Traditional Eastern Medicine May Soon Have Its AI Moment

Scientists should keep open minds and, in this age of AI, not shrink from complexity.

Richard Feynman famously stated "What I cannot create, I do not understand." Logically, the inverse statement ("What I create, I must understand") is neither implied nor true. Six-year-old children start building weight-bearing structures with Legos without really understanding physics or mechanical engineering, and the greatest minds in AI research today still do not understand how modern AI works to generate a particular response (AI interpretability). In mathematics, the Gödel Incompleteness Theorem counterintuitively proved that not all true statements can be proved to be true.

Given the above, we should expect, especially in fields as complex as biology and medicine, that there may be many effective drugs whose mechanism of action (MOA) we don't understand. So why is there a prevalent kneejerk reaction to reject traditional Eastern medicine that has endured for thousands of years? Particularly in recent times with the high enthusiasm around AI, it is worth pointing out the many parallels between traditional Eastern medicine and modern AI: both rely on complex pattern recognition rather than clearly explainable mechanisms, both evolved through iterative refinement rather than theoretical deduction, and both face skepticism from those demanding complete transparency in their operations.

Just as AI has proven its worth despite these challenges, traditional medicine has produced remarkable pharmaceutical success stories. Artemisinin, derived from the Chinese herb sweet wormwood, is now a cornerstone of malaria treatment worldwide. Dr. Youyou Tu won the 2015 Nobel Prize in Medicine for her contributions to discovering artemisinin as the active ingredient. Her discovery began with a careful study of a 1,600-year-old text from the Jin Dynasty, "Emergency Prescriptions Kept Up One's Sleeve" by Ge Hong, which described using sweet wormwood for fever. After testing over 2,000 Chinese herb preparations, Tu discovered that cold extraction, rather than the conventional boiling method, was necessary to preserve artemisinin's antimalarial properties – a detail that might never have been discovered without the guidance of traditional knowledge.

Tamiflu, developed from shikimic acid found in star anise (a common ingredient in traditional Chinese formulas), has become a standard antiviral treatment for influenza. These aren't isolated examples – the World Health Organization estimates that 80% of the global population still relies on plant-derived medicines for some aspect of primary healthcare.

Yet in my recent travels across India and China, I see a strong rejection by many people of traditional Eastern medicine, especially among the well-educated. They point to published scientific studies that acupuncture fails to produce results superior to random placebo jabs as evidence that acupuncture must be a sham, conveniently forgetting that a large number of published scientific studies are also retracted for flawed methodology or even malicious fabrication. Even some FDA approved drugs had their approvals revoked after years of public usage due to previously uncharacterized side effects (Vioxx, 2004 and Rezulin, 2000) or lack of real-world efficacy (Makena, 2023). We should not discard entire fields of study due to a few publications that fail to find statistically significant effects, and we should not consider the modern FDA approval process as an infallible judge of drug safety and efficacy.

Institutional Persistence Implies Value

The fact that traditional Eastern medicine has survived for thousands of years inherently suggests that they must deliver some value. Ineffective cargo cult superstitions historically are forgotten in a few generations when they fail to produce observed results above the statistical average outcome, particularly when the practice imposes time or financial costs (e.g. collecting and preparing specific rare plants). Consider the case of the Catholic Church, which survived and thrived in Europe for centuries, becoming at times wealthier and more powerful than monarchies. Whether or not one accepts Catholic theology is beside the point; the institution clearly offered practical benefits that supported its persistence – providing community, emotional support, education, and even primitive social welfare.

Institutions that survive for centuries typically do so because they provide real value, regardless of the accuracy of their explicit theoretical claims. The Catholic Church maintained hospitals, preserved knowledge through monasteries during turbulent times, provided a unifying cultural framework across diverse regions, and offered psychological comfort through rituals and community. These tangible benefits ensured its continuation even as specific theological interpretations evolved over time.

