The Doctor AI Cannot Replace
I dedicate this article, with gratitude, to all the healthcare professionals and administrative staff at the clinics and hospitals who are helping me through this difficult period. Their competence, patience, kindness, and care have meant more to me than they may realize—and, above all, to my brother, who, besides being my brother, has also been my physician and has accompanied me through every step of this journey.
An earlier version of this essay was written before generative AI became part of everyday life. This substantially revised 2026 edition asks a different question: not whether artificial intelligence will enter medicine—it already has—but what kind of medicine we will choose to build with it.
10 -15 min
Artificial Intelligence, Compassion, and the Future of Medicine
We Asked the Wrong Question
For years, the debate about artificial intelligence and employment was dominated by a frightening question:
How many jobs will AI eliminate?
The numbers were dramatic. Studies estimating the technical susceptibility of occupations to automation were often transformed, in public discussion, into predictions that enormous fractions of the workforce would simply disappear.
Reality has proved more complicated.
The International Labour Organization’s updated 2025 analysis estimates that roughly one in four jobs worldwide has some degree of exposure to generative AI. But its most important conclusion is not the number. It is that transformation is more likely than outright replacement.
That distinction changes the problem.
A profession is not a single task.
A physician diagnoses, interprets, writes, searches, explains, persuades, reassures, decides, takes responsibility and sometimes simply sits beside another human being who has received terrible news.
A nurse observes, measures, administers, records, communicates, comforts, notices what nobody else noticed and often recognizes changes in a patient before a machine or physician does.
Artificial intelligence may become extraordinarily good at some of these tasks without making the entire profession obsolete.
So perhaps the question was wrong from the beginning.
Instead of asking:
Which professions will AI replace?
we should ask:
Which parts of each profession will AI absorb—and what should humans do with the time that remains?
Medicine may become one of the most important experiments through which humanity discovers the answer.
The Future Has Already Entered the Hospital
When I first wrote about this subject, artificial intelligence in medicine could still be described largely in the language of the future.
It no longer can.
AI systems now assist with medical imaging, disease detection, clinical decision support, documentation, patient communication and other parts of healthcare. In January 2025, the U.S. Food and Drug Administration reported that it had already authorized more than 1,000 AI-enabled medical devices through established regulatory pathways.
The change is not confined to experimental laboratories.
A WHO/Europe report published in April 2026 found that 74% of European Union countries reported using AI in diagnostics, while 63% reported using chatbots to support patient engagement. The same report emphasized the growing need to train health professionals to work critically and safely with these systems.
The relevant question, therefore, is no longer whether AI will enter healthcare.
It has entered.
The interesting question is what happens next.
Kai-Fu Lee’s Epiphany
Long before ChatGPT and the current explosion of generative AI, Kai-Fu Lee proposed a provocative answer. In his 2018 book AI Superpowers: China, Silicon Valley, and the New World Order, Lee examined the approaching social and economic consequences of increasingly capable artificial intelligence and argued that the future of human work could not be understood simply as a competition between people and machines.
Lee had spent much of his professional life at the center of the technological revolution. He worked at Apple, Microsoft and Google, led Google China and later became one of the best-known technology investors and thinkers in China.
Then cancer interrupted the algorithm.
After being diagnosed with stage IV lymphoma, Lee began reconsidering a life that he had optimized heavily around productivity, achievement and work. Confronting mortality forced him to reconsider the importance of relationships, love and human connection.
That experience became central to one of the most important arguments in AI Superpowers.
Lee argued that increasingly capable machines would take over many tasks based on optimization, pattern recognition and routine cognitive work. Humans should therefore invest more—not less—in occupations and activities built around interpersonal connection, compassion and care. In healthcare, he imagined AI handling increasing amounts of technical analysis while humans devoted more attention to the patient as a person.
The proposal was not simply to protect existing medical jobs from automation. It was to redesign them around the capabilities that automation could not easily reproduce.
It was an elegant division of labor:
let machines do what machines do well; let humans become more human.
In medicine, the implication was radical.
Perhaps technological progress would not eliminate the doctor.
