Modern medicine rests on an assumption that with enough data (decades of clinical studies, millions of patient records, and now the immense pattern‑matching power of artificial intelligence), medicine can finally outsmart the body’s complexity.

Yet patients continue to cycle through specialists, accumulate diagnoses, and take ever‑more medications while their underlying health declines.

I think the reason is not a lack of enough data, but that both the academic data doctors are trained on, and the mass data that powers ‘proxy medical’ AI, operate with the same bias or incomplete mapping of the human body. As a result, they both have chances to lead to specific kinds of failures: misdiagnosis, treatments that make things worse, unsustainable costs, and the exhaustion of patients who are told by a second or third doctor that the first two were wrong.

The following is my view as a general patient, and as someone who has observed many other patients experiencing a specific case of cascading failures in medical treatments.

1. Diagnostic Knowledge, Data & Tests

The knowledge that guides medicine comes from two sources in the modern times.

The first data source is the body of knowledge doctors acquire in training and practice: textbooks, clinical trials, peer‑reviewed literature, and the accumulated experience of specialists. This data is filtered through a fragmented view of the body. Cardiology knows the heart; endocrinology knows hormones; gastroenterology knows the gut. The connections between these systems are treated as secondary, often discovered only when a patient’s experience contradicts the textbook.

The second is the mass data used to train artificial intelligence: electronic health records, billing codes, lab results, and millions of de‑identified patient journeys. AI learns to replicate the patterns it finds in this data, including the very fragmentation and errors that exist in routine practice. If most doctors treat a certain set of symptoms with a particular drug, the AI will recommend that drug. If the system routinely mislabels a medication side effect as a new disease, the AI will learn to do the same.

In both cases, the data reflects only what medicine measures, not what it misses. A doctor’s differential diagnosis is shaped by the specific specialties they were trained in. An AI’s recommendation is shaped by the narrow fields in the electronic record like labs, vitals, or prescription history, while the patient’s lived experience, the sequence of events, and the subtle ways one treatment altered the body’s response to the next are either missing or reduced to disconnected codes.

Inconclusive Test Results: Then a known and documented discordance is that medical tests evaluate readings within “normal range”, which makes the tests show Normal while body doesn’t function normally.

Standard reference ranges are population averages. “Normal” means within the range of the general population, not optimal for someone’s specific baseline before damage, and not accounting for:

– Any pre-medication baseline which is unknown because it was never tested
– Functional thresholds varying individually, e.g., some people need TSH at 1 to function, others at 3, but both “normal” on paper
– Some tests e.g., prolactin is frequently not included in standard hormone panels unless specifically requested
– Neural situations e.g., dopamine pathway disruption doesn’t have a direct test
– Metabolic set-point changes don’t appear as abnormal insulin on a fasting test if the issue is receptor sensitivity

This often puts patients in situations where:
– Doctors say tests are normal, therefore nothing is wrong
– But body clearly demonstrates something is wrong
– Standard testing cannot see the actual problem or the root cause
– Changing treatments didn’t help with visible symptoms or added more

2. Cascade: When the Treatment Becomes the Problem

One of these failure patterns is the cascade that begins with a reasonable intervention and ends with a patient far sicker than when they started, yet still under the care of a system that insists on adding more treatments.

Example:
· Linear Assumption: Drug A treats Symptom B.
· Non-Linear Reality: Drug A alters the overall state of the body, making its response to all future inputs (food, stress, other medications) fundamentally different from the population average.

Consider a few examples of this failure pattern:

· A misdiagnosed root cause: A patient presents with fatigue, weight gain, and brain fog. The first primary care doctor diagnoses depression and prescribes tests. When the fatigue worsens, a psychiatrist adds a stimulant. When anxiety develops, a second psychiatrist adds sedatives. Two years later, a different physician runs a comprehensive thyroid panel and discovers thyroiditis. The original cause was autoimmune hypothyroidism, but by then the patient is dependent on three psychotropic medications, each with its own metabolic and neurological side effects. The system treated the downstream symptoms while the root cause went unidentified.

· A specialist cascade: A man with mild psoriasis sees a dermatologist who prescribes a biologic. The biologic works well for the skin but, months later, the patient develops recurrent sinus infections and a persistent cough. He is referred to an allergist, who diagnoses new‑onset asthma and prescribes inhaled steroids and a rescue inhaler. When the patient gains thirty pounds and his blood sugar rises, he is sent to an endocrinologist, who diagnoses type 2 diabetes and starts metformin. No one connects the cascade back to the biologic, and the patient now carries three chronic disease labels that did not exist two years prior.

· Psychiatric: A patient is prescribed changing antipsychotics for what looks like symptoms of depression or mood instability. The medicine set affects thyroid function or raises prolactin levels, leading to hormonal dysfunction and significant weight gain. The system then treats the “new” weight with metabolic drugs & the “new” hormonal issues with hormone control, never resolving the neurotransmitter damage from the original prescription. The patient visits a second doctor who now identifies that the antipsychotics were not necessary since the patient was autistic or ADHD, simply requiring different care.