Similarly, traditional medical systems may have persisted because of practical benefits separate from their explicit theoretical claims. The five elements model in Chinese medicine, while not literally accurate by modern standards, may have accidentally captured effective treatments through patterns observed over centuries of practice. When a practitioner attributes a herbal remedy categorized as "Metal" to counter liver maladies typically associated with "Wood," the underlying reality might involve natural products affecting cytokines and cell state changes through molecular mechanisms unknown to traditional practitioners.

Take the example of Ma Huang (Ephedra sinica), classified in traditional Chinese medicine as a warm, acrid herb that disperses cold and alleviates wheezing. For over 2,000 years, it was used to treat asthma and respiratory conditions. Modern science later isolated ephedrine from this plant, which became the basis for bronchodilators used worldwide. The traditional practitioners correctly identified the therapeutic application centuries before the active compound or mechanism was understood – they simply expressed it through a different conceptual framework.

This perspective acknowledges that effective practices can emerge through observation and iteration even when the explanatory framework doesn't match modern understanding. In essence, the right answers can sometimes emerge from frameworks we now consider incomplete or incorrect. The value lies not necessarily in the explanatory model itself, but in the accumulated observations and refined practices that have proven beneficial over generations of clinical experience.

The AI-Eastern Medicine Parallel

The similarities between AI systems and traditional Eastern medicine go beyond superficial analogies. Both domains represent approaches to complex problems that have evolved through empirical refinement rather than theoretical deduction.

Consider these key parallels:

1. Effectiveness without full interpretability: Both AI systems and traditional medical approaches produce meaningful results without practitioners or users fully understanding the underlying mechanisms. Neural networks operate through complex statistical patterns that aren't easily reduced to simple explanations, much like how traditional medical systems might work through mechanisms not yet understood by modern science. The AlphaFold AI system, which predicts protein structures with remarkable accuracy, demonstrates this parallel perfectly. When AlphaFold made the surprising prediction that the bacterial protein GspB contained a previously unknown fold, scientists were initially skeptical. Laboratory verification later confirmed AlphaFold was correct – yet even its creators at DeepMind couldn't fully explain how the system arrived at this novel insight. Similarly, traditional practitioners have long recognized patterns of symptoms that respond to specific treatments without understanding the molecular mechanisms involved.

2. Pattern recognition over reductionism: Both rely heavily on pattern recognition rather than reductionist approaches. AI systems learn to recognize complex patterns in data, while traditional medicine practitioners identify patterns of symptoms and imbalances that may not map cleanly to isolated biological mechanisms.

3. Holistic approach: Both consider systems as interconnected wholes. AI models capture complex interdependencies between inputs rather than isolated variables, similar to how traditional medicine views the body as an interconnected system where imbalances in one area affect others.

4. Empirical development through iteration: Both have been developed through iterative empirical processes – AI through training on massive datasets and optimizing for outcomes, traditional medicine through generations of clinical observation and refinement. In 1578, Li Shizhen published the "Compendium of Materia Medica" (52 volumes) after spending decades testing and refining herbal formulations, documenting 1,892 substances and over 10,000 prescriptions. This massive empirical undertaking mirrors how modern machine learning systems improve through iterative exposure to examples – neither starts with a complete theoretical framework, but both improve through experience.

5. Difficulty with standardized validation: Both face challenges with validation using current scientific methods that expect clear mechanism explanations and standardized outcomes.

AI's recent breakthroughs offer an important lesson for how we might approach traditional medicine. For decades, symbolic AI systems that relied on explicit rules and clearly defined logic dominated the field. These systems excelled at narrow, well-defined tasks (e.g. complex mathematical calculations), but struggled with complex real-world problems involving ambiguity and contextual understanding (e.g. writing a sonnet about toxic love-hate relationships with each line starting with the letter "b"). Neural network-based artificial intelligence changed this paradigm, embracing systems that learn from examples rather than explicit programming, and the results have been transformative.

