AI Transforms Healthcare—But Not In Ways You Expect  

Imagine this: It’s time for your annual check-up, so you take photos of your eyes and the weird skin rash along your arms. AI imaging diagnostics check your retina for biomarkers of retinal disease1 and confirm the dry, scaly skin is mostly likely eczema—a detail that’s saved to your patient file. 

Your appointment begins with an annual mammogram before your doctor enters and asks about your recent eczema flare-up. They listen attentively as you discuss a recent change in your sleep pattern—eyes glued on you rather than a computer screen. Before your appointment ends, your doctor reviews the AI reading of your mammogram to confirm your x-ray is clear2, giving you a result within an hour rather than weeks. 

Once you leave, your doctor can turn to AI to summarize your patient notes and highlight any nuances they might have missed. Instead, they’ll spend extra time analyzing your responses to check-in questions and personalizing your treatment plan, including suggesting a special diet for eczema and topical creams.

Discussions about the healthcare system typically revolve around failing patients—but we often forget that it’s failing our healthcare practitioners, too. 

Burnout rates have steadily increased among physicians over the last decade. In 2024, more than half of doctors reported burnout in the Medscape Physician Burnout and Depression Report—and 83% of those recipients said their burnout was a direct result of their job. What’s more, women healthcare professionals were even more likely to deal with burnout.

Brian Ferguson, the founder of healthcare technology company Arena Labs, knows the stats well. His company provides data-driven solutions to burnout by offering training tools and tactics to clinicians on the frontlines.

Ferguson, whose own mother was a nurse, spent his career in high-performance organizations. He saw how elite athletes, creatives, and people in the military were all taught about human factors (i.e. understanding and managing the limitations of people).

"They're taught the connection between hydration and cognition, sleep, and the ability to focus. These very basic elements of our innate physiology," says Ferguson. Yet healthcare is one of the last industries to account for human factors.

"No part of nursing school or medical school trains clinicians how to really manage their nervous system, rest, recover, [or] balance energy," he adds.

Not only does this burnout cost the healthcare system more than $260 million annually, but it also puts patients at risk. Physicians dealing with burnout are twice as likely to be involved in a patient safety accident.

Many people may think of AI as a futuristic robot performing open heart surgery or a complicated spinal reconstruction (and yes, these do exist!) but the goal of integrating artificial intelligence into the health space isn’t to replace doctors. It's to help them.

"In the near term, healthcare is going to continue to be a human system," Ferguson says. "No matter how advanced people are able to get on the tech enablement side and the patient care side, there's still a human who is stewarding another human in health. Even the best AI cannot account for the need to help that human, the clinician, be their best self."

So what does that mean for AI? The smallest amounts of support could make the biggest difference. For example, doctors spend 15.5 hours per week on average on administrative tasks—and this tedious-yet-essential task became the first focus of many AI startups in the healthcare space.

Companies like Freed AI, a HIPPA-compliant AI scribe, launched in 2022 with the goal of reducing physician’s paperwork. Its tech claims to reduce doctors’ documentation time by 95%—and this dramatic drop is reflected in the rollout of AI scribes at large medical networks like Permanent Medical Group. The group rolled out AI scribes to more than 10,000 physicians in October of 2023—by April 2024, it reported AI scribes saved doctors within the group at least one hour of work per day. 

Beyond saving physicians essential time, AI also opens up the door to speedier diagnosis—which indirectly could help save lives. An estimated 795,000 patients per year die or are permanently disabled from a misdiagnosis. While there’s no guarantee AI can reverse these statistics, there’s promising evidence that it can help. 

An early assessment of ChatGPT in the Journal Of Medical Internet Research found the Large Language Model (LLM) had a 76.9% accuracy rate when diagnosing patients3. (Large Language Models are a subcategory of deep learning machines specifically that process and generate human language.)

