What’s eating healthcare AI? Not healthcare AI itself
When AI fails to thrive in healthcare, the problem is usually not with an algorithm. It’s with something deeper.
Three unrelated opinion pieces authored by subject matter experts hit on this theme just this week.
Consider an excerpt from each before following the links to the full articles.
1. AI in healthcare is like an iceberg viewed from an approaching ship.
There is a tendency to focus on the visible layer of AI—the algorithms, interfaces and outputs that capture attention. But those elements represent only a small portion of what determines success.
Below the surface is everything that makes AI viable in practice: data standardization, interoperability, governance, security and integration into clinical and operational workflows. When those elements are missing or underdeveloped, even the most advanced AI solutions struggle to deliver meaningful impact. Models trained on clean, curated datasets often encounter very different conditions when deployed in live environments. Inconsistent coding, incomplete records, and fragmented data sources can quickly degrade performance.
This is where many organizations hit a wall.
It’s not that the AI doesn’t work. It’s that the system around it isn’t ready.
—From “Healthcare doesn’t have an AI problem; it has a readiness problem” by HL7 International CEO Rachel Dunscombe (MedCity News, May 5)
2. Most health systems still lack IT components essential to AI.
Infrastructure is easy to overlook and underinvest in. Roads, bridges and electrical grids fade into the background—until a failure makes them impossible to ignore.
The same is becoming true of AI in healthcare. Over the past several years, health systems have developed or investigated hundreds of AI solutions. Many are showing strong technical performance but limited clinical or operational impact. Some have proved incompatible with existing systems; others couldn’t scale beyond pilots.
As a result, the landscape is now filled with isolated proofs of concept rather than sustained, system-wide tools that bring tangible and intangible value to the organization.
The shortfall isn’t in the tools themselves. It’s that most healthcare organizations lack the necessary infrastructure—such as reliable data architecture, governance, monitoring and workflow integration—that AI requires to operate safely and effectively at scale.
—From “AI won’t fix healthcare until we fix the infrastructure” by Stanford biomedical informatics instructor Justin Norden, MD, MBA, and colleagues (Healthcare IT News, May 5)
3. AI in healthcare does not fail because of weak models.
It fails because implementation does not align with requirements.
In fact, AI in healthcare is only useful when it is deployed as workflow, not just as a tool.
Whenever an AI tool is first introduced, and when an organization sees a demonstration of it, the feeling is one of excitement. Testing reveals amazing initial results, but the focus is rarely on the end goal that is to be achieved. The process is generally ‘in progress’—until the day comes when the technology meets the reality of everyday work. And when this happens, the output starts becoming inconsistent. It does not meet the requirements it was brought in for.
This is where trust drops—before the value of the technology is fully proven.
—From “Why AI in healthcare fails without workflow discipline” by business analyst Sai Teja Rayabarapu (The Hindu, May 3)
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.