AI is rapidly transforming oncology care by enhancing diagnostic precision in pathology, accelerating clinical trial matching, and enabling more precise personalized treatment strategies. The technology is also changing how patients with cancer receive health information, with about one-third of patients consulting AI chatbots to interpret lab reports, scans, and symptoms, often before speaking with their oncologists.

In this wide-ranging interview with Robert M. Wachter, MD, Professor and Chair of the Department of Medicine at the University of California, San Francisco, and author of A Giant Leap: How AI Is Transforming Healthcare, and What That Means for Our Future (Portfolio, 2026), ASCO AI in Oncology takes a deeper look at how this technology is likely to revolutionize cancer care over the next decade in terms of the oncology workforce, physician burnout, accountability, physician specialization, and more.

Using AI to Reduce Administration Burden

You have said that the American health-care system needs transformation. How might AI transform patient care, including improving access to care and patient outcomes and lowering cost in such a fractured and fragmented health-care system?

My core belief is that maintaining the status quo is unacceptable. The quality of health care is not what it should be, safety is not what it should be, access to care is difficult for many patients, costs are bankrupting patients and society, and many clinicians are burned out. We need help.

AI can help resolve many of these issues. For example, AI tools can reduce much of the administrative burdens that physicians face, including automatically drafting clinical notes, summarizing a 500-page medical record in minutes, and drafting prior authorization letters to insurance companies for new therapies. These can reduce physician burnout and improve patient communication.

At the same time, tens of millions of patients are using AI to get information about their symptoms or to interpret lab and imaging test results before they speak with their doctor. It gives them more access to expertise, and democratizes care in ways that are helpful.

I’m a generalist physician and I use the technology all the time to get information on a patient’s condition that I previously would have had to receive from a consultation with a specialist. AI doesn’t replace the specialist, but there are times when I need more precise or customized information than I can get from an electronic textbook—and quickly.

Now, whether this technology will actually lower health-care costs remains to be seen. It may allow patients to get some care from less expensive providers and to comparison shop. And AI will probably lower administrative costs.

We also have to remember that AI is not perfect, and users will have to get a sense of when to trust the information they receive from a chatbot and when they shouldn’t. That’s true when they see a physician, too. The burden will be on the makers of these tools and the institutions that employ them to ensure they are accurate and trustworthy.

Predicting the Future of AI Augmentation in Oncology

A report by ASCO, in 2025, found that the oncology workforce shortage is worsening, just as new cancer cases in the United States are rising. According to the report, only 4% of oncologists work in counties with high cancer mortality rates; and 38 states reported having fewer oncologists per capita in 2024 than they did in 2014. In addition, research from ASCO on physician burnout found that in 2023, three out of five oncologists (59%) reported symptoms of burnout vs 45% in 2013.

You have said that AI can help address these issues through productivity augmentation, such as documentation assistance and clinical decision support and through autonomous clinical substitution by providing clinical care without direct clinician oversight. Under what circumstances would this type of pathway be implemented?

We’ve got a long way to go before we accept truly autonomous clinical decision-making for AI. We are in a situation now in which AI tools are good enough to be useful for patients, especially for tasks like coordinating care, and for physicians, such as in scribing. But the tools are not good enough to be entirely trustworthy. So, for the foreseeable future for high-stakes clinical decisions, we are still going to need a human in the loop.

Autonomous AI has a role in certain back-office functions in cancer care, including automating repetitive, administrative tasks to reduce physician burnout and improve operational efficiency. But it’s going to be quite a while before we should be comfortable with letting AI diagnose patients or recommend therapy. The tools are getting better quickly, so we may reach that point eventually.

Credentialing AI to Practice Medicine

Would AI need to have a medical license to practice medicine?

Nobody knows the answer to that question yet. What is very clear is that the United States Food and Drug Administration (FDA)—the organization that determines whether a drug or device is safe and effective—is not the right model to oversee implementation of this technology when it comes to day-to-day clinical practice and decision support. One reason is the tools change too quickly.

I believe we will need a licensing system for AI that resembles the one we have for physicians. Physicians have to go through rigorous medical training to become licensed and board certified in their specialty. AI developers will have to demonstrate that their autonomous tool went through appropriate and trustworthy training and was able to pass a test that is a reasonable facsimile of the services it will perform in real life. Moreover, the tools will need to be retested periodically to ensure they are maintaining proficiency.

Establishing an Office of Clinical AI Oversight

You have suggested establishing an Office of Clinical AI Oversight staffed by individuals with dual expertise in clinical care and AI, including physicians, clinical informaticians, machine learning experts, and health policy experts, as well as an external advisory board. Would such an agency be housed within the federal government, and is this idea moving forward?

I don’t know that an Office of Clinical AI Oversight would be embedded in an agency within the federal government. Medical board certification is done by nonprofit organizations, not federal agencies.

