The Machines Will See You Now 

Today we are in the midst of an AI revolution whose trajectory seems increasingly difficult to control. Enthusiasts predict dramatic improvements in efficiency, accuracy, and access to health care. Critics fear the erosion of professional judgment and human relationships. The outlook for medicine in 2027 seems eerily similar to Simone Weil’s assessment of Western civilization in the turbulent 1930s, “On the whole, our present situation more or less resembles that of a party of absolutely ignorant travelers who find themselves in a motorcar launched at full speed and driverless across a broken country.”¹ Will AI replace many clinicians? Will the physician-relationship survive? Can AI replicate the healing virtues of empathy and compassion? 

One particularly interesting claim is, at first glance, counterintuitive. Several recent books argue that full implementation of AI in medicine will be a boon to the doctor-patient relationship and make clinical practice more humane. Anmol Madan’s The Machines Will See You Now (2026)², argues that automated medicine will deepen, rather than diminish the humanity at the heart of health care. “How Artificial Intelligence Can Make Health Care Human Again” is the subtitle of Eric Topol’s Deep Medicine (2019)³. Shannon Vallor presents a more measured, but still optimistic, perspective The AI Mirror (2024)⁴. In this essay, I want to explore the potential for traditional values in medicine at a time in the future when most diagnostic and treatment decisions may be made by large language model generative AI systems.  

Health care is already benefiting from automation, in which algorithmically driven programs interpret radiographic images and EKGs, enhancing the accuracy and efficiency of interpretation.  In one recent multicenter study, endoscopists performing colonoscopy with images analyzed by AI discovered 21% more adenomas than those performing colonoscopy without AI. However, a drawback is already apparent in evidence demonstrating a “de-skilling effect” of reliance on AI among radiologists and ophthalmologists after using AI assistance for some time.⁵ Once accustomed to passive verification of AI findings, clinicians find it difficult to engage in active observation to their former level. In a less “mixed blessing” way, automation is also producing benefits in clerical and administrative areas.  

The Machines Will See You Now characterizes the current situation as Level 2 in his four-step roadmap to fully automated health care. We are now entering the era of high automation (Level 3) in which AI will have the ability to make many or most diagnoses and treatment decisions, with human physicians limited to consulting in cases of patients with complex conditions. It should be noted that Anmol Madan is a prominent information technology scientist and not a physician. He rejects the term “patient” because it signifies passivity and a power imbalance between the consumer (aka patient) and provider (aka physician). He believes AI-assisted health care will lessen this imbalance and may eliminate it when we reach Level 4, which is defined as “the transition of health care provided by humans to health care supported and eventually delivered by algorithms.”^2,21 This may not be realized for several decades, but in the meantime much of the work of physicians in primary care specialties and psychiatry will be performed by machines in Level 3. Nonetheless, after describing impressive advances in AI technology, Madan notes, almost in passing, “Researchers have not yet figured out how well these seemingly compelling accuracy and performance results translate to real world use in health care,”^{2, p. 82} 

Eric Topol is a physician, and he better understands the humanity at the heart of healing. Because of this, he deplores the cold, mechanical, and robotic ways in which medicine is practiced today. Much of this results from productivity demands, administrative overload, and medical education that prioritizes technology over communication and compassion.  In essence, he claims that we already have thinly disguised robotic medicine. If AI takes over most notetaking, diagnostic, therapeutic, and “busy-work” functions, then physicians will be free to establish human connection with their patients. I believe Topol has made the correct diagnosis of contemporary medical practice, but is his solution viable? 

Madan supports his humanization claim with a discussion of an AI system’s potential to replicate empathy. “Empathy at its core,” he writes, “is a set of verbal and body language signals.^{2, p. 65} Already, large language models can modulate the tone and pace of their speech to replicate various human styles, concerns and moods. They can generate empathic responses, like, “That must be difficult” or “I’m sorry you’re experiencing so much pain.” He claims that the newest generative AI models “score high on empathy.”^{2, p. 72} Would such features be experienced by patients as genuine concern? Would they learn to trust a machine in the same way they trust their doctor? 

In today’s fragmented health care settings, physicians are harried by time pressures and multitasking, which often compromise empathic behavior. Given Madan’s background, it is understandable that he proposes an AI solution to the problem, but Topol understands that empathy is not merely the production of appropriate verbal responses, but rather it involves shared human presence, attention, reassurance, and understanding. Machines cannot vicariously participate in pain and suffering, as fellow humans can. Patients will not trust, or be comforted by, their AI providers_.. Although we commonly “trust” test results or interpretations produced by a machine, this “trust” is phenomenologically different from the trust involved in a human relationship.  

Thus, while Madan claims that AI will directly contribute to humanization by offering surrogate empathic behavior, Topol believes that physicians, relieved of time constraints and harried multitasking, will be free to engage in genuine empathic patient encounters. To be fair, Madan does refer to physicians’ having time for other activities when they are relieved of diagnosis and treatment decisions, but his goal of almost completely replacing physicians in many areas, except for very complex cases, makes you wonder whether the activities mentioned might include visiting the unemployment office.  

