As an emergency room doctor, I often find myself navigating chaos. Between alarms, split-second decisions, and endless documentation, the promise of artificial intelligence is hard to ignore. What if AI could shoulder some of the burden, allowing me to focus more on my patients?

But as a chief medical officer responsible for practice improvement, I approach AI with cautious optimism. While the technology holds promise, its real-world impact has yet to fully manifest.

The hype cycle

Like many new innovations, healthcare AI follows Gartner’s Hype Cycle—initial excitement leading to inflated expectations, followed by a phase of disillusionment when early implementations fall short. Eventually, through refinement and realistic applications, technology reaches a plateau of productivity.

Today, AI in medicine is moving beyond the hype. The focus has shifted from lofty possibilities to identifying specific, practical applications where AI can make a real difference—most notably in documentation and clinical decision support.

AI for clinical documentation

Documentation is a major time drain in clinical practice, making AI-powered ambient scribing an appealing solution. These tools listen to patient-clinician conversations and generate notes, potentially reducing screen time and improving patient interactions.

However, studies show mixed results:

In my experience, ambient scribing helps with straightforward cases like minor injuries or routine complaints, where the patient can provide most of the necessary information. By off-loading note-taking, I can maintain better patient engagement.

However, the limitations become apparent in complex cases. In the ED, I must sometimes gather input from multiple sources—patients, paramedics, family members, medical records—over several hours. AI scribing doesn’t effectively synthesize this information, and most platforms can’t help me generate medical decision-making notes, which are essential for accurate documentation and reimbursement.

One promising advancement is Sayvant, an AI scribing tool developed by Vituity’s Inflect innovation hub. Unlike first-generation AI scribes, Sayvant helps clinicians draft medical decision-making content, making end-of-shift documentation far less burdensome.

AI for clinical decision support

AI-driven clinical decision support (CDS) tools offer another exciting possibility. By analyzing vast amounts of EHR data, AI can provide insights that improve diagnosis and treatment. A frequently cited example is AI’s ability to detect sepsis, a leading cause of hospital mortality.

In theory, AI tools can identify sepsis early by analyzing vital signs, lab results, and other factors. However, in practice, these tools struggle with accuracy:

• High false-positive rates overwhelm clinicians with unnecessary alerts, contributing to alert fatigue.
• Some tools fail to identify true cases of sepsis, limiting their reliability.
• Even when AI detects sepsis early, research suggests that timing alone may not significantly improve mortality rates.

For AI-driven CDS to succeed, it must be accurate, actionable, and seamlessly integrated into clinical workflows. Too many pop-up alerts—many of which are redundant or inaccurate—lead to clinicians tuning them out, potentially missing the one critical notification that matters.

Other promising AI applications

In addition, AI offers potential in medical record summarization. Clinicians often struggle to sift through lengthy patient histories to find relevant details. AI tools that generate concise, high-impact summaries could be game changers.

Imagine a physician meeting a patient for the first time. Instead of combing through pages of records, they could review a brief AI-generated summary, highlighting essential information. Similarly, AI could assist in creating discharge summaries, making them clearer and more actionable for patients and families.

These applications might not grab headlines, but they address real inefficiencies in clinical workflows—saving time, reducing cognitive load, and improving patient care.

A measured perspective on AI in medicine

AI in clinical medicine is a work in progress. While the technology holds immense promise, its limitations are still very real. Rather than expecting sweeping change, the most meaningful progress will come from targeted innovations that address specific challenges.

As clinicians and healthcare leaders, we must engage thoughtfully with AI—experimenting with new tools, sharing real-world insights, and holding developers accountable for delivering solutions that add value.

AI in medicine isn’t a magic fix, but it’s moving in the right direction. The key will be focusing on incremental improvements rather than grand visions of transformation. While we’re not there yet, we’re certainly making progress.

Gregg Miller, MD, is the Chief Medical Officer at Vituity, leading performance-improvement programs and best practices across clinical specialties. Previously, he served as Program Director of Quality and Performance for Emergency Medicine and as Medical Director for emergency departments at San Joaquin Community Hospital and Swedish Edmonds Hospital. He earned his medical degree from UCSF and completed his emergency medicine residency at Harbor-UCLA. An administrative fellow at Vituity, he specialized in CMS quality programs. Dr. Miller also serves on the American College of Emergency Physicians Quality and Performance Committee.