Similarly, modern medicine has excelled with its reductionist, mechanism-based approach to well-defined diseases, but continues to struggle with complex, multifactorial conditions. For example, in cancer, sometimes there is one or more specific, well-understood DNA mutation that causes the cancer, such as BCR-ABL1 gene fusions for chronic myeloid leukemia. Drugs that target cells with this fusion are effective, and the efficacy can be objectively measured by the fraction of a patient's cells that still have the BCR-ABL1 fusion. This is the equivalent of a snippet of software code in traditional software engineering: precise, falsifiable, and highly effective. Traditional Eastern medicine's pattern-recognition approach, refined over centuries of clinical observation, might offer complementary insights for more complex diseases with many potential underlying causes and no perfect objective quantitative biomarkers – conditions such as pain, aging, and mental health.

The triumph of AI despite its "black box" nature suggests that effectiveness can precede complete understanding. This doesn't mean abandoning the pursuit of understanding, but rather recognizing that useful approaches can emerge from different epistemological frameworks, and that implementation and refinement can sometimes lead understanding rather than following it.

Emerging Biological Evidence

Recent research has begun to identify potential biological mechanisms behind traditional practices, offering intriguing prospects for bridging paradigms. For instance, studies on acupuncture have documented measurable gene expression changes not only at the needle insertion site but also in cells several diameters away, suggesting signal propagation effects that could partially explain the practice's reported benefits.

A particularly compelling example comes from research at the Karolinska Institute in Sweden, where scientists used functional MRI to observe brain activity during acupuncture treatments. They found that needle stimulation at specific traditional points activated precise regions of the brain associated with pain modulation – different points activated different brain areas in patterns consistent with traditional Chinese medicine's mapping of meridian functions.

These findings align with traditional practitioners' claims that effective treatment requires understanding individual variation. A friend who is a senior faculty at Shanghai Jiaotong University explained to me that meridian positions may vary slightly between individuals based on their physical development, health status, and even mental states ("dynamic meridians"). The identification of individual variations in treatment response could help explain why standardized acupuncture protocols in clinical trials often show limited efficacy compared to treatments from experienced practitioners. An expert traditional practitioner doesn't simply apply treatments "by the book" but adjusts techniques based on subtle observations and responses unique to each patient – much like how modern precision medicine seeks to move beyond one-size-fits-all approaches.

This individualized approach in traditional medicine anticipated by millennia what modern medicine now calls "precision medicine" – the recognition that treatments must be tailored to individual variation rather than applied uniformly. While modern precision medicine focuses on genetic and molecular biomarkers, traditional practitioners developed sophisticated observational techniques to identify individual patterns requiring personalized approaches.

These emerging biological findings suggest that traditional medical systems may have captured real physiological effects through different conceptual lenses. They don't necessarily validate the complete explanatory frameworks of traditional medicine, but they do suggest that dismissing these systems entirely would risk losing valuable insights accumulated through centuries of careful clinical observation. This calls for approaching traditional medical systems with scientific curiosity rather than dismissal, searching for valuable insights that might be obscured by differences in terminology and explanatory models.

The Biomarker Challenge

The difficulties facing traditional medicine parallel challenges in modern pain management, where subjective patient reporting remains the primary measure of effectiveness due to the lack of reliable objective biomarkers. This comparison reveals how even cutting-edge Western medicine sometimes struggles with the same validation challenges that confront traditional approaches.

Multiple US clinical pain specialists consistently emphasized to me that "a good pain biomarker is the Holy Grail" of their field. Without such biomarkers, pain assessment depends almost entirely on subjective patient reporting, making it difficult to standardize or quantify across individuals. This challenge is further complicated by phenomena like pain catastrophizing, where patients may report higher levels of pain than actually experienced. On the flip side, fibromyalgia is a condition characterized by widespread pain and fatigue. For decades, many physicians questioned whether it was even a "real" disease, despite affecting millions of patients worldwide.

Pain medicine thus occupies an interesting middle ground – somewhere between modern oncology (where mechanisms and disease progression are well-understood through clear biomarkers) and traditional Eastern medicine (where neither mechanisms nor disease progression have clear biomarkers). Vertex recently received FDA approval for suzetrigine, a drug that inhibits the NaV 1.8 ion channel, so the biological of the mechanism of action is well understood. But surprisingly, its efficacy as measured by patient Net Pain Reduction Score (NPRS) was only minimally better than placebo.