· Conflicting expert opinions: A young man with chronic abdominal pain sees a gastroenterologist who performs a scope, finds nothing conclusive, and diagnoses irritable bowel syndrome. The prescribed medicines provide no relief. A second gastroenterologist suggests small intestinal bacterial overgrowth and prescribes antibiotics, which temporarily improve symptoms but leave him with worsened bloating and new food intolerances. A third doctor, a functional medicine practitioner, points to a history of antibiotic use and poor gut motility, recommending a low diet and prokinetic agents. By this point the patient has spent money, undergone three conflicting treatment protocols, and remains with the original condition.

Once the body has been loaded with conflicting chemicals and shifted into a defensive state, the data becomes useless because the new data points (blood markers, weight, hormone levels) are no longer signals of a simple “disease,” but rather markers of structural damage that the system is not designed to measure.

In each case, the interventions were rational given the fragmented data each doctor saw. But because doctors or AI often are unable to connect the dots across time and specialty, the treatments are accumulated, the side effects multiplied, and the patient’s overall trajectory worsened.

The cascade is a recursive failure.

The medical system’s structure tends to maintain its own assumptions. When a patient experiences a net-negative trajectory despite ongoing treatment, the fragmented system does not conclude that the model is broken. Instead, it classifies the patient as a “non-responder,” adds new diagnoses, or attributes the iatrogenic damage to a “new condition.”

The medical system and by extension the AI are both also biased to assume every patient has the average statistical neurological or body type, which makes outliers at a greater risk of treatments which harm more than help.

This creates a recursive loop, seen across multiple clinical scenarios.

3. The Cost and the Fatigue

These failures carry a heavy toll beyond health outcomes. The financial burden grows as multiple specialist medications are added but rarely removed, expensive diagnostic procedures that yield little actionable insight, and, for many, a slow erosion of personal stability as chronic illness takes hold.

The mental and emotional toll is equally profound. Patients learn to navigate a system where each new doctor starts from scratch, where their own detailed history is often dismissed as “too complicated,” and where they are implicitly expected to be grateful for interventions that leave them worse off. The experience of being told by a second doctor that the first doctor was mistaken, and by a third that both were wrong, creates a deep sense of uncertainty. Many patients begin to feel that the system is not designed to solve their problem but to manage their participation in it.

4. Why AI Cannot Break the Cycle

There is a widespread belief that artificial intelligence, trained on vast datasets, may overcome the limitations of human doctors. But AI learns from the very same fragmented, incomplete records that human practitioners rely on. It cannot see what was never documented.

Moreover, AI is trained to reproduce the most statistically common patterns in the data. If the dominant pattern is to treat elevated blood sugar with medicine A, the AI will recommend medicine A, even in cases where the patient’s problem originates from another cause, or if their body does not respond positively to this treatment. If the data shows that most patients with a given symptom cluster receive a certain psychiatric diagnosis, the AI will propagate that diagnostic bias, even when the symptom cluster has an organic origin.

AI does not solve the problem of fragmented care; it scales it. It offers the same narrow solutions at greater speed and with the added authority of algorithmic certainty, making it harder for patients to argue with a recommendation that is presented as statistically optimal.

5. The False Certainty

I believe the first step is acknowledging that the current data, both the knowledge base of medical education and the massive datasets used to train AI, has not yet “solved the human body” with 100% certainty.

It is a knowledge system that has partially mapped many mechanisms, has interventions that work probabilistically across populations, and operates on incomplete models. But the medical system has not solved the human body completely yet.

The confidence with which it presents itself, including the “consult a doctor” directive in every conversation, is not proportional to its actual resolution rate for complex or chronic or iatrogenic problems.

Where it does work:
Acute trauma, infection with identified pathogen, some surgical problems, some single-mechanism diseases. These are areas where the intervention-to-outcome link is clean and the model is accurate enough to be useful.

Where it has chances for cascading failures:
•Multi-system chronic conditions
•Iatrogenic damage (the system has no good framework for undoing what it caused)
•Conditions where the patient is the outlier from population averages
•Neuorologically different patients who are frequently misread, overmedicated, and whose symptom reports are then treated superficially or dismissed

Until the underlying framework changes, both doctors and AI will continue to offer the same kinds of fragmented, reactive care. They may treat the blood sugar without asking why it rose. They may manage the insomnia without investigating what destabilized sleep or identifying the neurological type of the patient.

They may add medications to treat the side effects of other medications, all while the patient accumulates new diagnoses, mounting bills, and the feeling that the medical system is not functioning to fix the root cause, but only to keep responding to its own uncertainty.