Symptoms and the spiral

This is the question everybody asks: “I have these symptoms, what is it?”

It is worth understanding exactly why that goes badly, because “don’t” has never stopped anybody at two in the morning.

What a diagnosis actually requires

A clinician diagnosing you has things a model does not:

Examination. Looking, listening, pressing, testing.

Your history, properly — not the four things you thought to type.

Base rates. How common each possibility is in someone of your age, sex, history and location. This is the big one, and it does most of the work in real diagnosis.

Tests, when the answer is not clear.

Pattern from thousands of patients, including the ones who looked like one thing and were another.

An assistant has your paragraph. From a paragraph, many conditions match — and the model has no way to weight them by how likely they are for you, because it does not know you and was not built to estimate that.

The base rate problem

The technical heart of it, and it explains the experience.

Common things are common. A headache is almost always a headache. But the frightening explanations are heavily represented in the text a model learned from — medical writing discusses rare serious conditions far more than it discusses ordinary ones, because there is more to say about them.

So the frightening possibility is not merely mentioned. It is available, in proportion to how much has been written about it rather than how likely it is.

That is why symptom questions so reliably surface the worst option, and why that option is so rarely the right one.

The spiral

The pattern is consistent enough to name.

You ask about a symptom. The answer includes something serious. You ask about that. The answer contains more detail, all of it frightening and all of it plausible-sounding. You ask a narrower question. It finds a narrower, worse thing.

Each step feels like getting closer to the truth. What is actually happening is that you keep asking questions whose answers can only be alarming, and the fluency (E7·L1) makes every step feel authoritative.

If you notice you are three questions deep and more frightened than when you started, stop. That is the signal. Nothing at the fourth question is going to be reassuring, because you have selected a path where reassurance was never one of the available answers.

The two questions worth asking

If you are going to ask something at two in the morning, ask these.

What would make this urgent? What signs would mean I should seek help immediately rather than waiting?

This is genuinely useful. It gives you a threshold rather than a diagnosis, and thresholds are actionable. It is also the question the answer can be safely wrong about in one direction — an over-cautious threshold sends you to be checked, which is the outcome you want when unsure.

What should I do in the next 24 hours?

Practical, bounded, and it moves you towards a decision rather than a speculation.

Notice both are about action, not identity. That is the difference between useful and harmful in this domain.

Where to go instead

Most countries have a non-emergency health line staffed by people who triage for a living — 111 in the UK, 811 in much of Canada, similar elsewhere. Free, available at two in the morning, and specifically designed for “I do not know if this is serious”.

That is what you actually wanted. An assistant is a poor substitute for it and most people do not know it exists.

For emergencies, call your emergency number. Chest pain, difficulty breathing, sudden weakness or numbness, severe bleeding, a first seizure, thoughts of harming yourself. Do not type it into anything.

Try it now (4 minutes)

Look up your country’s non-emergency health line now, while nothing is wrong, and save it in your phone.

That is a two-minute task that will matter more than anything else in this lesson, on some future night when looking it up is the last thing you will manage.

Check your understanding

1. Why do symptom questions reliably surface frightening possibilities?

Models are designed to be cautious Rare serious conditions are over-represented in medical writing, so they are more available Common conditions are harder to describe

2. You are three questions deep and more frightened than when you started. This means:

You are getting closer to the answer Stop — you have selected a path where reassurance was never an available answer You should ask a more specific question

3. The most useful thing to ask at two in the morning:

What condition matches my symptoms? What would make this urgent, and what should I do in the next 24 hours? How rare is the serious possibility?

Check answers

Recap

A model cannot diagnose because it lacks examination, history, tests and base rates — and frightening conditions are over-represented in what it learned from, so they surface out of proportion to their likelihood. Recognise the spiral and stop when you notice it. Ask what would make this urgent and what to do in the next 24 hours. Save your non-emergency health line now, and for emergencies call your emergency number rather than typing anything.