Don’t Trust the Doctor. Don’t Trust the AI. Examine the Reasoning.
The real question in AI-assisted medicine is not who has authority. It is whether a medical conclusion can be examined, challenged, and justified.
A provocative debate is emerging around artificial intelligence and medicine.
U.S. Health Secretary Robert F. Kennedy Jr. recently argued that artificial intelligence could provide patients with a second opinion “much better informed than any doctor in the country” and could help free people from what he called “medical tyranny.”,
That statement naturally provokes a question:
Should we trust the doctor, or should we trust the AI?
I believe that is the wrong question.
The problem is not whether AI or the physician should be trusted.
The problem is whether the reasoning behind a medical decision can be examined, challenged, and justified.
The Problem With “Trust”
For decades, medicine has depended significantly on professional expertise.
A patient visits a physician because the physician possesses knowledge and experience that the patient normally does not.
Artificial intelligence changes this relationship.
AI systems can analyze large and complex datasets and generate information that may support diagnosis, disease detection and treatment decisions. The U.S. Food and Drug Administration explicitly recognizes these possibilities in describing AI-enabled medical devices.
But an important distinction must be made.
Access to more information does not automatically produce better judgment.
We should distinguish:
Information ≠ Evidence ≠ Reasoning ≠ Judgment ≠ Authority
A system may retrieve thousands of papers and still reach a poorly supported conclusion.
A physician may possess decades of experience and still make an incorrect inference.
Neither computational power nor professional credentials should make reasoning immune from examination.
From Authority to Justification
Imagine that a physician recommends Treatment A.
An AI system recommends Treatment B.
Who should the patient trust?
Simply asking which one is more intelligent or better informed does not solve the problem.
Instead, we should ask both:
What evidence supports the conclusion?
Which assumptions were made?
Which information contradicts the conclusion?
How was uncertainty evaluated?
What alternative explanations were considered?
Can the reasoning be challenged?
And finally:
Who is responsible for the decision?
These questions transform the problem.
The contest is no longer:
Doctor vs. Artificial Intelligence.
It becomes:
Claim vs. Evidence.
The AI Must Be Challengeable
This point is especially important.
An AI that simply produces an answer can become another authority.
Instead of:
“Trust your doctor.”
we risk replacing it with:
“Trust the algorithm.”
That would not represent epistemic progress.
It would simply replace one potentially opaque authority with another.
A genuinely useful medical AI should therefore do more than produce recommendations.
Ideally, its output should allow the relevant human decision-maker to examine a chain such as:
Patient information → Evidence → Interpretation → Inference → Alternatives → Uncertainty → Recommendation
This does not mean that every AI model currently exposes its internal computational processes. It means that an AI-assisted clinical architecture should provide enough decision-level evidence, assumptions, uncertainty and provenance for its recommendation to be critically evaluated.
This distinction matters enormously.
Explainability Is Not Enough
There is another subtle issue.
An explanation can sound convincing without being correct.
Therefore, we should not ask merely:
“Can the AI explain its answer?”
We should ask:
“Can the justification be independently examined against the evidence?”
That is a much higher standard.
A plausible narrative is not evidence.
Confidence is not evidence.
Authority is not evidence.
And consensus, by itself, does not explain causation.
The central principle should therefore be:
A medical recommendation should become more trustworthy because its justification survives examination, not because of who or what produced it.
Interestingly, Medicine Is Already Moving in This Direction
Get José. T Guevara’s stories in your inbox
Join Medium for free to get updates from this writer.
The American Medical Association recently adopted policies stating that AI should support evidence-based medicine while remaining under physician oversight. The policies specifically emphasize transparency, accountability and meaningful physician oversight.
The AMA has also called for disclosure of clinical logic, data sources and guidelines behind certain AI-supported determinations.
That is significant.
The debate is gradually moving from:
How intelligent is the AI?
toward:
How accountable is the decision architecture?
The FDA similarly treats AI-enabled medical devices within a lifecycle and risk-based framework, including development, validation, deployment, monitoring, maintenance and modification.
These developments point toward something larger.
The Human Still Matters
None of this diminishes the physician.
Quite the opposite.
Medicine involves much more than retrieving scientific literature.
The doctor encounters a person.
Symptoms exist within a history, a body, circumstances, preferences, fears and competing risks.
AI can contribute enormous analytical capacity.
The physician contributes clinical expertise and human contextual judgment.
The patient contributes something neither possesses:
the lived reality and values of the person whose body and future are affected by the decision.
The architecture should therefore not be:
Doctor → Patient
nor:
AI → Patient
but something closer to:
Patient + Physician + AI + Evidence → Examined Decision
with responsibility clearly defined.
A Different Kind of Second Opinion
This suggests that even the phrase “AI second opinion” deserves scrutiny.
A medical second opinion traditionally comes from another qualified professional.
AI produces something epistemically different.
Perhaps a better concept is:
AI-assisted clinical assessment
The purpose is not to establish another authority.
It is to introduce another analytical layer capable of identifying evidence, alternatives, contradictions and uncertainties that humans can examine.
That distinction protects both physician and patient from an important mistake:
confusing intelligence with authority.
From Trust to Epistemic Accountability
This principle does not apply only to medicine.
It applies to law.
Finance.
Government.
Scientific research.
And eventually to autonomous AI systems themselves.
Whenever AI participates in consequential decisions, society should increasingly demand four things:
Evidence.
What information supports the conclusion?
Reasoning.
How does that information justify the recommendation?
Uncertainty.
What does the system not know?
Responsibility.
Who remains accountable for the resulting decision?
This is what I would call epistemic accountability.
And it may prove much more important than asking whether an AI is more intelligent than a human expert.
The Wrong Question
Robert F. Kennedy Jr.’s comments are provocative because they challenge traditional medical authority. But replacing medical authority with algorithmic authority would not solve the underlying problem. [forbes.com], [bmj.com]
We need something more demanding than either:
Trust the experts.
or:
Trust the AI.
The standard should become:
Show me the evidence.
Show me the reasoning.
Show me the uncertainty.
Allow me to challenge the conclusion.
And tell me who is responsible for the decision.
The future of medical AI should not be built around deciding whom we trust.
It should be built around designing systems in which consequential reasoning can be examined, challenged and justified.
That is not the elimination of expertise.
It is the evolution of accountability.
For more information about Salomon-Prime, please contact me at josetguevara108@outlook.es or josetguevara108@gmail.com
José Tomás Guevara Calderón
Independent Researcher
San José, Costa Rica