Perhaps it could give the doctor back to the patient.
The Compassionate Caregiver
Modern medicine contains a paradox.
It has never possessed so much knowledge, yet the people practicing it frequently have too little time.
Healthcare professionals spend enormous portions of their working lives entering information, navigating electronic systems, preparing documentation, searching records, dealing with administrative processes and performing repetitive cognitive tasks.
Many of these activities are necessary.
Few are the reason someone decides to become a physician or nurse.
Imagine that artificial intelligence can remove a meaningful portion of that burden.
It summarizes the medical history before the consultation.
It retrieves relevant information from thousands of pages of records.
It drafts routine documentation.
It checks interactions between medications.
It identifies abnormalities in images.
It follows laboratory trends.
It suggests differential diagnoses.
It prepares an initial response to a patient’s message.
None of this automatically requires removing the physician.
Quite the opposite.
It could create something that has become scarce in modern healthcare:
time.
Time to listen.
Time to explain.
Time to ask the second question.
Time to notice fear that was never entered into the electronic medical record.
Time to explain why a technically optimal treatment may be unacceptable to this particular patient.
Time to hold the hand of someone who has just learned that life will not continue as expected.
This was the beauty of Lee’s argument.
Artificial intelligence could increase the value of what initially seemed least technological about medicine.
But something unexpected happened.
AI Learned to Sound Compassionate
The old argument depended on a comfortable boundary.
Machines were computational.
Humans were compassionate.
Then large language models arrived.
They did not acquire human emotions. There is no evidence that a language model experiences concern, fear, affection or suffering merely because it produces sentences associated with those states.
But they became remarkably good at producing the language of empathy.
A 2023 study published in JAMA Internal Medicine compared physician responses with ChatGPT responses to 195 medical questions posted on an online forum. Licensed healthcare professionals evaluating the answers preferred the chatbot responses in 78.6% of evaluations and rated them significantly higher for both quality and expressed empathy. The experiment had important limitations: it involved written online answers rather than real clinical relationships, chatbot responses were considerably longer, and the study did not demonstrate that AI could independently deliver safe clinical care.
Nevertheless, the result exposes an uncomfortable possibility.
A machine does not need to feel empathy to produce a response that a human being perceives as empathetic.
That distinction may become one of the central philosophical problems of AI-mediated medicine.
Suppose a frightened patient asks:
Am I going to die?
A machine might construct an exquisitely worded response—patient, sensitive, reassuring and perfectly adapted to the emotional context.
But nothing inside the machine necessarily fears the patient’s death.
Does that matter?
At first the answer seems obvious.
Of course it matters.
Compassion without feeling appears to be imitation.
But now reverse the situation.
Imagine an exhausted physician who genuinely cares about the patient but has twelve minutes for the consultation, twenty people still waiting outside and several hours of documentation ahead.
His concern is authentic.
His communication may nevertheless be hurried.
The machine’s compassion is synthetic but abundant.
The physician’s compassion is authentic but constrained.
Which one does the patient experience?
The question is more difficult than it first appears.
Empathy Is More Than a Sentence
There is nevertheless something dangerous about reducing compassion to linguistic performance.
Medicine is not simply an exchange of information.
A physician does not merely produce the correct sentence after receiving the appropriate prompt.
The relationship exists through time.
The doctor may have seen the patient healthy, watched the disease emerge, tried one therapy, watched it fail, delivered bad news, met the family and accepted responsibility for the next decision.
Human empathy is embedded in biography, vulnerability and consequence.
The patient can suffer.
The physician can also suffer.
Both know what mortality means because both inhabit mortal bodies.
An AI can construct the sentence:
“I understand how frightening this must be.”
But the word understand becomes philosophically complicated when spoken by an entity that does not fear illness, separation or death.
This does not make artificial empathy useless.
Quite the contrary.
AI may help physicians communicate better. It may suggest clearer explanations, detect unnecessarily cold language, translate complex medical information, draft responses and remind an exhausted professional that behind a laboratory result there is a frightened human being.
But assisting empathy and possessing empathy are not necessarily the same thing.
And perhaps they do not need to be.