This parallel suggests that the skepticism applied to traditional medicine might sometimes reflect a double standard. The regulatory flexibility shown toward pain treatments with limited measurable benefits indicates that our validation systems can accommodate approaches that don't fit neatly into ideal scientific frameworks. For conditions that lack clear biomarkers – not just pain, but also aging, cognitive function, and mental disorders – traditional medicine's approach of recognizing patterns across multiple subtle indicators might offer valuable complementary insights. Rather than dismissing these approaches for failing to provide the kind of evidence possible only with clear biomarkers, we might recognize that different validation standards are appropriate for different types of health challenges.

Different Approaches for Different Problems

The tension between traditional and modern approaches mirrors another familiar dichotomy in the technological world: programmed expert systems versus trained machine learning systems. This parallel offers illuminating insights into when each medical paradigm might prove most effective.

Programmed expert systems excel at well-defined problems with clear, precise answers – much like modern Western medicine shines when addressing conditions with identifiable pathogens, clear mechanisms, and specific molecular targets. The reductionist approach of isolating variables, establishing causal relationships, and developing targeted interventions works brilliantly for conditions like bacterial infections, certain deficiency diseases, or structural problems requiring surgical intervention.

Take the case of H. pylori infection and peptic ulcers. Once the bacterium was identified as the cause, a precisely targeted antibiotic approach became possible, revolutionizing treatment. This represents Western medicine at its best – identifying a specific pathogen and developing a targeted intervention with clear, measurable outcomes.

Conversely, trained systems like modern AI excel at complex pattern recognition where no single "correct" answer exists – similar to how traditional Eastern medicine may better address conditions characterized by complex, multi-factorial patterns and personalized symptom constellations. The holistic, pattern-recognition approach of traditional systems may better serve conditions without clear biomarkers, where subjective experience plays a major role, or where multiple factors interact in complex ways.

Irritable Bowel Syndrome (IBS) exemplifies this type of challenge. Despite affecting 10-15% of the global population, IBS lacks clear biomarkers and involves complex interactions between the gut, nervous system, microbiome, and psychological factors. Modern medicine struggles with IBS precisely because it doesn't fit neatly into reductionist frameworks. Traditional Chinese medicine, meanwhile, recognizes multiple patterns of IBS-like symptoms and treats them differently based on the individual's overall presentation – an approach that some studies suggest may offer relief for patients who haven't responded to conventional treatments.

This framework suggests a natural complementarity between paradigms, where:

1. Diseases with clear end-states (like infection elimination or tumor eradication) may be better suited for modern Western medicine's reductionist approaches.

2. Conditions with "fuzzy" acceptable states (like pain management, aging, or general wellbeing) may benefit from traditional Eastern medicine's pattern-recognition and systems-based approaches.

Rather than forcing a false choice between paradigms, this perspective suggests a more nuanced approach that applies each where it demonstrates greatest effectiveness. For acute bacterial pneumonia, antibiotics represent the clear first choice. For chronic pain conditions where conventional treatments provide incomplete relief, traditional approaches might offer valuable complementary strategies that improve quality of life.

The Distillation Problem

The process of extracting specific compounds from traditional herbal medicines parallels the challenge of distilling AI models. In both cases, simplification can lead to loss of complexity and effectiveness. When pharmaceutical researchers isolate a single "active ingredient" from a traditional herbal remedy, they may be capturing only one dimension of a multi-faceted therapeutic effect.

This distillation problem has two sides. On one hand, traditional herbal medicines contain numerous compounds that may work synergistically in ways not captured by isolating a single molecule. The complex interactions between multiple compounds in a traditional formula might produce effects greater than the sum of their individual actions, similar to how a complete neural network captures relationships that simplified models miss.

