For more than a decade, Big Tech companies and startups have heralded AI’s potential to solve healthcare’s problems. However, the remedy for the industry’s pressing workforce shortage and financial challenges proves to be far more intricate than merely embracing the allure of shiny new technologies.

As hospitals across the nation adopt more and more AI technology, there are some common hazards of which they should be wary, according to healthcare AI expert Suchi Saria, a machine learning researcher at Johns Hopkins and CEO of Bayesian Health, a clinical AI platform she founded in 2021. 

On Tuesday at Oliver Wyman’s Health Innovation Summit in Chicago, Saria laid out seven AI pitfalls hospitals should look out for.

Allowing Big Tech marketing campaigns to dictate your strategy

Saria has seen many hospitals succumb to vendors’ deceptive portrayals regarding the capabilities of their AI technology, she said in an interview the day after her presentation. In her view, hospitals need to have a resolute AI strategy of their own instead of latching on to these companies’ AI strategies. 

“Healthcare is a business where health systems don’t want to originate a strategy because they’re risk averse. They want to be part of a consensus group with some strategy,” Saria declared.

In the midst of the generative AI hype cycle that has taken the digital health world by storm, people seem to be grouping every piece of AI under the umbrella of generative AI, she pointed out. With this massive emphasis on generative AI, hospitals across the country are eager to adopt this technology — but not all of them understand why they should be implementing these tools. Before they go all in on generative AI tools, hospitals need to dig deeper into understanding why the recent advancements in generative AI are important to the healthcare field and what types of new problems this technology can solve, Saria stated.

Assuming AI isn’t ready for deployment at scale

Some more old-school healthcare stakeholders are still a bit wary of AI’s deployment in healthcare settings, but Saria asserted that most healthcare AI is ready for hospitals to use — with one big caveat.

“Doing it right is the caveat. AI done right is ready for primetime. AI done right means understanding the problem and the value chain you’re delivering, that the solution is designed with a deep understanding of the problem in mind, and you have metrics and rubrics for measuring performance, safety and bias,” she explained.

Doing it all by yourself

In order for healthcare AI deployments to be successful, the project team needs to have personnel with deep expertise on both the technical and clinical sides of things, Saria noted.

“IBM Watson is an excellent example. It had amazing technology — proven in the context of Jeopardy! — but extremely little experience in healthcare. It partnered with some of the leading systems in healthcare, like MSKCC and MD Anderson, but these partnerships didn’t get anywhere because the two sides were too far apart,” she said.

Confusing AI with automation software

Some health system leaders are still underestimating the power of AI, Saria argued. She denounced the notion some people may have that AI models are simply just pieces of software that can automate some of healthcare workers’ tasks.

Saria said part of the reason she loves AI is because it can truly act as clinicians’ copilot. When developed well, AI models can work alongside clinicians and cause them to interact with data in completely new ways, she explained.

Sitting idle and waiting for the perfect plan

If health systems sit around and wait for the perfect moment to start a new AI project, they will never make any progress, Saria warned. The hospital world is facing immense pressures — from a sweeping labor shortage to declining patient volumes to tight financial margins — and leaders need to abandon some of their risk aversion to solve these problems, she declared. 

She recommended that hospitals identify opportunities for quick wins while designing their future workflows with AI in mind.

Underinvesting in learning by doing

Sometimes hospitals underestimate how powerful of an asset their workforce is, Saria said. She recommended that health systems involve their staff members in the AI implementation process as early as possible.

“Even if I found the perfect plan, I cannot execute on this plan until my people are ready and my people are part of the solution,” she stated. “You might have to rely on your clinicians to give you feedback. That way, you can have a smoother rollout once they have used the technology a little bit and learned by having it in their daily workflows.”

Separating clinical and administrative

Many hospitals are interested in the AI tools sold by Big Tech companies, and these firms often tell providers to start out by using their technology in administrative use cases rather than clinical ones, Saria pointed out. This is seen as less risky by Big Tech companies, which don’t always have deep clinical expertise, she explained.

However, providers can’t solve their biggest problems if they focus their AI strategy on administrative use cases — after all, the core mission of healthcare is care delivery, Saria said.

“David Cook, who [is SVP United Health Group Strategy and Innovation], yesterday said to me that ultimately, administrative is only 10% of costs. We can do all the work in the world and the maximum bang we can get is up to 10%. We have to figure out ways to take costs out, streamline, improve and make progress on the remaining 90%,” she declared.

Correction: An earlier version quoted Saria incorrectly. She spoke to David Cook and not David Britchkow.