Junren Chen, MD, PhD, MBA
ROBERT PETER GALE, MD, PhD, DSc (hc), FACP, FRCP, FRCPI (hon), FRSM
In a recent article in The ASCO Post, we discussed increasing use of artificial intelligence (AI) in oncology and how physician-complementing AI can empower oncologists to be even better at what they do.1The reason AI is needed is that increasingly many variables need to be considered in cancer therapy decision-making,a challenge known as the curse of dimensionality. Also, people receiving cancer therapy may prefer outcomes like quality-of-life, time out-of-hospital, or lower cost, moreso than longer progression-free survival (PFS) or survival, endpoints that exist in most expert consensus statements and clinical practice guidelines. The curse of dimensionality, conflicting therapy goals, and the limited bandwidth of the human mind means oncologists need help in making therapy decisions.
If the motivations for using AI to assist cancer therapy decisions are clear, how shall we involve AI? We try to answer this below.
Understanding the Role of the ‘AI Doc’
Discussions of drug-prescribing AI often center on the question of who is responsible for mishaps.2 However, an “AI Doc” is not necessarily a fully autonomous physician but a partner to whom we sometimes delegate mission-critical tasks including drug prescribing. It is the responsibility of the AI Doc to diligently propose the most rational prescription option for someone whereas the responsibility of the oncologist-in-charge is reviewing the AI-generated prescription and accepting, modifying, or rejecting it.
There is considerable literature in business management on the effective execution of delegation which oncologists can learn from.3 The popular phrase “physician in the loop” in the AI literature does not accurately describe the world view of many people. The key issue is how to loop in AI within human-centered care workflows, not how to engage physicians in AI-centered care workflows.
First, oncologists need to decide which tasks relevant to cancer therapy they exclude delegating to an AI Doc. For example, an AI Doc probably should not communicate therapy decisions directly to patients most of whom will prefer consulting with their oncologist rather than an AI bot. Furthermore, an AI Doc communicating drug prescription decisions to pharmacies and patients can be confusing because its messages to them may differ from those from oncologists.
Second, oncologists need to decide which therapy-related tasks are better done by an AI Doc than themselves. That a task is better done by an AI Doc in no way implies AI is smarter than the oncologist. For example, an AI Doc may not have human-level understanding of a therapy protocol, but it can monitor protocol adherence more diligently and thoroughly than most humans, especially when the protocol is complex with many decision points.
Managing the AI Doc
Other therapy-related tasks possibly done better by an AI Doc compared with oncologists include: (1) checking drug inventories to ensure prescribed drugs are in stock and equivalent in potency and risk of adverse events; (2) reviewing electronic medical records (EMRs) to identify drugs likely to be associated with an adverse event in someone; (3) compiling and comparing outcomes of people with similar conditions who received different therapies; and (4) cross-checking clinical and genomics data of a patient with the scientific literature, expert consensus statements, and clinical practice guidelines to suggest therapy option(s).
Few senior oncologists would oppose delegating these tasks to junior physicians provided they are motivated, capable, well-read, and good communicators.The reason for concern about delegating therapy-related tasks to an AI Doc is how an AI Doc can be monitored and corrected when it makes mistakes.
The third and perhaps the most important consideration when delegating a therapy-related task to an AI Doc is the need for clearly defined expectations for how the AI Doc should report to the physician-in-charge, support therapy decision-making, and facilitate prescription ordering.
This subject was abstract until the availability of large language models (LLM) and AI agents. Conventional AI models were rigid in how they interacted with humans and often difficult to retool or customize. Scripted insights provided by conventional AI models omitted key data or were too expansive. In short, communication with conventional AI models felt unnatural.
New AI technologies give oncologists greater control over the format, scope, and depth of information an AI Doc presents to support final therapy decisions. The AI Doc may learn and adapt to oncologists’ evolving communication styles. After physicians confirm final therapy decisions the AI Doc enters prescription orders in the health information system, enabling oncologists to focus on higher-value tasks.
The AI Doc in Cancer Care
In summary, we propose a three-question framework for oncologists to conceptualize the use of an AI Doc in cancer therapy: (1) What therapy-related task should or should not be delegated to an AI Doc? (2) What therapy-related task is best done by an AI Doc? (3) How do I want an AI Doc to assist me in finalizing therapy decisions?
The job description of an AI Doc may vary between cancers. For some diseases, the complexity of implementing a therapy protocol is considerably higher than that of choosing a therapy protocol. An example is acute lymphoblastic leukemia in children where we predict an AI Doc will focus on assisting physicians in placing drug prescription and laboratory test orders defined in the protocol. For other cancers, we predict the AI Doc will focus on assisting oncologists and patients to choose a therapy based on therapy goals and patients’ clinical and genomics covariates.
In conclusion, there is considerable potential for the use of an AI Doc in cancer therapy. However, the proper benchmark for an AI Doc is not whether it will replace oncologists (it won’t). Rather, it is the advent of an AI Doc workforce ready to be hired by oncologists to assist in therapy-related tasks.
DISCLOSURE: Dr. Gale is a consultant for Antengene Biotech LLC and Shenzhen TargetRx; is Medical Director at FFF Enterprises Inc.; a speaker for Janssen Pharma, BeiGene, and Hengrui Pharma; on the Board of Directors for the Russian Foundation for Cancer Research Support; on the advisory board for StemRad Ltd; and has received support from the UK National Institute of Health Research (NIHR). Dr. Chen reported no conflicts of interest.
REFERENCES
1. Chen J, Gale RP: Physician-complementing artificial intelligence in oncology. The ASCO Post. Available at https://ascopost.com/news/october-2025/physician-complementing-artificial-intelligence-in-oncology. Accessed May 7, 2026.
2. Gilbert S, Dai T, Mathias R: Consternation as Congress proposal for autonomous prescribing AI coincides with the haphazard cuts at the FDA. NPJ Digit Med 8:165, 2025.
3. Landry L: How to delegate effectively: 9 tips for managers. Available at https://online.hbs.edu/blog/post/how-to-delegate-effectively. Accessed May 7, 2026.
Dr. Chen is Professor at the Institute of Hematology, Chinese Academy of Medical Sciences. Dr. Gale is Visiting Professor of Hematology, Imperial College London.
