Artificial intelligence scribes are transforming clinical documentation by recording and transcribing patient visits. While these tools save clinicians time and restore eye contact in the exam room, public health experts warn they risk omitting critical medical details.
Healthcare providers increasingly rely on ambient artificial intelligence tools to capture conversations during patient visits, raising concerns about data privacy, clinical accuracy, and patient consent. Katherine E. Goodman, PhD, JD, an associate professor of epidemiology and public health at the University of Maryland School of Medicine, addressed these challenges in a recent episode of “The UMB Pulse” podcast. Goodman studies artificial intelligence in healthcare and serves on the faculty of the University of Maryland Institute for Health Computing.
How Ambient Medical Scribes Capture Clinical Conversations
When a physician uses an AI scribe, the process begins long before advanced generative models engage. Typically running as an application on a smartphone or tablet, the software records the spoken dialogue between the doctor and patient during the appointment. Once the encounter concludes, the application applies statistical audio-to-text conversion tools to generate a raw transcript.
Generative artificial intelligence then parses this transcript to synthesize a draft summary of the clinical visit. The physician must review, edit, and sign off on this draft before it enters the patient’s permanent medical record as the official documentation. Goodman describes this setup as a “triple win” that satisfies clinicians, patients, and healthcare systems by reducing the burden on clinicians and improving billing procedures.
AI Transcription Tools Require Human Oversight and Risk Missing Vital Context
- Ambient Recording: AI applications listen to exam room conversations through a provider’s device, turning spoken words into written text.
- Human Oversight Required: Software generates only a draft summary; a licensed human must review and sign off before notes enter permanent health records.
- The Omission Hazard: AI tools sometimes drop vital context—such as recent travel history linked to blood clots—making omissions harder for doctors to catch.
Data Handling Policies Vary Across Healthcare Systems
Goodman notes that data handling policies vary widely across healthcare systems and software vendors.
Recognizing the Hidden Dangers of Clinical Omissions
While software hallucinations—where an AI model makes up information that was never spoken aloud—pose risks, Goodman highlights a more concerning hazard: clinically significant omissions. Medical diagnosis often hinges on precise circumstantial details mentioned during an encounter. If a patient mentions leg pain immediately followed by an overseas flight last week, an AI summary might record the pain while dropping the flight history.
Patients Can Opt Out of Automated Transcription During Appointments
Patients can opt out of AI transcription during their appointments, though doing so may involve completing consent forms or encountering providers who are not prepared to take notes the "semi-old-fashioned way." Patients uncomfortable with automated transcription should inform their provider before the examination begins.
References
- University of Maryland School of Medicine. “The UMB Pulse Podcast: AI in the Exam Room with Katherine E. Goodman.” University of Maryland, Baltimore.
- University of Maryland Institute for Health Computing.
