Dr. Anthony Chang

“The key question isn’t ‘What can AI do?’ but ‘What should AI do?”                                                                                               
John C. Havens, global lead for IEEE AI ethics initiative

Ambient AI is the general concept that AI works in the background by listening, watching, or sensing without requiring explicit prompts from uses. The ambient AI spectrum ranges from a relatively simple AI scribe to a sophisticated ambient clinical intelligence support. The applications can include: camera system in the waiting room to monitor fall risks, monitoring system for hospital bed flow and staffing in real time, and voice assistants for answering questions. Most of the ambient AI tools are integrated with EHR. Overall, there has been enthusiasm from this initial period amongst clinicians for these tools, especially AI scribes. Of note, human scribes have been effectively deployed in the clinical setting for a few decades, especially in the ED and subspecialty setting.

AI Scribes. These are digital tools that use AI to automatically document clinical encounters between a clinician and a patient. This AI tool, by recording the conversation and generating a structured medical note using natural language processing (NLP), obviates the need for a clinician to type a note during the interaction. The output from these tools is documentation only in the form of progress notes, H&P, discharge summaries, etc. While there is potential for time savings, burnout reduction, and accuracy improvement, the limitations include concerns about accuracy, privacy, and workflow integration. If these tools work in the background without continual commands from the clinicians, then these can be called ambient AI scribes (see below). The AI scribe is best for straightforward note-taking situations such as that in the outpatient clinic.
Companies in this space include: Abridge; Heidi Health (Australia); Nuance Dragon Medical One; and Suki.ai.

Ambient AI. This is a broader concept than the aforementioned AI scribe in that the ambient AI is deployed in the background of clinical encounters and functions without explicit prompts. The focus is not solely documentation as in AI scribes, but a myriad of functions that includes basic clinical support (flagging abnormal findings), billing and coding assistance, patient education summaries, and EHR task automation (orders, prescriptions, and reminders). Ambient AI, therefore, is a more capable assistant than the aforementioned AI scribe that passively listens with documentation as only one of its functions. Ambient AI is ideal for complex workflows and high acuity environments such as hospitals. The cost range is higher than the AI scribes and is estimated to be $50-$200 per month; the ROI, however, is likely to be higher with ambient AI.
Companies in this space include: Ambient Healthcare; Nuance Dragon Ambient eXperience (DAX) Copilot; and Avo.

Ambient Clinical Intelligence (ACI). A subset of ambient AI is ACI, AI that is specifically focused on clinician-patient interactions that includes not only clinical documentation and coding but also decision making. This AI, also powered by speech recognition, NLP, and large language model, is more narrowly focused on the clinical documentation and workflow support for clinicians. There is definitely some blurring of the boundary between ACI and ambient AI described above. Perhaps it is fair to state that ACI is the clinical workflow component of ambient AI in general. There is also some understandable confusion between ACI and agentic AI: the latter does not just generate outputs but also plan, decide, and act toward a pre-determined goal. Agentic (instead of ambient) clinical intelligence is the resource of the near future in healthcare.
Companies in this space include: Nuance Dragon Ambient eXperience (DAX) Copilot; Abridge; and Ambience Healthcare.

The ambient AI and its spectrum from AI scribes to ambient clinical intelligence with agentic features will be a hot topic presented at AIMed25 later this year.

Artificial Intelligence in Medicine (AIMed) is the longest running meeting (inaugurated in 2013) focused on artificial intelligence in medicine and healthcare. The meeting is usually attended by well over 1,000 clinicians (from over 50 subspecialties) as well as healthcare leaders, data scientists, students and trainees, and entrepreneurs and investors from all over the world. The 3-day meeting covers a broad range of topics directly or indirectly related to artificial intelligence in medicine and healthcare such as generative AI, agentic AI, large language models, cybersecurity, and intelligent extended reality with three special tracks this year: AI in pediatrics and neonatology, AI in health professional education, and AI and mental health of clinicians and patients. Several special features of this meeting of AI in health include: breakfast workshops focused on hot topics (as determined by our attendees as attendees vote on topics they like to hear), afternoon subspecialty breakout sessions (for over 20 different subspecialties or domain areas), abstract competition with scholarships for accepted scholars, and a special one-day American Board of AI in Medicine (ABAIM) course that has become very popular with our attendees. This year, there will be a special Chief AI Officer agenda during AIMed25.

See you there!