The historical development of Willow bark to aspirin illustrates this complexity. While modern medicine isolated salicylic acid (later synthesized as aspirin) as the active component, the original willow bark contains flavonoids and polyphenols that modulate salicylic acid's effects and may reduce gastrointestinal side effects. Some contemporary studies suggest that natural willow bark extracts may have different side effect profiles than isolated aspirin – indicating that something valuable was lost in the distillation process.

On the other hand, traditional medicines also contain compounds that may be harmful or counterproductive. The complex brewing and preparation methods developed in traditional medicine likely evolved to minimize these harmful elements while maximizing beneficial ones. Modern pharmaceutical processes can potentially improve this filtering, removing truly dangerous components while preserving synergistic interactions.

The traditional Chinese herb Ma Huang (Ephedra) provides a cautionary example. While effective for respiratory conditions when prepared traditionally, isolated ephedrine became problematic when concentrated as a weight loss supplement, leading to cardiovascular side effects and regulatory action. The traditional preparation methods, dosing guidelines, and combination with other herbs may have provided safety guardrails that were stripped away through pharmaceutical distillation.

A useful metaphor is principal component analysis. Some traditional medicines may have a single dominant active component that accounts for most of their effectiveness – like artemisinin from sweet wormwood, which led to effective antimalarial drugs. In these cases, pharmaceutical distillation succeeds brilliantly. But other traditional formulations might work through multiple compounds each contributing small effects that collectively exceed the threshold for clinical significance.

This "multiple principal components" challenge helps explain why some traditional medicines fail to translate into modern pharmaceuticals. If no single compound reaches the typical 20% overall response rate (ORR) threshold for FDA approval, but the complete formula provides 30% ORR through ten compounds each contributing 3%, the traditional approach might be clinically valuable despite failing conventional pharmaceutical development pathways. Given these challenges with pharmaceutical distillation, we need new approaches that preserve complexity while enhancing scientific rigor.

A Path Forward for Traditional Eastern Medicine

One of the greatest advantages of modern AI is its ability to make sense of enormous pools of data that far exceed the brain capacity of any one human. Unlike standard statistical programming approach that allow us to only "search under the lamp-post" with preconceived hypotheses to test, AI allows us to find novel patterns in data in a more hypothesis-free, open-minded fashion.

To really modernize traditional Eastern medicine with AI, we will need data, and lots of it. Ideally, the data will be collected following two core principles:

1. Whole formula studies: Rather than attempting to reduce traditional medicines to single active ingredients, researchers should evaluate complete traditional formulations that are most likely to be actually effective. This approach acknowledges the multicomponent nature of traditional medicines while still subjecting them to scientific validation, similar to how we evaluate complete AI systems rather than isolated layers of artificial neurons.

2. Systems biology measurements: Modern "omics" technologies can capture complex, system-wide responses to interventions. These approaches align well with the holistic perspective of traditional medicine, potentially identifying patterns of biological response that wouldn't be visible when focusing on isolated mechanisms – much like how neural networks capture patterns that simpler systems miss.

Let's work together to create the future of medicine. Traditional Eastern medicine and modern Western medicine both have a place in maximizing human health and life. For the past 100 years, we as a society and civilization have focused on reductionist Western medicine because that was the only thing we could do well, much like how software from 1993 through 2022 was exclusively precise software code following logical principles. With the advent of Big Compute and AI, the pendulum can start to shift in the other direction, not only because it is now possible to do AI-guided analysis of multi-component Eastern medicines and their effects on complex biology of human disease, but also because most of the low-hanging fruits of single-agent molecular medicines have already been picked.

If you are a practitioner or researcher of Traditional Chinese Medicine or Ayurveda, I would welcome a conversation. Please reach out to me personally at dave.zhang@biostate.ai with the Subject line "Traditional Eastern medicine collaboration" to see how we can work together to use AI and Big Data to help the FDA approve disease-curing and life-saving traditional Eastern medicines for patients.

---

By David Zhang and Claude 3.7 Sonnet
April 29, 2025

© 2025 David Yu Zhang. This article is licensed under Creative Commons CC-BY 4.0. Feel free to share and adapt with attribution.