Preventive Care: Turning Complex Medical Knowledge into Daily Habits

2026-09-05 Traditional Chinese Medicine · Learning · AI

If I Had to Name a Belief

I am uncertain whether AGI will arrive, but I am willing to help build it and to observe how it may arrive.

If I had to name something I hold closer to a belief, I would not name a technology. I would name preventive care: 治未病.

I hope that one day, complex medical knowledge will not exist only in classics, with doctors, or inside professional institutions. It may also become part of ordinary life. People should not have to finish an entire body of medical theory before they can notice changes in their bodies, retain useful information, make low-risk daily observations and preparations, or know when to stop handling something themselves and seek professional help.

I have previously used examples such as brown-sugar ginger tea and scallion-ginger water. This is not a judgment about their effect for any particular person, nor a recommendation that readers use them as a prescription. What draws me to them is that this kind of knowledge once entered daily life with a very low barrier to use: people did not need to understand the full theory or open a textbook; in a particular situation, an action would naturally come to mind.

Of course, habits can also pass errors from generation to generation. A practice being widespread does not prove that it works; an easy-to-remember saying may have omitted precisely the most important conditions for its use.

So the question is not how to put more “folk remedies” into everyday life, but this:

How can complex medical knowledge become a daily capability that ordinary people can use over time and correct, without losing its boundaries?

I do not yet have an answer. The several traditional Chinese medicine projects I have worked on over the past few years can be seen as very small pieces of infrastructure around that question.

First, Make a Book Something One Can Return To

My earliest Treatise on Cold Damage mini-program was not intended to be a medical product.

It was first a learning tool for myself. I wanted to structure material that was originally arranged linearly in a book, so that provisions, formulas, and related information would be easier to search, remember, and revisit. It later became a mini-program and was opened to the public. From what I observed of its use, several hundred people used it, but it was mainly a low-frequency, on-demand formula lookup tool: people opened it when they needed it and left when they were done.

By the usual standards of internet products, those numbers are not impressive. People did not open it frequently, and it did not show strong retention. I later spent more of my time studying traditional Chinese medicine and exploring AI applications rather than continuing to promote it.

But I later came to think that this pattern of use may not be a failure.

A learning and reference tool need not occupy a user’s attention. Being able to find it when needed, leave once the task is done, and return willingly when needed again is itself a relationship more appropriate to a tool.

What this mini-program accomplished was also limited. Making knowledge easier to access does not establish the medical efficacy of every item in it; structuring provisions and formulas does not amount to clinical understanding; and people looking something up does not mean their health improved as a result.

It showed me only the first layer of the problem: if complex knowledge remains hard to search and hard to review, it is difficult for it to enter long-term life. Structuring is not a medical conclusion, but it may be foundational work before a habit can form.

From Static Knowledge to Processes in Time

The Treatise on Cold Damage tool mainly dealt with static material. Actual diagnosis and treatment, however, are not a symptom-to-formula lookup table.

While studying my teacher’s cloud-based clinical follow-up videos, I gradually shifted my attention elsewhere: how the same patient moves from an initial consultation to a follow-up, what feedback they give after taking medicine, what the teacher changes in response, and how the next round of feedback changes the judgment again.

Diagnosis → Medication → Feedback → Re-diagnosis → Further medication → Further feedback

This is not a medical pattern I have discovered. It is only an object I hope to observe over the long term.

Traditional Chinese medicine has many schools. When different practitioners treat similar problems, their explanatory frameworks, terminology, and methods may differ greatly, and may sometimes even conflict. I have an unproven suspicion: beneath these apparently different explanations and practices, might there be some common structure in the course of continuous feedback?

It is too early to make comparisons across practitioners. What I am doing now is first learning longitudinally from one teacher and that teacher’s patients, organizing visits, clinical facts, original words, and changes over time that are scattered across long videos. Only if there are sufficient materials, authorization, and professional collaboration in the future could it become possible to compare the processes of different practitioners.

The cloud follow-up workbench is therefore not an automated diagnostic system, nor an already-launched clinical study. It is a local learning tool and source library: it helps me find patient boundaries, links between visits, original evidence, and my own study notes in hours of video.

This step is much slower than it sounds. In real project records, a single video can contain more than eighty cases; entering basic information for one case can take more than five minutes, and studying a video in full often takes two or three weeks. Transcription can be wrong, patient boundaries can be wrong, and the teacher’s explanations, my understanding, outside material, and unresolved questions cannot be mixed together.

The arrival of AI did not suddenly give me medical judgment. AI can reduce mechanical work in transcription, segmentation, retrieval, and organization, but it cannot automatically elevate a candidate organization into a clinical fact, much less prove why a treatment works.

Habit Is Not an Unbroken Check-In Streak

When people talk about turning knowledge into a habit, reminders, points, streaks, and daily tasks easily come to mind.

But I do not want people to open an app merely to keep a number going.

In designing the cloud follow-up learning tool, I care more about other questions: can someone spend only ten to fifteen minutes a day first working through the old material most in need of review? After missing a few days, can they return naturally rather than face a wall of overdue tasks? Can they first restate something in their own words, then look at the original text and outside explanations? Can uncertainty be preserved rather than turned into a conclusion too early just to complete a task?

