The patient sitting across from the oncologist has just heard the cancer is spreading. Until a few moments ago, they were discussing treatment options. Now they are discussing time.
Chemotherapy is available. It may prolong this disease’s course and perhaps buy more time, but it will also mean more appointments, more blood draws, more nausea, and likely more time in the hospital. The oncologist recites the statistics, but the patient is starting to hear them differently.
After an uncomfortable pause, they ask a question no survival curve can answer: “If I do this, will I still feel like myself?”
Imagine how useful artificial intelligence might become at this moment. AI could survey thousands of similar cases, aggregate genomic data with the patient’s medical history, identify treatments the doctor hasn’t considered, and provide ever-more-accurate estimates of both survival and toxicity. Maybe someday it could simply tell the doctor that treatment X will give this patient a 14 percent better chance of living another year. Information like that would be a genuine medical accomplishment. More accurate predictions can and do save lives.
But it still won’t help with that question.
To the patient, whether another year of life is worth the costs needed to achieve it is no mere variable awaiting calculation by a powerful enough computer. It is a question about what this particular person deems worth extending. Another patient might want every conceivable intervention that can prolong life. This patient might value remaining conscious enough to recognize their spouse, attending a daughter’s wedding, or simply spending their last months at home.
Medicine therefore faces a strange limit: no matter how much we know about what will happen we cannot by itself know what should be done.
This distinction between technical accomplishment and the good of the patient matters because one of the strongest arguments for accelerating artificial intelligence is the claim that AI will one day “solve” diseases like cancer. It has become such a familiar refrain as to sound almost like common sense: if human scientists have been trying without success for decades, then superintelligent AI will figure it out. Anthropic’s Dario Amodei has predicted that AI-augmented biology will allow us to make fifty to one hundred years’ worth of biological discoveries in just five to ten years. 1
The co-founder of OpenAI has made a similar claim about beating cancer when describing the benefits of ever-more-powerful AI. 2
Radiologist Emilia Javorsky recognizes why such statements are appealing: if preventing cancer becomes a realistic outcome of developing artificial superintelligence, then arguments for slowing that development begin to acquire an ethical cost of their own. Cautions about risks, regulation, or redirecting resources away from AI become no longer just impediments to technological progress but impediments to saving lives 3
This is why cancer is such a powerful rhetorical tool. Cancer is viscerally real to the public in a way that most other theoretical harmful uses of AI are not. Virtually everyone has seen a friend or loved one suffer through chemotherapy, await biopsy results, celebrate remission, or pass away from the disease. Promise cancer cures and watch crowds of previously skeptical onlookers become enthusiastic for technological progress. Promise to end cancer and watch them open their wallets.
Few anecdotes speak to the fears and hopes of so many as curing cancer. Javorsky knows this personally. When she was in medical school, her father sought treatment for what he thought was a chronic cough. He was diagnosed with esophageal cancer that had already begun to spread. After receiving a poor prognosis and reviewing difficult treatment options with his doctors, he chose to prioritize quality of life through radiation over chemotherapy and surgery. He lived another two years relatively healthy. But when the cancer began to metastasize to his brain, his oncologists offered another ray of hope. State-of-the-art proton therapy just might allow him to come home for Christmas if he started treatment immediately. He died in the hospital one week later. Javorsky recalls both the promise and its failure with equal clarity. 4
Her response is not to become afraid of medical progress. It is becoming wary of easy promises made in its name. This makes her critiques of claims that artificial superintelligence will “solve” cancer all the more important. There is no reason we have not already solved cancer because we lack enough intelligence to do so. Instead, Javorsky argues that time is constrained by incomplete data, biological complexity, years of required clinical experimentation, regulatory considerations, economic incentives, failures of coordination between institutions, and more. Living organisms do not improve according to Moore’s Law. You can’t run a clinical trial in your mind overnight simply because your computer does. Cancer is happening to humans, and humans have certain temporal and biological realities that AI cannot magically overcome. 5
This is not to say artificial intelligence has nothing to offer oncology. Far from it. We are already beginning to put AI systems to work on problems like new drug discovery, toxicity prediction, personalized treatment response, and improving clinical-trial efficiency. 6 Javorsky’s claim is more nuanced: we see substantial progress when AI is directed toward well-defined problems and identifiable bottlenecks where adding more computation and pattern recognition can make a clear difference. The danger is in assuming a specific application of AI that can improve cancer outcomes will ipso facto lead to a generally intelligent system that can cure cancer. The latter is a huge extrapolation supported by little evidence beyond our intuition that cancer involves patterns currently too complex for us but not for AI. The former is already here.
This brings us back to the patient asking whether the treatment will allow them to continue to be themselves. If curing cancer became a realistic near-term outcome of AI research, we would still face the problem that cancer is not merely a pattern to be optimized away. Javorsky points out that AI tends to do best at tasks with clear rules and win conditions, large amounts of available data, objective goals, and low latency between actions and feedback 7 Medicine suffers from none of these. Clinicians still work with incomplete information, cannot control for every variable that might affect health, face tremendous complexity in human biology, and deal with solutions that might take months or years to manifest successes or failures. Most critically for our purposes, medicine does not have clearly objective goals. Medicine can attempt to maximize years of survival, minimize toxicity, shorten time spent in hospitals, or identify the treatment with the best chance of shrinking tumors. But how we weigh these goods against each other depends upon knowing which is most important for this patient’s health. And we can only know that by understanding what healing will look like for them.
Why does this matter? Let medical ethicists Pellegrino and Thomasma answer.
