When every client arrives with a ChatGPT diagnosis, does the vet's job move from diagnosing the animal to adjudicating the owner's AI?

The shift

Plausible medical synthesis of a pet's symptoms — the thing owners used to have no way to produce and could only buy from the vet — goes from scarce to free and instant. But the animal still can't self-report, so the model is reasoning from what the owner typed, not from the patient. The vet's information monopoly breaks; the physical access to the actual patient does not.

The axioms

• The vet is the sole source of medical knowledge about the animal — the visit exists to transfer what only the vet knows.
• Diagnosis starts from a blank slate: the owner brings symptoms, the vet supplies the interpretation.
• The consult is where information gets delivered, so it's structured as the vet explaining what's wrong.
• The patient can't describe its own symptoms, so someone must read a non-verbal body — palpate, observe gait, interpret behavior the owner normalized.
• Treatment is physical: the exam, the bloodwork, the extraction, the spay, the needle.
• A licensed vet must own the diagnosis and be accountable for the treatment call.
• The owner's trust in the vet is what converts a recommendation into consent and payment.
• The owner is a channel for the vet's judgment, not a party holding a competing diagnosis.

Invalid axioms

1. The vet is the sole source of medical knowledge about the animal. A model will produce a differential, a drug-dose estimate, and a prognosis from a paragraph of symptoms, for free, before the appointment. The habit-trap: the practice still prices and structures the consult as knowledge transfer — the fee, the appointment length, and the vet's self-image all assume the owner arrives empty-handed and leaves informed.
2. Diagnosis starts from a blank slate. The owner now arrives anchored on a specific answer, emotionally invested in it, and often halfway to a decision ("I read it's probably lymphoma, so I want to know about treatment"). The habit-trap: the consult is still designed to build a diagnosis forward from history-taking, when the real starting point is now an existing diagnosis the vet has to engage with — the intake doesn't ask "what have you already concluded?"
3. The consult is where information gets delivered. Explanation is the abundant thing now; the owner had the explanation before they walked in. The habit-trap: vets still spend consult minutes explaining the mechanism of the disease as the core deliverable, when the scarce act has moved to reconciling the owner's version with the exam.

Unchanged axioms

1. The patient can't self-report, and the model can't touch it. Every AI diagnosis is built on the owner's secondhand description — "he's been off his food," "she's limping" — which is exactly where lay observation is unreliable. The scarce act is getting hands on a non-verbal body: palpating an abdomen, feeling a mass the owner couldn't, spotting that the "limp" is neurological. This is the sharpest line in veterinary medicine specifically, because the information source the model needs is the one thing it structurally can't access.
2. Treatment requires physical action on a physical animal. The exam, the radiograph, the blood draw, the surgery, the euthanasia. None of it is token generation, and an anxious, unrestrained animal makes it harder, not easier, than the human equivalent.
3. A licensed, accountable vet must own the diagnosis and the treatment call. Liability doesn't transfer to a chatbot. When the animal is harmed, a named professional answers for it — and being confidently wrong causes direct physical harm, not a redo.
4. Reading a non-verbal patient is judgment, not lookup. Interpreting gait, body condition, a subtle behavioral shift the owner normalized, the significance of what the animal isn't doing — this is pattern-matching against a physical presentation the model never sees, on a species where "the patient seems fine" from an owner is frequently wrong.
5. The owner's trust is what converts a recommendation into consent. The owner pays, decides, and lives with the outcome for a creature they love and can't consult. Trust built through relationship — being believed, being treated as competent, having the vet clearly on the animal's side — is what carries the hard calls, and it doesn't come from the quality of an explanation.

New axioms

1. The consult now opens with adjudication: is the owner's AI answer right, wrong, or wrong-in-a-way-that-matters? The vet's first move is triaging someone else's diagnosis under emotional load — before any exam, they're managing an owner anchored on a confident answer that the animal's actual body may contradict. No consult model, appointment length, or fee structure accounts for this as the opening task.
2. Correcting a confident-but-wrong AI carries emotional labor the old job didn't. Telling an owner their pet doesn't have the terrifying thing the model suggested is a relief; telling them the reassuring AI answer was wrong and it's worse than they think is the hard version. Either way the vet is now managing the gap between what the owner already believes and what the exam shows — and dislodging an anchored belief costs more time and trust than delivering a fresh one.
3. Liability shifts when the owner insists on the AI's answer. The owner who refuses a recommended workup because "ChatGPT said it's just a diet issue," or who demands a drug the model named, creates a new failure mode: the vet is accountable for the outcome but no longer the sole author of the decision. Consent, refusal, and documentation frameworks were built for an owner without a competing diagnosis.
4. When the AI is anchored on the owner's inaccurate report, the vet has to debug the input, not just the output. The model said "probable arthritis" because the owner described a limp; the exam says neurological. The scarce skill becomes knowing which owner observations to distrust and re-derive from the animal — a step the owner assumes the AI already handled.
5. Trust now has a rival source of authority in the room. When the vet's read disagrees with the owner's AI, the practice needs a way to win that disagreement without condescension — the owner can re-query the model in real time and get reinforced. Nothing yet governs how a vet establishes standing over a tool the owner trusts and can consult mid-consult.

Where it breaks

"The consult is where information gets delivered" (invalid) collides with "the consult now opens with adjudication" (new): the vet is still booked, timed, and paid as though the deliverable is explaining the disease, but the actual opening work is arguing with a diagnosis the owner already trusts — an unbilled, emotionally heavier task wedged into an appointment slot sized for the old job.

A second collision: "the owner's trust converts a recommendation into consent" (still holds) meets "the owner insists on the AI's answer" (new). The whole model assumes trust flows to the vet by default because the vet knew more. Now a second authority sits in the room that the owner can re-consult mid-appointment, and when it contradicts the exam, the vet is fully accountable for an outcome while sharing authorship of the decision with a tool that can't be — with no framework for who owns the call when the owner overrides the hands that actually touched the animal.

ai-axioms.vercel.app/axioms/healthcare/veterinary-ai-axioms