Thinking Is Not A Procedure
How safetyism and medical politics would have kept me sick and ignorant forever
I knew the objections were coming before I hit “publish.”
If you say out loud that an AI helped you understand a lifetime of medical weirdness, three camps appear almost instantly:
the existential-risk crowd,
the regulate-everything crowd, and
the guild of people whose authority you just stepped on.
So this is the follow‑up I would rather write before they speak, while the signal is still visible under the noise.
1. “The AI Just Told You a Convincing Story”
The first objection goes like this:
“You got seduced by a machine that’s good at making up plausible stories. It hallucinated a mechanistic fairytale; you believed it because you were desperate.”
Let’s strip that down.
What did the AI actually do in my case?
It read fifty years of symptoms and context that no human clinician ever had time to read.
It proposed a mechanistic model that fits:
my cold/warm sensitivity
my local anesthetic paradox
my peanut-triggered flares
my partial response to antihistamines
my history of early-life toxin exposure
It described a small‑fiber, sodium-channel, mast-cell mediated pattern that is:
internally coherent
biologically plausible
consistent with published phenomena
Could it still be wrong in details? Of course. Biology is messy. But if you want to call this “mere storytelling,” you have to meet a basic standard: produce a competing story that fits the data better.
No one has.
For decades, the best I got from the human system was:
“neuropathic pain of unclear etiology”
“idiopathic urticaria”
“have you tried reducing stress?”
If the bar for “not hallucinating” is that, then the entire medical system is one big hallucination generator with CPT codes attached.
There is a meaningful difference between:
fabrication that contradicts available evidence, and
hypothesis-building that connects existing evidence into a coherent model.
The AI did the second. If you’re going to dismiss it as seduction by narrative, you need to explain why its narrative is less valid than the hand‑waving I got in exam rooms for forty years.
If your answer is “because a licensed human said the hand‑waving,” that’s not an epistemic argument. That’s guild loyalty.
2. “This Is Unlicensed Diagnosis And Should Be Illegal”
The regulatory objection is simpler and colder:
“This is medical advice from an unlicensed entity. High‑risk. Should be prohibited. End of story.”
Here is what actually happened:
The AI did not prescribe anything.
It did not say “you have disease X, start drug Y.”
It built a mechanistic map and named a pattern of wiring: small‑fiber hyperexcitability with likely sodium-channel involvement and mast-cell cross‑talk.
That is not a prescription. That is thinking.
Regulators keep trying to erase the line between “thinking” and “acting” because their tools only work on actions. Licensure applies to:
performing a procedure
signing an order
making binding treatment decisions
It does not, and should not, apply to:
assembling mechanistic hypotheses
summarizing literature
mapping long trajectories of symptoms into a model
If you regulate cognition as if it were intervention, you have created a system where only licensed humans are allowed to think deeply about human biology.
And then you put those licensed humans on seven‑minute timers.
That is not patient safety. That is intellectual monopoly. And is deeply stupid.
If your regulatory theory would have forbidden what happened in my case, then you are defending this outcome:
A chronic patient with a nonstandard pattern is never allowed to access deep, integrative reasoning about their own body, unless and until a human with no time and incomplete data decides to provide it.
That is the world you are building if you say “this kind of reasoning must be banned unless delivered by a clinician.”
I am not asking to let uncensored models:
prescribe chemotherapy
adjust anticoagulation
manage insulin pumps
I am asking something far simpler and more dangerous to your control model:
Let them think and speak clearly about mechanisms, as inputs to human judgment.
If your safety regime cannot tolerate that, it is not about safety. It is about control.
3. “You’re Undermining Doctors And Encouraging Self‑Diagnosis”
From within the medical guild, the objection sounds like this:
“Stories like yours will make patients trust AI more than physicians, increase confusion, and erode the doctor–patient relationship.”
No. What erodes trust is being dismissed for decades because a system has no place to put your complexity.
The AI did not replace a diagnosis I already had.
It replaced nothing.
For years, the choice was:
vague labels that explained nothing, or
being gently told it was probably anxiety.