This kind of habit is not high-frequency use itself. It is a recoverable cycle:

Encounter material → Active recall → Return to evidence → Revise understanding → Preserve questions → Encounter it again

If it were someday to enter ordinary health-related life, the actions in this cycle would not necessarily be only “what to eat” or “what to drink.” They could include recording changes, recognizing warning signs, preparing information for a consultation, understanding a doctor’s instructions, observing feedback after an intervention, and knowing when one should not make a judgment alone.

Complexity has not disappeared as a result. It has only been broken into small actions that people may be able to carry out in daily life, with each step preserving a way back to evidence and correction.

That is the opposite of compressing medicine into a few universal rules.

Compression Must Preserve a Route Back to Complexity

I have a strong tendency to compress. I instinctively look for common structures behind different phenomena, hoping to use a smaller framework to hold a large body of material.

In medical questions, this tendency is both valuable and dangerous.

Without compression, people are overwhelmed by endless detail; knowledge is hard to spread and even harder to turn into action. But when compression goes too far, individual differences, conditions of use, dosage, timing, interactions, and warning signs can all be erased by one catchy phrase.

So any piece of medical knowledge that enters daily life should retain at least several routes back:

Ordinary people cannot reconstruct an entire medical system before every action, and experts cannot examine every health judgment made across society each day. What is really needed is a mechanism in which low-risk actions can happen simply, high-risk problems can be escalated promptly, and errors can be discovered and corrected.

This is still only a direction, not a system I have already built.

AI Can Lower the Threshold, Not Bear the Consequences for Knowledge

AI makes this direction seem less remote for the first time.

It can turn long videos into searchable text, help organize structural candidates awaiting verification, connect a patient’s multiple visits, generate review cards from old material, and make potential relationships between different expressions easier to see. The project tried automatically identifying patient boundaries, but it did not reach a usable level; case start and end points still need to be confirmed by a person returning to the video and audio.

But these capabilities can also magnify errors.

A transcription error can change the name or dose of a medicine; a segmentation error can join two patients together; a fluent model-generated summary can erase the hesitation and conditions in the original video; and clinical material that has not been de-identified involves the privacy and authorization of real patients.

The workflow I use now is therefore closer to this:

Source material
  ↓
AI produces transcription and structural candidates
  ↓
Human confirms case start and end points, and returns to the video to correct transcription and structural candidates
  ↓
Separate facts, the teacher's explanations, personal understanding, and questions
  ↓
Produce searchable, reviewable, traceable learning material

The output of this workflow is still learning material, not diagnostic or treatment conclusions.

At present, I do not send this kind of material to cloud models. Original medical records, media, transcripts, and personal notes remain local. Even if cloud capabilities are considered in the future, it would first be necessary to confirm appropriate authorization and independently validate outbound de-identification, auditing, vendors, and deletion boundaries. Only then could the smallest necessary de-identified copy possibly be sent; model output would still be only a candidate derivative. Until these boundaries are in place, expanding automation may not be expanding capability at all—it may only be expanding exposure.

AI can make complex knowledge easier to process, but it cannot automatically decide which forms of compression are safe, which errors are tolerable, or who should bear the consequences.

What I Really Want to Build Is Not a Product People Cannot Leave

Measured as commercial products, the traditional Chinese medicine tools I have made are all small.

One mini-program that, by my observation, was used by several hundred people and mainly served low-frequency, on-demand lookup; one cloud follow-up workbench still in development, where even the correctness of transcription and patient segmentation needs further correction; and one idea about common structures across schools that has not yet reached a stage that supports conclusions.

These boundaries need to be stated clearly.

But they have also gradually clarified that what I may want to build is not an app that keeps people longer each day, but a quiet layer of infrastructure for everyday life.

It does not ask people to become fanatical about medicine, nor does it turn everyone into a doctor. It simply lets knowledge appear when it is needed, gives daily observations somewhere to accumulate, lets personal understanding return to evidence, and makes professional help easier to bring in at the right time.

Ideally, people would not even need to notice that there is a large knowledge system behind it. Some habits are already woven into families and society: the action is simple, but the simplicity is not arbitrary. Behind it is a complex structure that has been tested over time, retains boundaries, and can be updated.

That is a different measure of value from how many times users open an app each day.

Technology Is Only a Middle Layer

I am interested in AI, and I am willing to study Auto Research, formal verification, and human-machine collaboration. But these are not my final purpose.

Models will change, ways of programming will change, and capabilities that seem astonishing today may soon become infrastructure. For me, binding the direction of a life to a particular model, architecture, or forecast about AGI is still too short-sighted.

Preventive care—治未病—offers a longer scale.

When I use this term, I am not declaring that any medical system has solved disease, nor am I packaging unverified experience as traditional wisdom. It points to a plain hope: that people can understand their changes earlier, that reliable knowledge can enter daily life more easily, that small problems may be noticed before they become larger, and that ordinary people can be better prepared rather than more deluded when facing a complex medical system.

What I am doing now is only structuring one book, organizing a set of videos, designing some review and feedback processes, and raising a research question that remains to be tested.

They are still far from that hope.

But if AI can truly expand a person’s ability to enter unfamiliar fields, handle complex material, and build tools, what I most hope it will serve in the end is not proving how powerful AI is, but helping complex knowledge cross the distance between professional life and everyday life.

I cannot be certain about the endpoint of a technology that has not yet appeared.

I am willing to give my time to a more concrete and more difficult direction: let knowledge become habit, let habit retain evidence, and let ordinary people have more possibilities for understanding and action before illness truly arrives.