The future of medical note-taking
͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
|
|
|
|
|
|
|
|
Hello and welcome to Making AI Work, MIT Technology Review’s new limited-run newsletter. In this series, we share practical guidance for working professionals about how generative AI is being deployed today across industries, and what you need to know about applying it at work.
|
|
|
|
|
|
|
Various AI tools are already being used in health care—to do everything from streamlining hospital admin to diagnosing cancer. Some of the AI giants are also getting into the health care space. Within the first few weeks of the year, OpenAI debuted ChatGPT Health, designed for consumers, and Anthropic introduced Claude for Healthcare, which is aimed more at health care providers.
But the tools that have taken off the most rapidly are those that are designed to ease the burden of medical note-taking.
|
|
|
|
|
“We are taught in medical school that if you didn’t write it down, it didn’t happen,” says Colin Walsh, a practicing internist and clinical informatician at Vanderbilt University Medical Center (VUMC) in Nashville, Tennessee. “Documentation is a really important part of [a doctor’s] job.”
But a doctor might see 30 or more patients in a day and often won’t get around to writing up each person’s notes until the evening or weekend. “They’re doing it in what’s called pajama time,” says Walsh.
VUMC is one of the medical centers that piloted Microsoft’s Dragon Ambient eXperience (DAX) Copilot—an “AI scribe” to help make documentation a lot easier. In practice, doctors first ask their patients if they are comfortable with them using the technology and then might use a smartphone, for example, to capture a conversation. The scribe will automatically transcribe and summarize that conversation and the summaries can be reviewed and edited before they are added to a patient’s medical record.
|
|
|
|
|
|
|
It is still early days for the technology, and the usefulness of scribes is under research. But surveys suggest clinicians like using them. VUMC first trialed DAX Copilot among 10 doctors back in March 2024. That trial expanded, and by August 2025, the technology was offered to all faculty, staff physicians, and other health professionals in the ambulatory (outpatient) clinics and emergency settings. 78% of the 226 users surveyed said that the tool improved the quality of their documentation, and 74% said it improved their patients’ experience.
In addition to reclaiming their pajama time, doctors using the tool have reported being able to see more patients, focus more fully on those patients, and find it easier to speak to patients in other languages, because the tool will automatically translate conversations to English. “It’s been popular,” says Walsh.
|
|
|
|
|
|
|
|
|
|
|
King’s College Hospital transformed its ability to manage critical healthcare data during the COVID-19 pandemic with AI.
The hospital was able to consolidate multiple data sources into one searchable database supported by natural language processing (NLP) using Elastic's Search AI Platform and Cogstack.
This allowed healthcare staff to access relevant data without the need for specialized terminology. Clinicians got real-time insights that accelerated decision-making, supporting the hospital’s ability to respond to public health crises.
See how AI improved patient outcomes →
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
DAX Copilot isn’t the only AI scribe in the game; alternatives include DeepScribe, Suki, Abridge, and Nabla, all of which are being used in health care settings across the US. A recent trial of DAX and Nabla found that they seem to improve doctors’ ability to engage with their patients. Doctors who used them reported less burnout and work exhaustion by some measures.
How these scribes are used is expanding. In March 2025, Microsoft unveiled its Dragon Copilot, which combined DAX Copilot with speech recognition software Dragon Medical One. The tool allows users to search for medical information and automate a range of tasks including visit summaries and referral letters.
And in November, Microsoft described a Dragon Copilot tool to support radiologists, while DeepScribe has a tool specifically for cancer care, too.
|
|
|
|
|
|
|
But we are still learning just how helpful these tools are, and how receptive patients are to their use. One recent survey conducted in Canada found that while many people trusted their use with human oversight, and could see how their use might benefit patient-doctor interactions, most were reluctant to use them.
And AI scribes aren’t perfect, either. The DAX and Nabla trial found that the tools sometimes leave out important information or use the wrong pronouns. So it is important that their outputs are fact-checked by a human.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Health care providers considering the use of AI scribes—or any other AI tool for that matter—should bear in mind a few key principles.
|
|
|
Run pilot trials. Before adopting any new tool, you want to make sure you’re not inadvertently adding to the workload of clinicians or harming patients.
Keep humans in the loop. AI can hallucinate, and garble information. In a health care setting, this can be potentially dangerous. Any outputs should be checked by a person.
Get patient input. Make sure patients are on board with any new tools, and that they will also see benefits from their adoption.
Keep on evaluating. Health care providers should continually assess how well any AI tools they’re using are working. Are they delivering measurable benefits? It’s also important to keep an eye out for problems that might only become visible once a tool has become more widely adopted. “Every time we go a little bigger, you find those edge cases, or … those bugs you didn’t expect to see,” says Walsh. “In Silicon Valley, they often say ‘move fast and break things,’ but in health care, that’s harm. That hurts someone.”
|
|
|
|
|
|
|
|
|
|
|
|
|
Next time, we’ll learn how Westinghouse is using AI to advise crews as they construct nuclear reactors. Later in the series, we’ll look at how several small business owners are applying AI to save time and help manage their inventory. In the meantime, check out our staff’s predictions for what’s next for AI in 2026.
|
|
|
|
|
|
|
MIT Technology Review’s
flagship event at the Frontiers of Technology
|
|
|
|
|
|
|
Step inside the hub of innovation
|
|
|
Attend EmTech Future September 29-October 1 to explore the technologies transforming tomorrow and gain first-hand insight into how they are moving from breakthrough research into real-world impact.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Be at the forefront of emerging technology
|
|
|
|
|
|
|
|
|
Unlock essential reporting
|
|
|
From AI to climate change to biotech, learn how technology is shaping the future.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Download the MIT Technology Review app
 
|
|
|
|
|
|
|
|
|
|
|
|
|
|