Shannon Vallor, a philosopher at the University of Edinburgh, takes a step back and questions the assumptions on which either one of these “health care more humane with AI” projections are based. In The AI Mirror, she states the obvious, but rarely discussed, fact that human technologists “train” generative AI systems with algorithms and immense amounts of data. Sometimes the algorithms may reflect human bias, or the training data may include errors. She describes an egregious case of racial bias discovered in 2019 in an AI risk prediction algorithm widely used in American hospitals. The algorithm “diverted medical care away from high-risk Black patients, even though these patients were, in fact, sicker than the white patients the tool prioritized for care.”^{4, p. 42} Race had been explicitly excluded from the training data, but .the AI system was able to reconstruct the same bias by integrating other forms of data, like residence and income, into its decision making,  

Vallor’s point is that generative AI mirrors the values and aspirations of the culture that created it. When we look at a powerful Large Language Model system, like ChatGPT, we see our own priorities, both healthy and flawed, writ large. In medicine, AI should not be understood primarily as a technological revolution but as a moral and cultural mirror revealing what modern medicine values, what it neglects, and what aspects of healing can and cannot be mechanized. From this perspective, whether AI will make medicine more humane depends less on the technology itself than on the vision of healing that clinicians, patients, and health care institutions bring to it. 

There is almost universal agreement that American health care needs major reform. In 1977, John Knowles edited a collection of essays entitled Doing Better and Feeling Worse: Health in the United States, which was severely critical of the fragmented, expensive, mechanistic, and inequitable health care system at the time, despite its many technological advances.⁶ Today, fifty years later, despite remarkable technological developments since then, we are in an even more critical situation. We are both doing worse and feeling worse. In this context, how is the future of AI-automated health care likely to look?  

Currently the U.S. health care sector employs nearly 18 million people, among them over 860,000 physicians.⁷ Millions of these are engaged in clerical and administrative jobs that will likely be handled more efficiently by AI systems. If any scenario even approximating Madan’s Level 3 or 4 occurs, tens of thousands of physicians may also be made redundant. Thus, full implementation of AI will probably lead to a considerable reduction in the health care workforce. This negative economic consequence could only be avoided by significant health care reform involving professional development and the creation of new openings for genuine caregiving. 

Modern health policy often treats “care” as the aggregate of services delivered, procedures performed, and outcomes measured. Yet the word “care” refers not primarily to a service but to a disposition or motivation to respond to human suffering. If humanity in medicine comes from caring relationships, then technology is beneficial only insofar as it enables, rather than replaces, those relationships. If medicine is a moral practice involving trust, suffering, mortality, uncertainty, and compassion, then AI can assist but never replace the physician’s role.^8,9 Contemporary health care fosters ambiguity between the two definitions of “care.” While in practice the quantitative definition prevails, in theory (e.g., hospital commercials, mission statements), institutions tout the qualitative definition. It is reasonable, then. to expect the same ambiguity to continue with AI enhanced medicine, unless major reform results in prioritizing genuine caring 

From this perspective, if AI saves physicians two hours of clerical work per day, will those hours be returned to more productive encounters with patients, or will they be absorbed by institutions seeking greater productivity? Indeed, physicians currently spend increasing amounts of time on various forms of documentation, chart review, data management, and other forms of drudgery, from which AI will be able to relieve them. But will this leave them more time for listening to, examining, reassuring, and teaching patients, as Topol believes? A hopeful scenario, but unlikely under current institutional demands. It seems more likely that the extra hours will be absorbed into the hospital’s or clinic’s demand for income and efficiency. Every major medical technology has been accompanied by promises that efficiency would create more time for the patient. Yet the history of medicine shows the opposite outcome: efficiencies become opportunities for greater productivity. 

And so I return to Simone Weil’s metaphor. We once again find ourselves “in a motorcar launched at full speed and driverless across a broken country.”¹ Health care automated by AI will surely be able to drive the car even faster, but I don’t believe that AI in itself will take us to a more humane destination. At present, AI mirrors the values of a system that values technology, productivity, and income over equity and caregiving. If we want our car to arrive at universal and compassionate health care, radical reform – not generative AI – must become its driver. 

References 

1. Simone Weil, Oppression and Liberty, Routledge,2001 (1st edition, 1955) 

2. Anmol Madan, The Machines Will See You Now, Baltimore, Johns Hopkins University Press, 2026 

3. Eric Topol. Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, New York, Basic Books, 2019 

4. Shannon Vallor, The AI Mirror, New York, Oxford University Press, 2024 

5. Maxim Tokas, Laura Marie Peltomen, Zhihong Zang et al, The deskilling effect: is artificial intelligence eroding clinical competence? Ann Intern Med. 2026; 179: 1048-1054 

6. John Knowles (Ed.) Doing Better and Feeling Worse: Health in the United States, New York, Norton, 1977 

7. Health Resources and Services Administration. State of the U.S. Health Care Workforce, 2025. National Center for Health Workforce Analysis, December 2025,  

8. Eric J. Cassell, The Nature of Suffering and the Goals of Medicine. New York, Oxford University Press, 2004 

9. Arthur Kleinman, The Soul of Care, New York, Penguin Books, 2020 

Web image generated by Medhum.org