Benefits for Healthcare From AI Are Coming—Over Time

Artificial intelligence (AI) is in use all around us. Smart speakers like Alexa and Siri couldn’t function without it, nor could industries like finance, entertainment, and retail. The healthcare industry overall has yet to embrace AI and its myriad subtypes. Some institutions have implemented multiple AI-powered tools, whereas some have implemented none. Because machine learning thrives on large datasets, this patchwork approach limits the ability of AI applications to predict, analyze, or advise.

Yet healthcare’s late-to-the-game position may mean that healthcare can learn from other industries that have gone all in, says Erik Brynjolfsson, PhD, Professor and Senior Fellow at the Stanford Institute for Human-Centered AI, Director of the Stanford Digital Economy Lab, and a Research Associate at the National Bureau of Economic Research.*

General-Purpose Technologies

“Every once in a while, a technology comes along that literally bends the curve of human history,” says Dr. Brynjolfsson. The steam engine and electricity both contributed to automated production, which led eventually to more wealth and healthier lives for many people. Of the steam engine, says Dr. Brynjolfsson, “We are all 30 to 50 times richer than our ancestors largely because of that technological innovation.”

Such leaps forward, Dr. Brynjolfsson says, are described by economists as general-purpose technologies (GPT), and they are “pervasive, affecting many different parts of the economy. They rapidly improve. And perhaps most importantly, they spawn complementary innovations.”

However, society doesn’t benefit immediately from these life-changing technologies. It takes time, sometimes a generation, because everything from familiar ideas to physical workspaces must adapt to the innovation.

Potential Applications in Healthcare

Healthcare systems are steeped in data, which is the lifeblood of training AI systems. Healthcare systems also need help organizing their data for daily tasks like scheduling, billing, culling prescription data, and more—and in all these areas, AI could deliver. AI “can do prediction and classification,” Dr. Brynjolfsson says.

Dr. Brynjolfsson drives the point home to clinical leaders and healthcare executives: “Every one of your organizations has dozens, if not hundreds of opportunities to apply these machine-learning techniques.” Doing so will open doors for organizations and partnerships “to solve a whole set of problems that previously only humans could solve, or for that matter that previously couldn’t be solved at all.”

First, the Definitions

AI has expanded in sophistication and popularity. One hundred million users began using ChatGPT within two months of its release.

A few definitions to digest:

• Artificial intelligence is the so-called umbrella term for a computer system that can reason, sense, or act like a human.
• Machine learning uses algorithms to pinpoint patterns and relationships in the data; this system can make predictions or decisions without being explicitly programmed.
• Generative AI, a subtype of machine learning, can create images, language, text, and more.
• Large language models, a type of generative AI, are trained to predict the next word in a sequence; the data used to make this happen is derived from huge amounts of information coming from sites like Wikipedia or parts of the internet. ChatGPT is one example (with “GPT” here used in the computing sense to mean “generative pretrained transformer”).
• Deep learning, based on artificial neural networks, has been described as an attempt to mimic the human brain. Layers upon layers of digital information are fed into the system, allowing the linking of like information.

History Lessons

A general purpose technology makes a positive, permanent effect on daily life—but the challenging part is getting people to overhaul current concepts, infrastructure, and methodologies. “We see this pattern repeat itself over and over, that the technology advancement alone is not enough. It's rethinking your business process to take full advantage of it.”

When electricity became available, factories installed only one electric motor, because prior to electricity, the factories were powered by one steam engine. All production was clustered around that one power source. In time, electric motors were built in different sizes for different placements, thereby accommodating each production step and facilitating the assembly line. With that redesign, Dr. Brynjolfsson says, the gains in productivity were “massive—100 percent, 200 percent gains in productivity.”

Healthcare spaces may be ready for a parallel wave of advancements through new technologies that call for reimagining the work.

AI in Use

Dr. Brynjolfsson says generative AI systems are getting good at answering questions. The third version of GPT, he says, got a 5 percent score on the U.S. medical licensing exam; GPT-4 nailed a score of over 90 percent. He emphasizes, “This is not indicative of how one would perform in the clinic, but it is indicative of how broad and deep this capability is.”

When Dr. Brynjolfsson and others analyzed 950 job roles, considering which parts of those roles machine learning could perform, “We didn’t find a single one where machine learning or any combination of technologies just ran the table.” That said, lower-paid occupations like cashier were more likely to involve some tasks that could be performed through machine learning. But higher-paying professional positions like lawyer and investment banker could be more affected by a GPT, the researchers’ results showed.

At this moment, Dr. Brynjolfsson emphasized, the real benefit from AI comes from machine and human working together. Healthcare spaces have begun to benefit from AI-powered assistance with billing, referrals, prior authorizations, staffing predictions, and other administrative lifts or business insights. Is healthcare ready, on a larger scale, to reap the benefits of broader use of clinician decision support tools and other modes of machine-clinician collaboration?

AI systems today, Dr. Brynjolfsson says, come closer to capturing those unwritten, noncodable human bits of information that prior AI systems could not. He quoted a philosopher named Michael Polanyi, who articulated Polanyi’s Paradox: “We know more than we can tell.”

*Erik Brynjolfsson, PhD, spoke at the 2023 gathering of the TDC Group Executive Advisory Board Meeting in Napa, CA, which brings together top healthcare executives, academic researchers, and clinical leaders to discuss the changing landscape of healthcare as part of the Leading Voices in Healthcare initiative.