The Real Reason AI Stalls in Healthcare Has Nothing to Do with the Technology

Everyone assumes AI adoption in healthcare slows down because the tools aren't good enough yet. Fix the accuracy. Improve the interface. Add more training data. Then people will trust it.

That's not what the evidence shows.

A 2023 Nature Medicine study found that even when AI diagnostic tools outperformed clinicians in accuracy, adoption rates stayed low. Researchers gave it a name: algorithm aversion. See one AI mistake, and the distrust sticks, even if the human error rate is higher. That's not a technology problem. That's a psychology problem.

And it runs in both directions. The WHO flagged this in 2021, what they called automation bias. Some clinicians over-trust AI outputs and stop questioning them. Others reject AI reflexively. Neither response is driven by evidence. Both are driven by how we're wired.

A 2023 Wolters Kluwer survey put numbers to the skepticism. Sixty-one percent of physicians worried AI would make clinical errors. Only 38% felt confident they could even interpret AI-generated recommendations. That's not a training gap you close with a product demo. That's a trust gap built over decades of how medicine has taught people to think about authority and error.

Here's the part worth sitting with: the barrier isn't confidence in AI. It's confidence in themselves while using AI.

NEJM Catalyst found something useful here. Patient willingness to accept AI-assisted diagnoses jumped from 38% to 62% when a human clinician was described as overseeing the AI. Same tool. Same output. Different framing. The technology didn't change. The story around it did.

That should matter to everyone working in HR and benefits, not just health systems. We're in the same trust landscape. The people we serve...employees, plan members, HR admins are skeptical of automated tools for the same reasons clinicians are. They've seen a chatbot give bad information. They've watched a system fail a colleague. One bad experience overwrites a hundred good ones.

So the question isn't how do we build better AI. The question is how do we build better context around AI.

That means being honest about what the tool does and doesn't do. It means keeping a human in the loop, not as a disclaimer, but as a real part of the design. It means treating transparency as a feature, not an afterthought.

My shovel analogy holds true here. A good shovel doesn't replace the person digging. It just means the work that shouldn't take this long, doesn't. But if the person holding the shovel doesn't trust it, doesn't understand it, or got hit with it once...they're going to use their hands instead.

That's where most AI adoption efforts get stuck. Not in the algorithm. In the room where the person decides whether to pick up the tool at all.

What's your experience been, when skepticism about AI tools comes up with your team or your employees, what's actually underneath it?