volume 37  |  No. 2  |  2026

Why Before How: Person-centered Care In the Age of Artificial Intelligence

Why this diagnosis? Why this treatment plan? Why this trade-off for this patient? Healthcare is navigating a crisis of complexity.¹ The volume of medical and dental knowledge is expanding faster than any clinician can reasonably process, while our time and attention remain fixed. In that environment, AI will likely become an important tool for retrieving information, organizing clinical context and standardizing routine tasks. Used well, it can help us see more clearly.

Used poorly, it can simply help us move faster.

That distinction matters. Our training and compensation systems already reward the “how” of care: procedures, production, speed and throughput. If we introduce AI mainly as an engine for efficiency, we should expect a predictable result. We will accelerate care delivery without necessarily improving the quality of clinical judgment behind it.

For implant clinicians, the “why” is not a philosophical extra. It is the foundation of care. It is the reasoning that comes before intervention: what we believe to be true about the diagnosis and prognosis, what evidence might change our minds and what matters most to the person in the chair.

Confidence Is Not the Same as Accuracy
One reason this matters so much is that clinicians are not always well calibrated to their own diagnostic accuracy. Studies have shown that physicians’ confidence does not reliably rise and fall with actual correctness or with difficulty. In more complex cases, accuracy may decline without a corresponding drop in confidence.²

That should give all of us pause.

When we introduce modern AI into this dynamic – i.e., systems that are inherently fluent, persuasive and outwardly confident – the risks compound. The combination of human overconfidence and machine fluency makes it dangerously easy to act decisively on a clinical narrative simply because it feels coherent rather than because it is rigorously supported. The solution is not to become more certain. It is to become more disciplined about how certainty is earned.

AI should help us see better, not simply decide faster.

Not All Clinical Variation Is Bad
One of the most difficult realities in healthcare is that different clinicians may recommend different paths for the same patient. From the patient’s perspective, those differences can easily look like inconsistency or incompetence. But not all variation means the same thing.

Some variation is unwarranted. It reflects diagnostic noise: missing data, incomplete records, inconsistent documentation, unstable disease labels or poor information transfer. That kind of variation creates avoidable harm and should be reduced.³

Some variation is warranted. It reflects the fact that patients value different things.

One patient may prioritize fewer visits. Another may prioritize lower cost. Another may prefer the least invasive option, even if it brings compromises in durability or maintenance. Another may accept a higher burden to maximize long-term stability. Those are not failures of the system. They are the essence of person-centered care.

AI will predictably affect both. While it can reduce noise by retrieving missing context and standardizing evaluations, it also threatens to flatten warranted variation. Without deliberate design, AI will mathematically push diverse patient needs toward a single “optimal” plan, particularly when the system’s underlying objective function remains implicit.

Three AI Challenges to Anticipate
To prevent this flattening and preserve the space for individualized care, our integration of AI must carefully navigate three pertinent challenges:

1. Data Mismatch (Representation and Dataset Shift): Algorithms inherit the world in which they were trained, and datasets do not travel seamlessly. Differences in imaging devices, acquisition protocols, documentation habits, and patient demographics can severely degrade a model’s performance outside its original setting.⁴ Even populations that look similar structurally (demographics, education level, etc.) can be very different culturally. Clinicians must be mindful of these potential differences when applying model suggestions to all patient types. A heuristic that can help mitigate the concern of overgeneralization is shown in Figure 1.
2. Objective Mismatch (The “Missing Value”): AI models optimize mathematical metrics, but patients live inside complex human values. As Kohane and Manrai note, medical AI can be “perfectly right” regarding a specific metric while entirely misaligned with what the patient cares about.⁵ The antidote is to make the model’s objective explicitly clear: we must label what is being optimized (e.g., durability, success rates, low burden, cost efficiency) and clearly state what is being traded away to achieve it.
3. Cognitive Mismatch (Automation Bias and Drift): There is a common assumption that combining a human with AI will automatically produce a better decision. That is not always true. In fact, recent evidence suggests that human-AI combinations can underperform the best human or the best AI acting alone, especially in decision-making tasks.⁶ The mechanism here is a familiar one: automation bias, or over-reliance on an authoritative-seeming tool, especially under the pressure of a busy clinic.

What Good Use Looks Like
AI has real potential in implant care. It can help us retrieve information, reduce inconsistency, and create more reliable systems. Those are meaningful advances.

But for AI to earn its place in dentistry, it must do more than impress us. It has to help secure diagnostic truth, clarify tradeoffs, and preserve room for individualized care. It has to support decisions that are biologically sound, evidence-based, and aligned with the goals of the person sitting in the chair.

That means the clinician’s role does not become less important. It becomes more important.

Our task is not to resist new tools, nor is it to surrender judgment to them. Our task is to make sure these tools serve the right end. In implant dentistry, the “why” must remain primary: why this diagnosis, why this intervention, and why this plan for this patient.

AI is fully capable of thinking without caring. Our responsibility is to make sure that, in adopting these tools, we do not learn to care without thinking.

References for this article can be seen here.