The relationship of trust begins with the sick person’s good entrusted to the professional. Trust is reliance upon another’s competence and commitment to further what is entrusted to him rather than to harm or misuse it. Trust necessarily includes both vulnerability and discretionary power. The trustee is vulnerable in that he must rely upon another. He also has discretionary power in how to exercise his skills on behalf of the trustor’s good. 8
The vulnerability created by illness and the patient’s subsequent need to trust someone competent to heal becomes all the more important with cancer. Cancer might give patients fewer opportunities to independently assess genomic data, pathology reports, treatment risks, survival probabilities, or responses predicted by machine learning. The patient will need to trust someone. But that does not mean every use of AI in oncology is an ethical use of AI.
Trust is not the same thing as fidelity. An algorithm might someday become more trustworthy than a physician at predicting which cancers will respond to which therapies. There would be nothing unethical about preferring it for that task. But notice how Pellegrino and Thomasma describe the nature of trust: it is a relationship in which someone’s vulnerability becomes the moral responsibility of another. To be faithful to that trust requires medical knowledge and power be exercised for the sake of that patient’s good. Without fidelity, the relationship between doctor and patient can become something worse than useless. It can become exploitative. 9
AI can never be faithful in this sense. An algorithm can be made more reliable but not more caring. It cannot recognize vulnerability as a morally significant fact about the world, nor can it appreciate the responsibility of responding to that vulnerability. The limitation on AI is therefore not computational power or improvements therein. We need not wait for more powerful computers to realize this limitation because it is not computational.
We need medicine.
Compassion allows that vulnerability to become knowledge. For Pellegrino and Thomasma, compassion is not simply feelings we supplement to clinically excellent care. Compassion is the virtue by which the suffering of another person is permitted to influence the judgment of the healer. 10 It prevents us from defining this patient by the cancer that needs curing. Cancer treatments do not inherit the patient’s goals, and an intervention that is objectively successful at curing cancer can be objectively horrific for the person being cured. Without compassion, we have no way of understanding how this illness and its treatment enter into the life of the person who has cancer. Healing does.
Artificial intelligence will allow us to learn more about cancer and develop better predictions about how treatments will work. But we will still need to know the patient in order to understand what counts as healing. When a machine tells the oncologist that Treatment A has a slightly better probability of extending survival than Treatment B, compassion forces us to consider what that additional survival will cost this patient. It is not sentimental. It is part of knowing whether a successful treatment is truly good for the patient. Trust and compassion therefore guide us toward what Pellegrino and Thomasma identify as phronesis.
Medical facts do not vanish when physicians begin moral deliberation. But practical wisdom tells us how those facts should bear on our decisions in particular cases. 11 Increasingly capable AI will not make the practice of medicine any less in need of good judgment. If anything, it will make good judgment all the more important.
The problem with claiming accelerated AI will cure cancer is not just that it misunderstands cancer; it misunderstands medicine. One imagines a single disease to be solved by the right technical expertise. But curing cancer will never absolutely “solve” the problem of cancer care. As long as there are patients sitting across from oncologists, there will be times when we know how to prolong life but cannot know what kind of life a patient will want to prolong. At that point we must trust the patient knows their life better than we do. Unfortunately, appealing to future cancer patients cannot absolve current institutions from the responsibility of justifying how AI will actually benefit the sick.
AI will transform oncology. To hope otherwise is foolish. Let us use that intelligence to transform medicine for the better.
Notes
- Emilia Javorsky, “How AI Can, and Can’t, Cure Cancer,” AI vs Cancer, March 16, 2026. Javorsky discusses Dario Amodei’s prediction that AI-enabled biology and medicine could compress fifty to one hundred years of biological progress into five to ten years and situates it within the broader promise that artificial superintelligence will produce major medical breakthroughs. Recent reporting also documents similar cancer-cure claims associated with Amodei, Sam Altman, and Demis Hassabis.
- Javorsky argues that once cancer cures are attached to the pursuit of ASI, regulatory and resource constraints can be framed as unconscionable impediments to medical progress. She also raises the opportunity costs created by the enormous capital flowing toward ASI rather than existing biomedical research.
- Javorsky recounts her father’s diagnosis, his decision to prioritize quality of life, the later metastasis to his brain, and the ultimately false hope that proton therapy would allow him to return home by Christmas.
- Javorsky argues that biological systems impose constraints that cannot simply be overcome through additional computation. In particular, clinical research remains bounded by the timescales of cells, organisms, disease progression, and human populations.
- Javorsky identifies existing applications of AI to drug discovery, toxicity prediction, and clinical-trial efficiency as examples of targeted applications capable of addressing real bottlenecks in oncology.
- Javorsky contrasts domains favorable to AI with medicine, which involves incomplete information, stochastic outcomes, physical constraints, delayed consequences, and subjective “win conditions.”
- Edmund D. Pellegrino and David C. Thomasma, The Virtues in Medical Practice (New York: Oxford University Press, 1993), 67. Their discussion of trust emphasizes competence, vulnerability, discretionary power, and confidence that the person entrusted with another’s good will not exploit that vulnerability.
- Pellegrino and Thomasma, The Virtues in Medical Practice, 75. They argue that fidelity to trust remains a central obligation of professional relationships and that its absence turns vulnerability into the possibility of exploitation rather than healing.Pellegrino and Thomasma, The Virtues in Medical Practice, 79–83. Their treatment of compassion emphasizes its intellectual and moral dimensions and its relationship to the healing encounter.
- Pellegrino and Thomasma, The Virtues in Medical Practice, 90. They describe clinical decisions as requiring the integration of scientific and moral reasoning and identify prudence as indispensable to this form of clinical judgment.
- Javorsky does not reject AI in oncology. Her proposed alternative is to support and scale targeted AI tools while addressing the systemic bottlenecks that limit medical progress.
Views and opinions expressed by authors and editors are their own and do not necessarily reflect the view of AI and Faith or any of its leadership.