The mistake would be to conclude that because AI can reproduce one expression of humanity, the human relationship itself has become unnecessary.
The Real Battle Is Not Human Versus Machine
This leads to what may be the most important question.
The conflict in healthcare will probably not be:
AI versus doctors.
It will be between two different ways of using the same technology.
Consider two hospitals.
Hospital A
AI reduces documentation time.
Diagnostic systems increase efficiency.
Administrative processes become automated.
Management calculates the productivity gain and concludes that fewer professionals can now treat more patients.
Staff numbers fall.
Appointments become shorter.
Workload rises.
The hospital becomes more computationally efficient and more psychologically brutal.
Hospital B
The same technologies are introduced.
But the productivity gain is used differently.
Doctors spend less time typing.
Nurses spend less time entering repetitive information.
Administrative delays fall.
Professionals are given more time with each patient.
AI performs more computation.
Humans perform more medicine.
Technologically, the two hospitals may be almost identical.
Morally, they are different institutions.
And this reveals something frequently forgotten in discussions about artificial intelligence:
technology does not determine how productivity gains are distributed. Institutions do.
AI can be used to reduce the burden on healthcare professionals.
It can also be used to increase the burden on the professionals who remain.
It can create more time for patients.
Or it can become the justification for putting more patients into the same amount of time.
Artificial intelligence does not make that decision.
We do.
What Should We Teach the Next Generation of Doctors?
This also changes medical education.
For centuries, becoming a physician required acquiring an enormous internal library of knowledge.
That will remain important. A doctor who cannot reason independently cannot safely supervise an artificial intelligence system.
But memorization alone will become progressively less valuable when a machine can retrieve and synthesize enormous volumes of information almost instantly.
Other skills become more important.
Knowing when the machine may be wrong.
Understanding uncertainty.
Recognizing bias.
Asking better questions.
Explaining probabilities.
Combining conflicting evidence.
Taking responsibility when an algorithm recommends one thing and clinical judgment suggests another.
Understanding the patient rather than merely the disease.
And communicating when no calculation can produce the answer a patient actually wants.
The physician of the AI age therefore should not know less medicine.
The physician may need to understand medicine more deeply precisely because superficial retrieval is becoming cheap.
At the same time, medical education may finally have to take communication, psychology, ethics and human relationships as seriously as it takes technical competence.
The paradox is beautiful.
The more intelligent our machines become, the more seriously we may have to teach humans how to be human.
The Doctor AI Cannot Replace
The question “Will AI replace doctors?” is no longer particularly useful.
AI will replace some tasks performed by doctors.
It will create others.
It will outperform humans in certain narrow activities and fail unpredictably in others. It will become an assistant, a second opinion, a documentation system, a diagnostic instrument and perhaps eventually something much more autonomous.
Conversational diagnostic systems are already being studied at increasingly sophisticated levels, but experimental performance is not equivalent to independent clinical practice; real medicine involves multimodal information, uncertainty, accountability and relationships accumulated over time.
The physician who merely transfers information from medical knowledge to the patient is increasingly vulnerable to automation.
But that was never the complete physician.
The irreplaceable doctor is not a database with a stethoscope.
It is the person who combines knowledge with judgment, judgment with responsibility and responsibility with concern for another human being.
Perhaps one day machines will imitate every visible component of that relationship convincingly enough that we will again have to reconsider where the boundary lies.
We should remain humble about that possibility.
But we are not there yet.
And meanwhile there is a much more immediate danger.
We may develop machines capable of freeing healthcare professionals from enormous amounts of mechanical work—and then use those machines merely to demand more mechanical work from fewer humans.
That would be an extraordinary technological achievement and a profound failure of imagination.
Kai-Fu Lee’s insight therefore remains relevant, although for a different reason than it did in 2018.
The great opportunity of artificial intelligence in medicine is not simply to create a machine that diagnoses faster.
It is to reconsider what humans should do once machines can perform more of the computation.
We can use AI to reduce the number of people working in medicine.
Or we can use it to reduce the amount of mechanical, repetitive work that healthcare professionals are forced to do.
