AI and Traditional Chinese Medicine: Can AI Diagnose TCM Patterns?

At first glance, Artificial Intelligence and Traditional Chinese Medicine would seem to have very little in common. One represents the newest technology; the other has been practiced for thousands of years.

But they share something fundamental:

Both depend heavily on pattern recognition.

An AI system looks at large amounts of information, recognizes relationships and patterns, and uses those patterns to arrive at a conclusion.

Traditional Chinese Medicine does something remarkably similar.

A TCM practitioner listens to a patient's symptoms and looks for a pattern. Fatigue, poor digestion, loose stools and a pale complexion might suggest one pattern. Night sweats, dry mouth and insomnia might suggest another. Irritability, headaches and muscular tension might point us in still another direction.

We don't usually treat a symptom by itself. We look at the collection of symptoms and ask:

What pattern do these symptoms create?

That makes AI seem almost tailor-made for Chinese medicine.

Pattern Recognition Versus Tongue and Pulse Diagnosis

Traditional Chinese diagnosis generally includes questioning the patient, observing the tongue and feeling the pulse.

Personally, I have always found pattern recognition from the patient's history and symptoms more reliable than relying heavily on the tongue or pulse.

Tongue diagnosis can be straightforward and extremely helpful. A very red tongue, a pale tongue, a thick greasy coating or an unusually dry tongue can provide useful information that supports what we are already learning from the patient.

Pulse diagnosis is another matter.

Traditional Chinese medicine describes as many as 28 different pulse qualities, including wiry, slippery, thready, rapid, slow, floating, deep and many others. An experienced practitioner can certainly obtain useful information from the pulse.

But distinguishing among all of these pulse qualities can also be more than a little subjective.

I learned this lesson while I was an acupuncture student in San Francisco.

I attended a demonstration of pulse diagnosis in which three experienced practitioners examined volunteer patients. Each practitioner privately felt the patient's pulse and wrote down a diagnosis without knowing what the other practitioners had concluded.

I expected their diagnoses to be reasonably similar.

They weren't.

For patient after patient, we ended up with three entirely different diagnoses.

That demonstration stayed with me. If three trained practitioners could feel exactly the same pulse and arrive at three different conclusions, perhaps we shouldn't pretend that pulse diagnosis is always as precise as we would like it to be.

This is one reason I place considerable emphasis on the overall pattern created by a patient's symptoms and history.

Enter Artificial Intelligence

This is where AI becomes particularly interesting for Chinese medicine.

AI is extraordinarily good at sorting through information and recognizing patterns. Give it a collection of symptoms and it can compare those symptoms with patterns it has learned and suggest possible TCM diagnoses.

I recently tried this with one of my own patients.

Before his consultation, the patient had provided information describing his symptoms. I entered that information into ChatGPT and asked it to consider the case from the perspective of Traditional Chinese Medicine.

The diagnosis it suggested seemed quite reasonable.

In fact, based on the information I had given it, I thought it made very good sense.

Then the patient came into the office.

During our face-to-face interview, he mentioned something that hadn't appeared in the information he had previously provided:

All of his complaints had begun after he became infected with COVID.

That one piece of information changed the entire case.

Suddenly, symptoms that had appeared to form one pattern needed to be reconsidered in the context of a post-infectious illness. The diagnosis I had been considering—and the diagnosis suggested by AI—had to be turned almost upside down.

ChatGPT hadn't necessarily made a foolish diagnosis.

It had made a reasonable diagnosis from incomplete information.

And therein lies the problem.

AI Only Knows What You Tell It

Artificial intelligence can recognize patterns remarkably well.

But it cannot recognize information that it doesn't have.

If the patient forgets something important, misunderstands a question, leaves something off a questionnaire or simply doesn't realize that a particular event might be medically significant, AI cannot take that information into account.

Neither, for that matter, can the practitioner—until somebody asks the right question.

This is one of the great advantages of a real consultation.

During a conversation, I can ask:

When did this start?

What was happening at the time?

What makes it better?

What makes it worse?

Did anything change just before this began?

What else was happening with your health?

And sometimes the most important information appears almost accidentally.

The patient says, “Come to think of it, this all started after…”

And suddenly the pattern changes.

Where AI Can Be Extremely Useful

None of this means that AI has no place in Chinese medicine.

Quite the opposite.

I think it can become an extraordinarily useful tool for practitioners.

AI can help organize a complicated case, suggest patterns that deserve consideration, point out relationships among symptoms and remind a practitioner of possibilities that might otherwise be overlooked.

It may be especially useful as a second opinion for the practitioner.

Here's the information I collected.

Here's the pattern I think I'm seeing.

What other patterns should I consider?

What doesn't fit?

Those are excellent questions to ask AI.

But there is an important difference between using AI as a tool and turning clinical judgment over to AI.

AI Is a Tool, Not the Practitioner

Perhaps the most interesting lesson is that AI and Traditional Chinese Medicine share both a strength and a weakness.

Both depend on recognizing patterns.

And both depend on having the right information from which to recognize those patterns.

AI can help us see connections. It can help organize our thinking. It may even catch things we overlook.

But it cannot look across the desk and notice that a patient hesitated before answering a question.

It cannot know what the patient forgot to mention.

And it cannot ask a follow-up question unless someone realizes that question needs to be asked.

At least for now, that's still the practitioner's job.

So I am enthusiastic about using AI in Chinese medicine.

I use it myself.

But I don't believe we should blindly trust it.

Use AI to help recognize the pattern. Then use experience, questioning, observation and judgment to decide whether that pattern is actually the right one.

After all, a very intelligent answer to incomplete information can still be the wrong answer.