Beyond Decision Support: The Case for Collaborative Clinical Intelligence
By Maka Tsulukidze, MD, PhD, Associate Professor | Health Services Researcher | AI Strategy & Leadership in Higher Education & Healthcare, Florida Gulf Coast University
Much of the current conversation about Artificial Intelligence in Healthcare centers on trustworthy AI. That goal is essential. Intelligent systems used in medicine should be safe, reliable, transparent, interpretable, and accountable. But trustworthy AI, by itself, is not enough. Explainable models do not necessarily produce explainable clinical reasoning.
A system may reveal how it generated a prediction while leaving the clinical judgment around it largely invisible. A clinician may see which variables influenced an algorithm without seeing how evidence, professional standards, contextual knowledge, uncertainty, and patient priorities were weighed to justify a course of action.
The challenge, therefore, is larger than making intelligent systems explainable. Healthcare also needs trustworthy reasoning.
Clinical reasoning is more than an answer
Two physicians reviewing the same history, labs, imaging, and vitals for one patient may recommend different next steps. This happens every day in medicine.
The difference is not necessarily that one physician has better information or is clearly right while the other is wrong. Clinical reasoning is not the mechanical application of medical knowledge. It requires recognizing what matters, distinguishing signal from noise, weighing competing risks, interpreting incomplete evidence, adapting as new information emerges, and accepting responsibility under uncertainty. When clinicians disagree, the revealing question is often not which conclusion is correct, but why their reasoning diverged.
Medicine has always accommodated thoughtful disagreement because patients, evidence, and clinical contexts rarely fit standardized answers. The problem is not disagreement. The problem is disagreement without inspectable reasoning.
From decision support to reasoning support
Most intelligent systems in healthcare are designed around answer production: predictive models estimate risk, algorithms classify disease, language models summarize records. The shared assumption is that AI’s role is to produce a better answer.
A more consequential role is to strengthen the reasoning process through which an answer is reached.
Intelligent systems should not compete with clinicians to produce better answers. They should help clinicians produce better reasoning.
That is the difference between decision support and reasoning support.
Decision support asks: What should we do?
Reasoning support asks: Why does this course of action make sense, given the evidence, standards, context, and uncertainty involved?
The shift can be summarized this way:
Rethinking Healthcare AI
| Traditional Decision-Focused Perspective | Proposed Reasoning-Centered Perspective |
| Can AI make better decisions? | Can intelligent systems improve clinical reasoning? |
| Is AI explainable? | Is the reasoning process inspectable? |
| Can AI replace clinicians? | How should intelligent systems participate in professional reasoning? |
| Decision support | Reasoning support |
| Answer generation | Collaborative reasoning |
| Agreement | Explainable disagreement |
| Automation | Strengthening clinical judgment |
But trustworthy AI, by itself, is not enough. Explainable models do not necessarily produce explainable clinical reasoning.
This reframing does not move attention away from safety, explainability, or interpretability, which remain indispensable. It expands the focus to the reasoning process: whether it is visible, inspectable, challengeable, and open to revision.
Trustworthy AI is therefore one part of a larger objective: trustworthy clinical reasoning.
Artificial intelligence as a participant in professional discourse
Medicine advances through professional discourse: rounds, case conferences, and peer review that expose assumptions, compare interpretations, test arguments, and reconcile evidence.
If the goal is not simply better answers but better reasoning, then AI should support that discourse, not stand outside it.
The future role of intelligent systems may therefore be less about becoming another diagnostician and more about being designed into professional discourse.
That does not mean pretending an intelligent system thinks or bears responsibility like a clinician. It means designing systems that contribute distinctive capabilities to shared reasoning.
Collaborative Clinical Intelligence
This leads to a different vision for the relationship between clinicians and intelligent systems: Collaborative Clinical Intelligence, or CCI.
CCI is not the addition of human intelligence and artificial intelligence. It is the quality of clinical reasoning that emerges when clinicians and intelligent systems contribute complementary capabilities to a shared process grounded in evidence, while accountability remains with human experts.
The unit of analysis is therefore not the clinician, the intelligent system, or even the clinical decision itself. It is the reasoning process.
That distinction matters. A clinician using an algorithm has not necessarily achieved CCI, nor has a system achieved it merely because its recommendation agrees with the clinician’s judgment.
CCI emerges when the interaction makes reasoning more transparent, defensible, context-sensitive, and capable of improvement. It should help participants see not only what conclusion was reached, but how evidence was interpreted, where uncertainty remains, why reasonable disagreement exists, and what might justify revision.
Making judgment inspectable
One architectural approach to enabling CCI is computable judgment: making relevant elements of expertise, standards, evidence, and decision logic computationally explicit while preserving their relationship to clinical context.
The purpose is not to reduce medicine to rigid rules or transfer authority to software or suggest that every dimension of clinical judgment can or should be formalized. It is to make enough of the reasoning process explicit that it can be examined through questions such as: How was evidence interpreted? Which standards were applied? What assumptions shaped the recommendation? Where did professional judgment enter? What uncertainty remains?
In resuscitation training, for example, an AI-supported system could make visible not only whether a learner followed an ACLS protocol, but how evidence, timing, sequence, and clinical priorities shaped the assessment of performance.
When these elements become inspectable, intelligent systems can do more than generate outputs. They can contribute to reasoning that clinicians understand, challenge, teach, refine, and defend.
Nothing in this vision depends on today’s large language models, tomorrow’s algorithms, or any particular technical architecture. Technologies will change. The underlying challenge will remain.
Computable judgment, however, carries its own risks. Formalizing part of the reasoning process can create a false sense of rigor, encouraging clinicians to trust a trail of logic because it is visible, not because it is sound. It can also turn reasoning into a checklist. These risks do not argue against inspectability; they argue for treating it as a starting point for scrutiny, not a substitute for judgment.
Preserving medicine
Medicine has never been defined simply by the answers clinicians reach. It is defined by the reasoning they can explain, the responsibility they accept, the uncertainty they acknowledge, and their willingness to revise their thinking when evidence changes.
Instead of weakening those attributes, intelligent systems should enhance them. The purpose is not to automate medicine until the clinician becomes peripheral, but to create conditions under which better clinical reasoning can emerge. Earlier technologies often consumed clinician attention; AI, if designed well, could help restore space for deliberation and humane care.
That possibility is not yet assured. The task now is to design intelligent systems not merely to generate clinical outputs, but to support reasoning that is more transparent, inspectable, and trustworthy.
Preserving medicine remains a human responsibility. AI can help by strengthening the reasoning through which care remains responsible, explainable, and human.
Author Bio:
Maka Tsulukidze, MD, PhD, is an Associate Professor in the Department of Health Sciences at Florida Gulf Coast University. Her work focuses on healthcare innovation, artificial intelligence in healthcare and higher education, and the development of intelligent systems that support transparency, accountability, and human-centered practice.