If we choose the second path, AI may give doctors and nurses something increasingly scarce in modern healthcare: time—to think, to listen, to explain, and to care.
Perhaps the greatest achievement of artificial intelligence in medicine will therefore not be a hospital with fewer humans, but a hospital in which humans have more time to be human.
Reference List
Ayers, John W., Adam Poliak, Mark Dredze, Eric C. Leas, Zechariah Zhu, Jessica B. Kelley, Dennis J. Faix, et al. 2023. “Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum.” JAMA Internal Medicine 183 (6): 589–596. https://doi.org/10.1001/jamainternmed.2023.1838.
Butlin, Patrick, Robert Long, Tim Bayne, Yoshua Bengio, Jonathan Birch, David Chalmers, Axel Constant, et al. 2026. “Identifying Indicators of Consciousness in AI Systems.” Trends in Cognitive Sciences 30 (6): 488–501. https://doi.org/10.1016/j.tics.2025.10.011.
Gmyrek, Paweł, Janine Berg, Karol Kamiński, Filip Konopczyński, Agnieszka Ładna, Balint Nafradi, Konrad Rosłaniec, and Marek Troszyński. 2025. Generative AI and Jobs: A Refined Global Index of Occupational Exposure. ILO Working Paper 140. Geneva: International Labour Organization. https://doi.org/10.54394/HETP0387.
Lee, Kai-Fu. 2017. “A Blueprint for Coexistence with Artificial Intelligence.” WIRED, July 12, 2017. https://www.wired.com/story/a-blueprint-for-coexistence-with-artificial-intelligence/.
Lee, Kai-Fu. 2018. AI Superpowers: China, Silicon Valley, and the New World Order. Boston: Houghton Mifflin Harcourt.
Liévin, Valentin, Anil Palepu, Wei-Hung Weng, Khaled Saab, David Stutz, Yong Cheng, Kavita Kulkarni, et al. 2026. “Towards Conversational Artificial Intelligence for Disease Management.” Nature 655 (8125): 1292–1299. https://doi.org/10.1038/s41586-026-10764-5.
Sinsky, Christine, Lacey Colligan, Ling Li, Mirela Prgomet, Sam Reynolds, Lindsey Goeders, Johanna Westbrook, Michael Tutty, and George Blike. 2016. “Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties.” Annals of Internal Medicine 165 (11): 753–760. https://doi.org/10.7326/M16-0961.
Teo, Zhen Ling, Arun James Thirunavukarasu, Kabilan Elangovan, Haoran Cheng, Prasanth Moova, Brian Soetikno, Christopher Nielsen, et al. 2025. “Generative Artificial Intelligence in Medicine.” Nature Medicine 31 (10): 3270–3282. https://doi.org/10.1038/s41591-025-03983-2.
Tu, Tao, Mike Schaekermann, Anil Palepu, Khaled Saab, Jan Freyberg, Ryutaro Tanno, Amy Wang, et al. 2025. “Towards Conversational Diagnostic Artificial Intelligence.” Nature 642 (8067): 442–450. https://doi.org/10.1038/s41586-025-08866-7.
U.S. Food and Drug Administration. 2025. “FDA Issues Comprehensive Draft Guidance for Developers of Artificial Intelligence-Enabled Medical Devices.” Press release, January 6, 2025. https://www.fda.gov/news-events/press-announcements/fda-issues-comprehensive-draft-guidance-developers-artificial-intelligence-enabled-medical-devices.
World Health Organization Regional Office for Europe. 2026. Artificial Intelligence Is Reshaping Health Systems: State of Readiness across the European Union. Copenhagen: WHO Regional Office for Europe. WHO/EURO:2026-12707-52481-81471. https://www.who.int/europe/publications/i/item/WHO-EURO-2026-12707-52481-81471.
Editorial transparency note: This article, as with all articles published on this site, was conceived, directed, written, and reviewed by Prof. Maurício Veloso Brant Pinheiro. Artificial intelligence was used as an assistant for editorial refinement, formatting, image generation, SEO metadata, and publication workflow.
Copyright 2026 AI-Talks.org
