Artificial Intelligence: 5 Considerations for Health Systems

Following are 5 takeaways from our conversation.

1. Many Health System Leaders Don’t Understand AI

While AI has become an industry buzz word, “people don’t really know what it means,” Ziemianski says. “I would say there’s a lack of education in this area.”

“I think the best definition of artificial intelligence is the capability for machines to replicate some or all human intelligence capabilities,” Amarasingham says, yet there are many reasons for confusion. “Because the definition of artificial intelligence is so broad in popular imagination, as well as what’s available, it does create misconceptions with healthcare leaders.”

The most advanced forms of AI in development are often referred to as artificial generalized intelligence, he says. This form of AI mimics the full scope of human thought and enables a computer to interact with the world and not be perceived as a machine, a phenomenon known as the Turing Test.

“That capability is not available in medicine,” Amarasingham says. “In healthcare, the use of AI is much narrower. It’s focused on building on statistical and computer science tools that we’ve had for quite a while now—one or two decades of machine learning, deep learning, and neural networks.” Capabilities in medicine focus on “relatively narrow tasks,” such as interpreting a chest x-ray or determining the right medication. Yet the potential is expansive.

“It’s not an exaggeration to say that the promise of AI in medicine is immense, and that we’re at the earliest stages of it,” Amarasingham says.

2. AI Can Harness Vast Quantities of Data Generated in Healthcare

Healthcare is generating massive quantities of data that need to be accessed and analyzed, Vygantas says. Machine learning and other computational tools, including artificial intelligence, provide the resources to “tackle the opportunity in front of us,” he says.

“The pace of progress over the last three to four decades has been tremendous, starting with the genomic revolution, followed by the proteomic revolution, right into data interoperability driven by [electronic] health record adoption,” Vygantas continues. “There has been an explosion in the amount of information [available]—from the most granular molecular level, all the way up to population health management. AI allows us to harness the insights that would otherwise be nearly impossible or very limited in scale on a human level.”

3. AI Can Reduce Cognitive Burden and Provider Burnout

Another benefit AI can deliver is the potential to reduce cognitive burden and diminish provider workloads, Amarasingham says.

“Right now in medicine, there’s friction in every aspect of the healthcare delivery process,” he says. Time-consuming tasks related to documentation take away from caregiving. “I see vast potential for AI to take on many or most of those tasks and allow more space for the doctor‒patient relationship. That’s going to reduce burnout and reduce the burden on our healthcare professionals, while also improving the outcomes [by ensuring] that things are getting done correctly. That’s the biggest area of improvement we’ll see over the next decade—and just making medicine much more joyful, too.”

4. Reliance on AI Could Diminish Clinician’s Cognitive Abilities

“One area of concern among AI researchers and individuals building AI tools and products is what I would call the ‘GPS effect,’ ” Amarasingham says. Drivers who rely on using GPS (global positioning systems) for directions while driving lose their own cognitive ability to navigate “because your mind’s not actively trying to work out the geospatial coordinates and landmarks,” he says. “The GPS system sort of takes over that requirement for your brain, and then you remember less, and you learn less.”

Some fear a similar phenomenon could occur in the practice of medicine as providers become more reliant on AI tools.

“As AI gets more advanced in medicine, is it possible that physicians, nurses, and others will become too reliant on the AI, and not learn the territory, so to speak, for themselves? How do we properly protect that? Because machines will have errors,” Amarasingham says.

Problems may arise with statistical errors, as well as issues with conflicting data or nuances in how data is applied. “If you’re not exercising those mental faculties all the time, you could default to what the machine [recommends],” he says. “That’s going to be a serious challenge going forward in every industry, but particularly cognitive industries like medicine.”

Ziemianski says that this issue has already occurred in the airline industry. “Pilots have become too reliant on autopilot, and they started to lose their skills.” To address this problem, the industry is “putting programs in place where pilots have to physically do a certain quota of landings and take offs … to keep their skills sharp,” he says. “The same thing’s going to happen in healthcare if we become too reliant on what the machine tells us.”

5. AI has Already Made Significant Inroads Into Healthcare

Children’s Health launched numerous AI initiatives before the pandemic, but COVID-19 spurred additional uses, Ziemianski says. Among the ways the Dallas hospital is using AI:

• AI is being used to maintain personal protective equipment (PPE) supplies. The system examines five years of historical data related to PPE usage during surgeries and ambulatory visits to forecast against current inventory.
• The technology is being used to analyze the number of patients in the emergency department and inform inpatient floor staff how many admissions to expect based on historical data related to acuity level and chief complaint.
• AI solutions are also being used to address sepsis and asthma, as well as to assess readmission risks and excess length of stay.

At OSF HealthCare, Vygantas says the technology is being used to drive efficiencies in multiple areas, including the operating room and in monitoring blood loss. In addition, a radiology AI solution helps radiologists more efficiently process CT lung scans. The health system also employs AI to better manage sepsis, and it is using other clinical support tools aimed at improving outcomes and avoiding adverse effects.