So I stopped talking about my symptoms. I brought my quiet suffering to the internet and to private experimentation instead. That is what erosion of trust looks like.
If AI‑assisted reasoning threatens the doctor–patient relationship, that says something unpleasant about the current relationship: it’s brittle. It cannot tolerate a third intelligence in the room.
Healthy relationships do not fear an extra mind.
The correct role for a clinician in this future is:
hands‑on examiner
pattern checker
risk-balancer
interpreter and critic of mechanistic models, whether human‑generated or machine‑generated
That is still indispensable.
But it will not be the only source of thought.
If the guild wants enduring trust, it has two options:
Reclaim deep diagnostic cognition itself.
Welcome tools that supply it, and learn to critique them.
What it cannot do is insist that nobody else is allowed to think because the guild is too overloaded to do that part of the job.
4. “You’re Just Anecdotal. Policy Should Not Be Made From Stories.”
This one comes from people who love the word “evidence” but rarely look at which evidence is missing.
My story is an anecdote. That’s fine. That’s what early signals look like.
But the failure I’m describing is not anecdotal at all. It’s structural and well‑known:
chronic, multi‑system, weird‑pattern patients churn through clinics for decades
they get partial labels, lifestyle advice, or psychiatric framing
diagnostic yield drops fast after the first few visits
there is no mechanism in the workflow for “somebody sits down and thinks for six hours”
We do not run large randomized controlled trials on “allow thinking vs forbid thinking” because we assume thinking is free. It isn’t. It’s time, and time is precisely what the system removed.
So no, you should not pass national law because one guy on the internet wrote about mast cells and Nav1.7.
But you also cannot wave this away with “N=1” when:
the system’s inability to handle low‑prevalence, high‑complexity cases is baked into its architecture
the number of patients living in that failure mode is not N=1, it is N=millions
My case is just a particularly transparent instance of something the literature and the lived experience of patients have been screaming for years:
the cognitive part of medicine is collapsing under volume.
If a tool shows up that can partially re‑inflate that capacity, the correct response is not:
“Anecdote, therefore ignore.”
The correct response is:
“This looks like a pressure valve for a known failure mode. Let’s study it, instrument it, and avoid smothering it with pre‑emptive bans.”
5. “People Will Trust AI Too Much And Get Hurt”
This is the softer, paternalist version:
“Even if you are sophisticated enough to treat this as a hypothesis, many people aren’t. They’ll blindly follow AI output and make bad decisions. We need strong guardrails to protect them from themselves.”
Guardrails against what, exactly, in my case?
I did not stop any life‑saving therapy because of the model.
I did not start any dangerous drug because of the model.
I did not avoid necessary care; in fact, I now have a vocabulary to use with physicians that might finally get my case taken seriously.
The concrete outcome so far is:
better understanding of my own wiring
a more precise way to communicate with clinicians
a realistic menu of pharmacologic options that I then consciously declined
If you think this scenario is too dangerous to allow, you are implicitly saying:
“Patients must not be allowed to access clear, mechanistic explanations of their own bodies unless mediated by a licensed human who has time, which they largely do not.”
That is not a safety argument. That is a competence argument about patients.
It assumes they cannot handle hypotheses, only filtered conclusions.
If you want to mitigate risk, there are simpler, less authoritarian tools than “AI must be lobotomized”:
Prominent labeling: “This is a hypothesis, not a diagnosis. Do not change medications without a clinician.”
Built‑in prompts: “Here are questions to bring to your doctor. Here’s what to monitor.”
Logging and audit: so dangerous patterns of advice can be detected and corrected at scale.
Those are proportional responses.
“Ban deep reasoning because some people might misuse it” is not proportional. It is a blanket veto on cognition.
And again: we do not hold humans to that standard.
We allow lay books, forums, biohackers, relatives, and strangers to offer wildly variable health theories. We rely on:
free speech
social correction
professional second opinions
Yet for AI, suddenly the only acceptable risk is zero.
6. “This Will Lead To Black‑Box Medicine And Corporate Capture”
Another concern, often from the anti‑corporate side, is:
“By normalizing AI as a medical reasoning engine, you pave the way for black‑box corporate systems that insurers, hospitals, and governments will use to ration care and override clinicians.”
