Artificial intelligence is rapidly reshaping health care: clinical decision support, documentation, operations, patient engagement. Most of the conversation about it focuses on the technology itself: the quality of the data, the sophistication of the models, the accuracy of the outputs.
These are real questions. But they can obscure a more basic one: Health care is a human system, and every outcome it produces depends on what people do. That includes patients, but also physicians, nurses, staff, and caregivers. An AI tool that doesn’t account for human behavior is solving only part of the problem.
Take two examples physicians will recognize. An AI system flags a patient as unlikely to complete a recommended screening. That’s useful information. But it doesn’t tell you why. Logistical barriers, competing priorities, fear, a bad past experience with the health system, or some combination that looks identical from the outside. It also doesn’t tell you what to do differently with that patient at the next visit.
Or: An AI documentation tool cuts your charting time in half. That’s a real win for burnout, right up until the system also starts generating more follow-up tasks, more messages to triage, more alerts to review, and the time you got back gets absorbed by different administrative work. The tool solved a technical problem. It didn’t necessarily solve a behavioral one.
Real physician voices, twice a week
Free, and one click to unsubscribe.
This is the gap behavioral science exists to close: not what is happening, but why, and what might actually change it.
Behavioral scientists have spent decades building frameworks for understanding motivation, decision-making, habit formation, self-efficacy, social influence, and emotional response: the mechanisms behind why people act, or don’t. Much of that work rests on three observations that sound almost too obvious to state, and yet are routinely ignored in how health care technology gets built.
- People are different from one another: The patient who skips a screening and the patient who skips a medication refill may look identical in a dataset and be responding to entirely different underlying causes. The same message, timed the same way, will not work equally well for both of them.
- People are shaped by context: The physician who counsels patients on preventive care all day may be behind on their own. A patient who manages one chronic condition well may struggle with a second, layered on top of the first and competing for the same limited attention and energy. Behavior is not a fixed trait; it shifts with role, setting, and competing demands.
- People change over time: A diagnosis, a hospitalization, or a difficult conversation can make one patient more engaged and another more avoidant of follow-up care entirely. What worked for someone six months ago may not work now.
Health care technology has historically treated this variability as noise: something to standardize away because responding to it at scale wasn’t practical. A clinician can tailor a conversation to one patient in the room. Doing that consistently across a panel of thousands has never been feasible by hand.
This is precisely where AI’s real value lies: not simply in identifying problems faster, but in responding to that variability at a scale humans can’t. Done well, AI can help identify patterns associated with different barriers and tailor outreach based on what appears most likely to be effective for a given patient. Without a behavioral model underneath the hood, AI just produces faster, more confident-looking guesses.
That distinction matters because every AI system that tries to influence behavior, whether a patient’s, a physician’s, or an organization’s, is already making assumptions about how people think and act, whether or not anyone designed those assumptions on purpose. The question isn’t whether a behavioral model is present. It’s whether that model was built deliberately and informed by evidence, or whether it’s an accident of whatever data happened to be available.
Too much of health care AI gets framed as a technology initiative first. In practice, most of these systems are behavior-change initiatives: They exist to alter a decision, an action, a workflow, an interaction. Achieving that requires more than a well-trained model. It requires a working theory of why the person on the other end behaves the way they do. Physicians, more than most, already know how often that “why” gets flattened by systems built for speed and scale instead of understanding.
None of this is an argument against AI in health care, or a case for going slower. It’s an argument for what AI needs alongside it. Computational intelligence gives us the ability to process information, personalize interactions, and operate at a scale no clinician could manage alone. Behavioral intelligence gives that capability direction: a way of knowing which lever to pull, for which person, and when.
For physicians, that combination should be reassuring rather than threatening. It doesn’t replace clinical judgment or the relationship with a patient. Used well, it clears away some of the friction and guesswork so more of a physician’s attention goes toward the parts of care that actually require a human: the diagnosis, the conversation, the judgment call.
Health care has made extraordinary progress in diagnosing disease, predicting risk, and personalizing treatment. The harder problem left is getting people to act on what we now know. Solving it will take more than better models. It will take taking human behavior as seriously as we take the technology built to serve it. The most successful AI won’t be the systems that understand data best. It will be the systems that understand people best.
Amy Bucher is a health care executive.
