The more capable AI becomes at answering clinical questions, the more important it is to ask whether we are asking the right questions. Medicine is not becoming less human; if anything, it is becoming more dependent on the human qualities it has always needed: judgment, responsibility, and the ability to act with care amid uncertainty. AI may change how information is delivered, but it does not decide what matters. That still rests with the clinician, sitting with another person and working out the best way forward together. In this post, I argue that we do, and that clinical judgment now matters more, not less.
The information paradox
For years, many clinicians have relied heavily on guidelines, tests, and multidisciplinary teams. AI has accelerated that tendency. It is fast, fluent, and always available.
For straightforward cases, it can seem decisive. Yet for a growing number of patients, the problem is rarely straightforward. People arrive with incomplete stories, complex multimorbidity, diverse backgrounds, and lives that do not fit neat pathways. Consultations have become more emotionally and cognitively demanding, not less.
The real work of the consultation has never been pure information retrieval. It has always been the labor of building trust, listening to what is said and unsaid, demonstrating empathy, and tailoring evidence to a particular person at a particular moment.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The accuracy of any tool, AI included, still depends on asking the right questions, assembling a complete picture, and translating population data into a decision that fits this life. Medicine is no longer limited by access to knowledge. It is limited by the need for judgment under uncertainty, which is held within relationship.
When information stops being the bottleneck
AI answers the old questions almost instantly:
- What is the diagnosis?
- What does the evidence say?
- What are the recommended next steps?
The consultation itself has not been automated. In many ways, it has become harder. Clinicians now face an overload of information, some useful, some distracting, and some misaligned with the person in front of them.
Ask most clinicians five years ago what the hardest part of their job was, and many would have said keeping up with the evidence. Today, their answer has changed; it’s not just about the evidence anymore, but about figuring out what to do when the textbook case doesn’t quite match the unique person in front of you.
That question cannot be answered by an output alone. It requires judgment.
A clinical vignette: lung cancer and the limits of answers
Consider an elderly patient with symptoms suggestive of lung cancer. Imaging is performed. AI-assisted tools interpret the scan, suggest a high probability of malignancy, and outline standard next steps complete with survival curves and treatment pathways.
On paper, the pathway looks clear. In the room, it does not.
The patient is frail and is the primary carer for a spouse living with dementia. Their days are already full of medications, meals, and moments of confusion. They are frightened not only of dying but also of what treatment might mean for the small routines that still hold their life together. They look up and ask a question no algorithm can meaningfully answer.
“What would you do if this were your family member?”
There is no dropdown for that. No confidence interval captures the guilt of leaving a loved one alone, or the fear of spending what remains of life in hospital corridors rather than at the kitchen table.
At this point, the limiting factor is no longer diagnostic accuracy. It is judgment shaped by values, relationships, and the responsibility of sharing another person’s decision.
What this reveals
AI is increasingly adept at trawling through and organizing data, recognizing patterns, stratifying risk, and synthesizing guidelines. It can illuminate options with extraordinary speed.
What it cannot do is integrate the full context of a human life into a decision. It does not understand what it feels like to trade precious time for the burden of treatment, how caregiving determines what is actually possible, or when “best practice” in theory is not the best decision for this person right now.
These are not peripheral details to be considered after the “real” medical work is complete. They are the work of clinical medicine.
Clinical expertise is structured judgment under uncertainty, exercised in relationship. It means discerning which questions matter today, recognizing when more data will not resolve the ambiguity a patient lives with, balancing probabilities against lived experience, and accepting responsibility for outcomes that can never be fully predicted.
AI can broaden the view of possibilities and reduce certain errors. It cannot bear the moral and relational burden at the heart of these decisions.
Reframing the consultation
If we take this seriously, the consultation is no longer a data-retrieval exercise but a space for shared decision-making. The clinician’s role is not to compete with AI in providing information. It is to listen for what truly matters to this patient, in this life, at this moment, and to let that understanding shape the next step.
This includes:
- Knowing when guidelines apply cleanly and when they do not
- When to escalate and when to pause
- When certainty is possible, and when honesty requires us to say, “We don’t know, but we will face this with you.”
The consultation becomes less about having all the answers and more about being a trustworthy companion through the questions.
The enduring work of judgment
As AI becomes better at answering clinical questions, it becomes even more important to ask whether we are asking the questions that matter most to our patients. Medicine is not becoming less human. It is becoming more openly reliant on the human elements it has always required: the work of judgment, the weight of responsibility, and the courage to act in uncertainty alongside the people whose lives are most affected.
AI may reshape how information is delivered. It does not decide what matters. That responsibility remains, as it always has, with the clinician, sitting with another human being and choosing a path forward together.
I’d love to hear your thoughts:
- What do you think the clinician brings to the consultation that AI can’t?
- What will be the future role of clinicians?
- What is your experience of working with AI in the consulting room?
Adam Widdison is a surgeon and medical educator.