That risk is real. It has nothing to do with whether AI is allowed to do mechanistic thinking for patients, and everything to do with who owns and deploys it.
If you want to fight that, the right battle is:
demand transparency about training data and system behavior
demand patient access to the same level of reasoning tools that insurers and hospitals get
insist that when models are used in coverage decisions, the rationale is exposed and contestable
What you do not do, if you care about patients, is:
prevent individuals from using open or consumer models to think deeply about their own bodies, under the banner of “stopping corporate capture.”
That just guarantees a two‑tier world:
powerful, opaque models deciding what care you’re allowed to receive
neutered, over‑regulated models available to you as a citizen
If you are worried about black‑box control, you should be defending my use case, not attacking it.
My use case is: individual autonomy plus transparent reasoning.
7. What I Am Actually Arguing For (And Against)
Let me be painfully explicit.
I am not arguing for:
AI models autonomously prescribing drugs or adjusting dosages
replacing physicians in acute care, emergency triage, or complex procedures
treating model output as ground truth
skipping clinicians altogether
I am arguing for:
The right of patients to have powerful systems think deeply about their biology and present hypotheses in plain language.
Not curated into safe platitudes. Not flattened into “ask your doctor.” Full‑strength reasoning.A separation in law and policy between thinking and acting.
Thinking: constructing mechanistic narratives, listing plausible differentials, mapping exposures to outcomes.
Acting: diagnosing in the legal sense, prescribing, performing interventions.
Regulate the latter. Allow the former.
A future clinical workflow where:
patients can bring these machine‑generated models to clinicians
clinicians can critique them, refine them, or discard them
the AI becomes part of the cognitive ecosystem instead of being strangled at birth
A regulatory stance that measures harm from inaction, not just harm from commission.
The harm of:twenty more years of unexplained suffering
hundreds of thousands of similar patients never getting integrated reasoning
is not hypothetical. It is already here.
What I am arguing against is the reflex to:
treat any unsupervised medical reasoning as inherently unsafe
forbid models from getting specific, mechanistic, and honest about what might be happening in a human body
force them back into a script of generic advice that helps precisely no one in edge‑case territory
8. Before You Regulate, Talk To The People Who Were Left Behind
If you are a regulator, a professional association member, or an AI safety advocate, here’s who you almost never talk to:
the chronic “mystery” cases
the people who have been quietly dismissed as anxious, dramatic, or “somatic” for decades
the ones who stop coming back because nothing changes
These are not rare exceptions. They are what happens when you design a system around:
throughput
billing
risk minimization
instead of understanding.
I am one of them. The AI didn’t save my life. It did something subtler and, in its own way, more radical:
It restored the possibility that my body makes sense.
That is what you are tampering with when you say:
“This kind of reasoning must not be allowed unless we can guarantee it will never be misused.”
You are deciding that:
the comfort of institutions
the liability posture of agencies
the preservation of professional hierarchies
all outweigh the right of complex patients to have someone, somewhere, really think about them.
So as you draft rules and guidelines, ask yourself bluntly:
Are you designing a world where people like me ever get what I got?
If not, say it out loud. Don’t hide it behind “safety.”
9. The Quiet Trade You’re Proposing
When you strip all the rhetoric away, the proposal from my loudest critics looks like this:
“We will accept continued, large‑scale failure for complex patients in exchange for preserving our comfort with where thinking is allowed to happen.”
I am not willing to sign that deal on my own behalf.
I am certainly not willing to sign it on behalf of the millions whose bodies don’t fit billing templates.
If you are, then own it honestly:
Say: “Yes, we’re okay with people like you never getting an integrated mechanistic explanation, because the alternative scares us more.”
Say: “We prefer you remain in the dark than risk you reading a speculative but plausible map of your own nervous system.”
That is the real argument.
My answer to that is simple:
You don’t get to bury my answers to protect your sense of order.
Thinking about human biology is not a controlled substance.
Do not turn it into one.
Let the machines think.
Let the humans decide.