What is clear is that we’re going to need some new organizations that have appropriate competencies in both the relevant clinical field and in technology to create the appropriate structure to license and oversee these tools. Where such an office might live in terms of whether is it in the public or private domain remains to be determined.

The FDA is a public agency, but the organization that accredits over 22,000 health-care organizations and programs in the United States, including hospitals, is the Joint Commission. The Centers for Medicare and Medicaid has its own role in deeming hospitals as trustworthy, and states have a role in certifying clinicians. So, often our system of certification and accreditation is a complicated mix of government and private organizations. We just haven’t figured out yet what the licensure framework for autonomous clinical AI will look like and who would be in charge of oversight.

Determining Accountability for AI Use in Clinical Care

Who do you believe would be responsible for AI’s performance and safe workflow integration into clinical care: medical institutions, AI developers, or physicians?

Potentially all three, because it’s hard to figure out how you would split that accountability and risk. The Joint Commission recently published its Responsible Use of AI in Healthcare Certification, which recognizes health-care organizations in the United States that demonstrate they have the governance, safeguards, monitoring processes, and education in place to use AI responsibly. So, I think we’ll begin to see some medical institutional accountability in the implementation of AI clinical tools to ensure their safety and effectiveness and that they are being used appropriately. There will also need to be appropriate product liability for clinical AI manufacturers who put out tools with clinically relevant defects.

What responsibility physicians may bear after a bad outcome in the setting of a flawed AI diagnosis or treatment recommendation is unclear. I’m guessing that the standard practice of holding physicians responsible for their clinical decisions will still apply, even if they relied on the output of an AI tool.

Introducing the Generalist-Specialist in Oncology Care

In your paper “How AI will redefine care delivery: the rise of the generalist-specialist,” you argue that AI could fundamentally reorganize the clinical workforce with the development of AI-augmented clinicians that could manage a range of patients’ chronic and complex conditions within broader, disease-based domains, such as cardiometabolic, infectious, and inflammatory, rather than organ-specific specialists. Cancer encompasses thousands of diseases and genomic sequencing is fundamentally personalizing care for each patient with cancer. Would an AI-augmented generalist-specialist apply in cancer care?

It depends. I imagine that there are certain cancers that are so common and straightforward that an AI-augmented generalist-specialist may be able to manage these patients’ oncology care, especially in this setting of oncology workforce shortages. Will all patients want and accept care from a generalist? Probably not. I’d imagine that many patients will say, “I want to see a specialist.” But some of the decisions might be determined by cost—the patient can see a specialist, but would be hit with a substantial additional payment to do so.

It may turn out that the generalist can provide reasonable care for certain kinds of tumors without a lot of specialist involvement. Or it may be that the specialist acts as a second opinion, and that more of the care is carried out by the generalist practitioner, guided in part by AI. I could see that happening.

In a world in which everyone could afford the cost of health care and there were enough oncology specialists, of course, we would want every patient to be seen by the most skilled practitioner in the field. But in the world that we actually have, one in which the cost of health care is bankrupting individuals and society, and there are shortages of oncologists and other specialists in many regions of the country, it seems inevitable that we are going to expand the boundaries of what a generalist can do.

I’m a generalist myself and there are patients I see with conditions that 2 years ago I would have asked a specialist for a consultation, because I want to be very sure of the best treatment strategy, but today I’ll consult with an AI tool, such as OpenEvidence, instead. And generally, the answers I get back are very good.

My criterion now for getting a human specialist consult is that either the patient needs a procedure or that the question I have about the patient’s care is complex enough and the stakes are high enough that I wouldn’t trust AI to give me an accurate answer. But often I am confident that AI is giving the answer that I would have gotten from a specialist.

Supporting the Oncology Workforce Through Augmented AI Care

How do you envision AI changing the oncology workforce?

I don’t think many physician roles will be eliminated over the next decade. Beyond the next decade, it’s hard to predict what will happen because this technology is advancing so fast.

Currently, there are so many unmet patient needs. Oncologists are under tremendous time pressure, and it can be difficult to get a timely appointment with an oncologist. Right now, I believe that AI is beginning to allow generalist-specialists to perform some routine cancer care, such as analyzing genomic data and navigating complex treatment guidelines with greater accuracy. There is still plenty of work for oncologists, and I don’t envision any unemployed oncologists in the foreseeable future.

It’s natural that oncologists are concerned about how AI augmentation may impact clinical care. But there are so many unmet patient needs, so many barriers to health-care access, and so few specialists to meet the increasing demand that if AI can reduce the oncologists’ workload either by AI assisting in documentation processes or in clinical decision support, it will free up their time to concentrate on their most complex cases.

DISCLOSURES: Dr. Wachter is a board member of The Doctors Company, Second Wave Delivery Solutions, Arbiter, Coral, and the Josiah Macy Jr. Foundation. He is a scientific advisor to Commure, Curai Health, Roon, Notable, and Qualified Health. He has also received speaking fees from Google and Sanofi.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.