a more passive learning experience and a superficial understanding of the patient’s condition. Ambient AI may be prone to errors of interpretation and to the omission of nuanced details. Faculty will need training on how to effectively teach and supervise students when using ambient AI and when best to introduce these tools.

Value of Evidence-Based Practice

Research on the impact and utility of ambient AI in medical education and the developmental acquisition of skills is needed. Priority should be given to proposing and implementing a pragmatic study that allows for controlled exposure to ambient AI while adjusting for individual variations in learning outcomes. Such a study should use rigorous methodologies, such as multicenter randomized crossover trial or step-wedge or parallel-group designs, to carefully determine the potential for impact on the fundamentals of medical education. Specific educational outcomes of interest need to be studied, such as note quality and clinical reasoning, patient communication, and time spent in direct patient care, as do changes in cognitive load and burnout from use of ambient AI, among other areas of impact. Additionally, vendor neutrality will also have to be considered, though studies may need to be conducted across vendors based on ambient AI features available for study, such as clinical decision support or diagnostic reasoning.

Furthermore, concerns about consent as well as patient and trainee privacy will need to be addressed when such technology is implemented in medical education, especially if audio recording is used to teach trainees and give feedback. Lastly, there is increasing evidence that ambient AI is used differently across specialties; therefore, trainee education will also have to be modified by specialty [12,13]. A one-size-fits-all training plan is unlikely to translate across all specialties, especially for fields such as pediatrics, where clinical practice has specific nuances, such as developmental stage of the patient and complex caregiver relationships.

Summary

Ambient AI holds the promise of reducing the documentation burden of modern medical practice and offering benefits in both clinical workflow and medical education. The impact of these tools on our medical students, residents, and fellows must be actively studied. Medical schools themselves need guidance on how to incorporate these technologies to the maximal benefit and least harm of their students and trainees. Accrediting agencies such as the Liaison Committee on Medical Education need evidence to set policy on the impact of ambient AI on medical education. We call for a thoughtful and evidence-based approach to realize the potential of ambient AI while mitigating the risks to the acquisition and retention of fundamental skills for the next generation of physicians. Robust research studies are urgently needed to navigate this new frontier and ensure that the integration of ambient AI in the clinical learning environment ultimately enriches, rather than diminishes, the practice of medicine.