I was diagnosed with benign paroxysmal positional vertigo (BPPV). The diagnosis was not surprising. I had experienced several brief but intense episodes of vertigo, sudden, disorienting sensations that made it difficult to remain upright.
During one episode, I slumped against a bookcase. I did not lose consciousness, but the event was concerning enough that my physician suggested enhanced head and neck MRIs to rule out vascular causes. I have no fear of MRI machines. I have undergone many scans in my life. Yet after leaving the appointment, I felt unsettled, not by the imaging itself, but by what it represented.
I take sedative medications at bedtime for chronic insomnia and anxiety. I am also neurodivergent and live with generalized anxiety disorder. It seemed to me that a far simpler explanation was available: positional vertigo, possibly compounded by medication effects and ordinary imbalance. The proposed imaging, however, was aimed at searching for far less probable but more catastrophic possibilities.
I began thinking not only as a patient, but as someone trained to reason under uncertainty. What is the base rate of a dangerous arterial abnormality presenting this way? What is the probability that imaging would reveal something clinically actionable? What is the false positive rate? What is the likelihood of incidental findings, abnormalities that are technically “real” but medically irrelevant?
And perhaps most importantly for someone with anxiety: What is the psychological cost of opening diagnostic doors that may not need to be opened?
After researching these questions carefully, I wrote to my physician and declined the imaging. I explained that I was not refusing care. I was refusing a test whose expected value, for me, did not justify its potential harms, especially the harm of non-actionable findings that could generate months of distress. This was not an act of defiance. It was an act of reasoning.
That experience illustrates a deeper issue in modern health care: The uneasy relationship between medicine, probability, and power. Note that this article was structured by me, but the composition was aided by ChatGPT.
Defensive medicine and the expansion of testing
Physicians today practice in an environment saturated with liability pressure. Missing a rare but catastrophic diagnosis can have severe professional consequences. The incentive structure therefore often favors ruling out worst-case scenarios, even when their probability is extremely low. This phenomenon is widely known as defensive medicine.
From the physician’s perspective, ordering an MRI may feel prudent. From the patient’s perspective, it can initiate a cascade: incidental findings, follow-up scans, specialist referrals, biopsies, and sleepless nights.
The harms of testing are not limited to procedural complications. They include psychological distress, false alarms, overdiagnosis, and overtreatment. In statistical terms, modern medicine often prioritizes minimizing false negatives (missing something) at the cost of increasing false positives (finding something meaningless).
For many patients, that tradeoff may be acceptable. For others, particularly those predisposed to anxiety, it is not trivial. When defensive logic dominates, patients risk being treated as potential malpractice scenarios rather than as whole persons with differing tolerances for uncertainty. This is not a moral failure of individual physicians. It is a structural feature of the system in which they work.
Iatrogenic illness and the cost of speed
Medicine contains a sobering truth: Medical care itself can cause harm. Iatrogenic illness, illness caused by medical intervention, is not rare. It includes medication side effects, procedural complications, hospital-acquired infections, unnecessary surgeries, and the psychological burden of overdiagnosis.
At the same time, primary care physicians are often given 15 minutes per patient. Within that narrow window they must review history, interpret symptoms, document electronically, consider guidelines, manage liability, and move to the next room. There is little time for probabilistic reasoning. Little time to ask: What is most likely? Little time to incorporate the patient’s psychological context. Little time for shared deliberation.
Speed and defensive caution combine in a troubling way. The result can be a mechanized style of care: test broadly, rule out catastrophe, move on. The human cost of this approach is rarely visible in billing data.
AI, diagnostic reasoning, and patient agency
Recent research has suggested that large language models can perform impressively on certain diagnostic reasoning tasks in structured comparisons with physicians. This does not mean that AI should replace doctors. AI cannot perform physical examinations, cannot assume responsibility, and remains susceptible to error.
But it does suggest something important: Access to medical reasoning is no longer confined exclusively to professionals.
Tools such as ChatGPT can help patients explore differential diagnoses, base rates, test characteristics, and guideline-based questions. Used carefully, they may reduce information asymmetry and help patients participate more actively in their own care. No one should be a passive recipient of medicine.
Patients should not be expected to comply blindly with every test recommendation, nor should physicians be expected to practice alone under crushing time pressure. The goal is partnership: informed dialogue rather than unilateral authority. AI is not a substitute for clinical expertise. But it may serve as an educational aid that helps patients ask better questions.
Probability as a form of self-protection
I work in machine learning, computer vision, and pattern recognition. When I evaluate a medical recommendation, I look for specific quantities:
- What is the mortality risk of the procedure?
- What is the false positive probability?
- What is the probability of a correct diagnosis?
If those numbers are unavailable, vague, or unsatisfying, I ask for clarification. If the expected value remains low relative to the costs, physical or psychological, I decline and seek a second opinion. This is not rejection of medicine. It is Bayesian reasoning applied to one’s own body.
Medicine often speaks in categorical language: “We need to rule this out.” But the world is probabilistic. Tests do not eliminate uncertainty; they merely shift likelihoods. Humanizing medicine means acknowledging that patients may rationally choose to live with a small degree of uncertainty rather than pursue every improbable possibility. It means respecting informed refusal as much as informed consent.
Power in the exam room
The modern exam room contains an asymmetry of knowledge, authority, and institutional backing. Historically, patients have been expected to comply. But power without shared reasoning can drift toward paternalism. Even benevolent paternalism can produce unnecessary interventions.
Rebalancing power does not require diminishing physicians. It requires recognizing patients as cognitive agents capable of engaging with probability, tradeoffs, and risk tolerance. Different patients will make different decisions. That diversity is not dysfunction; it is humanity.
Toward a more human system
There are many dedicated, compassionate physicians who practice careful medicine under immense pressure. This essay is not an indictment of individuals. It is a critique of incentives.
A more human system would:
- reduce liability pressures that drive defensive testing
- allow physicians more time per patient
- normalize probabilistic discussions of risk
- respect informed refusal
- encourage responsible use of AI as an educational support tool
Humanizing medicine does not mean rejecting science. It means applying science, including probability theory, without losing sight of the person sitting in the chair. Medicine, at its best, is not the elimination of all uncertainty. It is the wise management of it. And wisdom requires shared power.
Martin Bello is an expert on computer vision and machine learning.