Further research supported this claim, showing AI was almost twice as accurate as physicians in making a correct diagnosis (59.1 versus 33.6%). What’s more, early research from 2019 suggested AI could be an optimal way to help diagnose Acute Kidney Injuries (AKI) up to 48 hours earlier4, a challenging feat that often prevents patients from receiving necessary care. 

Yet, for all the excitement that may exist around AI in the medical space, there are still limitations—namely, the physicians themselves. A small study published in JAMA Network Open last month uncovered LLM didn’t actually make a significant difference5 in assisting physicians compared to conventional resources (even though the LLM chatbot outperformed physicians significantly when diagnosing solo). Physicians were often hesitant when an LLM offered a differing opinion from their own and many didn’t understand the correct way to prompt AI. 

These findings contradicted the hypotheses of researchers, including Jonathan H. Chen, M.D., Ph.D., an assistant professor at Stanford Center for Biomedical Informatics Research. The team expected the chatbot AI systems to perform well based on pilot studies—and assumed doctors empowered with a similar tool should perform better.

Even though LLMs tend to be at the center of discussion in diagnosing patients, they’re not the only tools available to physicians and nurse practitioners. Deep learning systems are also a key part of the discussion. 

Deep learning systems refer to learning machines that learn from data using artificial neural networks. These AI systems can detect anemia from retinal images6 and reveal signs of diabetic retinal disease1, as well as elevated glycated hemoglobin (an indicator of high blood glucose levels and increased risk of diabetes-related health issues). 

In fact, a deep learning model even exists to help rapidly and accurately read mammograms. And yes, it demonstrated a similar assessment to trained radiologists when reading the full-field digital mammographic images.

Improving diagnostics is the immediate focus of most AI research, but the countless use cases for artificial intelligence ensure it’s far from the end. AI is also expected to play a pivotal role in transforming radiation therapy. 

A 2023 review published in the Journal of Radiation Research found AI could help drastically reduce the time needed to create a cancer treatment plan from several days to just minutes, if not seconds. 

The impact of AI also goes beyond how we receive treatments to the creation of the treatments themselves. The AI-driven biotechnology company Insilico Medicine, valued at $895 million, is currently undergoing human trials for the first-ever drug developed using generative AI technology. 

As we look towards 2025, we’ll continue to understand the ways AI can help with some of the biggest complications in the healthcare space: misdiagnosis, burnout, and general research. Yet it’ll also become clear how much training is necessary to ensure the AI available at our fingertips is effectively utilized.

Tech skill shortages already plague the healthcare industry. As AI investment increases from $20 billion in 2024 to nearly $150 billion within the next five years, training a skilled workforce who can utilize AI to improve documentation, communication, and workflows will be a necessity for the healthcare industry. 

"Healthcare in many ways is a legacy organization. It's tech-saturated, so adopting new technology is a lot of work, and a lot of times that work isn't done well," Ferguson explains.

He points to the introduction of the Da Vinci Robotic Surgical Systems, which allows for a minimally invasive surgical approach. When introduced to the operating room, many doctors complained the technology was bad or didn't work well. The robotic systems were simply added into the existing OR setup, when they actually required a completely different setup to thrive.

To ensure AI can help clinician success, it's important for the technology to be properly implemented (or we defeat the purpose).

"If I'm using AI to take my notes, that means my workflows have to change, and I've got to make room to take that automation and somehow bring it back in the loop. That usually takes more work than is accounted for, and if that's not done well, it puts more stress on the human system."

In the coming years, healthcare practitioners will need to learn how to use AI—something Chen likens to the introduction of the Internet.

"While we take for granted daily online activities like searching the internet, browsing through articles, and submitting online transactions, these are all acquired skills that we learned how to do (and our former selves 35 years ago would not have known how to)," says Chen. "Similarly, I expect we will all need to learn new skills in how to interact with chatbot AI systems to nudge and negotiate them to behave in the ways we wish."

So what does that mean for the aforementioned AI-infused day at the doctor? Well, we're likely a far way off before it becomes a reality. For now, Ferguson argues: "The best role of artificial intelligence is to make the human more human."