The most peculiar thing about prescription bottles is how many of them remain empty. They are found in cabinets all over the nation, half-used, expired, and tucked away behind the vitamins. Insurance companies pay for them, doctors distribute them, pharmacists count them out, and then patients just stop for reasons that rarely fit on a clipboard. Nurses are aware of this. It has been measured for decades by researchers. And the price, both in terms of money and lives, is enormous.
The type of tool being used to address the issue is currently subtly changing. Reminders were the solution for years. The phone buzzed. The pharmacy texted me. Perhaps the pill itself has a sensor that broadcasts confirmation that the dose did indeed decrease. At least not in the way that anyone had hoped, none of it actually worked. Remembering is not a problem for people. They have conflicting feelings. They’re afraid. They’re occupied. They feel fine and don’t want to be reminded that they aren’t, or they don’t trust the medication or the diagnosis. A buzz doesn’t retaliate.
A more recent generation of research is taking an uncommon turn in that direction. Some teams are developing what computer scientists refer to as “argument-based digital companions”—AI systems that are intended to genuinely reason with a patient, weigh their motivations against their resistance, push back when necessary, and yield when not—instead of bothering users. The framework originates in the field of computational argumentation, which may seem uninteresting until you see what it aims to accomplish. It views a person’s justifications for missing a pill as valid points of contention that should be discussed rather than as unacceptable behavior that needs to be addressed.
This concept was investigated in a recent study from a Swedish research program called STAR-C, which worked with both common users and medical professionals to determine what people truly desired from such a companion. The results were more intriguing than what is typically found in user surveys. Not everyone wants the same coach. A gentle presence that nods along is what some seek. Some desire something more akin to a sparring partner, who is assertive, even provocative, and prepared to confront flimsy justifications. The researchers came to the conclusion that the intensity of the agent’s arguments and pushes might be more significant than the content of its statements.

This could be the piece that’s missing. For a long time, behavior-change applications have viewed their users as passive recipients of guidance. Consume this. Take this many walks. At eight, take this. However, anyone who has attempted to break a habit understands that the true dialogue takes place internally and is louder than any notification. There’s a feeling that an AI that can respond to that noise, effectively saying, “I hear why you don’t want to, but consider this,” could accomplish something that the buzzers could never.
The world of clinical trials is also keeping an eye on things. The CEO of a company called Cognivia, Dominique Demolle, recently argued that AI’s true potential in clinical trials lies in predicting, prior to enrollment, who is likely to quit and why, rather than monitoring whether patients swallow the medication. Stress, fear of side effects, personality traits, and health literacy are all combined into a model that allows researchers to customize participation rather than exclude individuals. Recruiting patients is challenging. It is costly and distorts the data to lose them in the middle. The entire industry’s economics could be altered by an AI that knows when to step up and when to back off.
Nothing has been resolved. The studies are small, the systems are early, and no one has fully mapped the ethical territory of an AI that argues with vulnerable people about their own health. Even so, it’s difficult to ignore the sense that something more sincere is at last being tried as the field moves from reminders to reasoning. The medicine cabinet hasn’t been altered in fifty years. The discussion about it may be about to begin.⁖※

