From the Confusion Over 'Wind' to the True Mission of AI for Science

I. The Starting Point of Confusion: What Exactly is "Wind" in TCM?

Every researcher who has seriously contemplated fundamental questions in medicine will sooner or later encounter a common confusion: What exactly is the "Wind" (风) spoken of in Traditional Chinese Medicine?

If "Wind" refers simply to the naturally flowing air, how can it be considered a cause of disease? The movement of air itself is neither pathogenic nor therapeutic. Listing "Wind" as the first of the "Six Excesses" (Wind, Cold, Summer Heat, Dampness, Dryness, Fire) makes no sense in terms of physics.

If "Wind" has no physical entity that can be found under a microscope, does this mean TCM theory is "unscientific"? This question has troubled both the TCM community and modern medicine for at least a century. The debate between the two sides often falls into an either-or impasse: one side insists that "Wind" is a real pathogenic factor, while the other considers it merely a naive imagination of ancient people about pathogens.

However, when we temporarily set aside the question of "whether Wind is an entity" and instead ask another question — "How is the concept of 'Wind' actually used in TCM clinical practice?" — the answer becomes clear.

TCM theory's description of wind pathogen focuses on its "pathogenic characteristics": Wind is active in nature, moves quickly and changes rapidly; Wind is a Yang pathogen, characterized by opening and dispersing. None of these descriptions characterize the physical or chemical properties of "Wind" itself; rather, they characterize the response patterns exhibited by the human body when facing specific environmental perturbations: symptoms that come and go rapidly, locations that are not fixed, and a tendency to cause the body surface to open and disperse — these are characteristics of human reactivity, not intrinsic properties of wind itself.

TCM's understanding of etiology, guided by holism, sets aside the complex pathogens that can only be seen with microscopic tools, and instead focuses on grasping the specific reactive states of the organism caused by pathogens. Through abstraction of the organism's specific reactive states, TCM recognizes disease etiology and its pathogenic nature and characteristics — the so-called "differentiating syndromes to identify causes" (审证求因) or "differentiating patterns to identify causes" (辨证求因).

> "Wind" is not an object of physics, but a response pattern of the human body.

This cognitive shift is crucial. It means that TCM's "Wind" does not name a "thing," but names a "state." It is not an etiology in the etiological sense, but a characterization in the system response sense.

And once we start asking questions from "Wind," we further discover: the "Wind, Cold, Summer Heat, Dampness, Dryness, Fire" of the Six Excesses in TCM can essentially all be understood as classifications of typical response patterns of the human body when facing different environmental perturbations; TCM's "Qi, Blood, Yin, Yang, Deficiency, Excess, Cold, Heat" even more directly describe system states rather than naming objects. More fundamentally, the core of the entire TCM theoretical system is not an object description language, but a response description language.

This is not merely a discovery about TCM. When we examine the terminology system of modern medicine with the same lens, we find a structurally significant fact that has long been overlooked: although modern medicine is known for object description, its terminology system objectively contains a class of response description concepts. Irritable bowel syndrome, chronic fatigue syndrome, and fibromyalgia — these functional diagnoses lack clear organic lesions or single molecular markers and are essentially cluster descriptions of abnormal system response patterns. The "compensated" and "decompensated" phases of cardiac insufficiency describe not the damage of a specific part, but the capacity state of the system maintaining homeostasis. Systemic inflammatory response syndrome (SIRS) and multiple organ dysfunction syndrome (MODS) are names for the response patterns of the entire system to severe insults.

> The terminology system of medical knowledge has never been singular.

At its deep structure, it has always contained two paradigms — object description and response mapping. The former paradigm excels at slicing the world into objects, asking what the pathogen is, where the lesion is, which gene mutated; the latter paradigm excels at organizing the world into states, asking what mode the system is currently in, how it responds when facing perturbations. The historical evolution of modern medicine has continuously strengthened the former, while the latter has long lacked a formalized language and has thus often been misunderstood as vague, empirical, or even excluded from "hard knowledge" altogether.

It was precisely this discovery that catalyzed our original motivation for proposing the "Response Mapping Hypothesis."

II. Deducing from First Principles: Why Must We Introduce the "Response" Perspective?

To judge whether the Response Mapping Hypothesis is valid, we cannot rely solely on intuition or experience. We need to start from first principles and use deductive reasoning to establish its logical necessity.

Step One: Establishing the Starting Point.

What is life? If we attempt to give the most basic answer with the fewest presuppositions, it is this: Life is the result of the interaction between life and its environment. A cell, an organ, a complete human being — all must continuously exchange matter, energy, and information with the outside world. Without an environment, there is no life; without continuous response to environmental changes, life cannot maintain its ordered state. Therefore, the essence of life can be understood as an open system that continuously interacts with its environment to maintain its own ordered state.

Step Two: Deriving the Nature of Disease.

If the essence of life is "continuous interaction with the environment," then what is disease? The nature of disease cannot be understood merely as damage to a part. Part damage — such as trauma, infection, single-gene mutations — is of course an important form of disease, but not the only one. The more universal situation is: there are problems in the system's interaction with the environment — the system's mode of response to environmental perturbations has deviated from the healthy trajectory, manifesting as declining adaptive capabilities, narrowing of regulatory range, and shifting of homeostasis.

Hypertension, type 2 diabetes, depression, and aging all fall into this category. They have no single "lesion" and no unique "cause." They are trajectory deviations of a system running under multiple environmental pressures over extended periods.

Step Three: Identifying the Cognitive Blind Spot.

Following this deduction, we can discover a structural blind spot in modern medicine.

Modern medicine has become highly developed in describing the "environment" — pathogen databases, nutrition science, toxicology, and environmental epidemiology all provide detailed dictionaries of environmental variables. In describing the "human system itself," modern medicine has also achieved enormous success — genomics, proteomics, anatomy, and pathology have established a refined object description system.

But in describing "what response the environment has produced in the human body" — this intermediate link — we lack a systematic, computable language.

Network medicine has already mapped disease genes onto protein-protein interaction (PPI) network topology, and Barabási et al. have pointed out that complex diseases are essentially perturbations of network modules rather than abnormalities of single genes. But this is still a description of "fault location." What is truly missing is a description of "fault mode" — not "which node is broken," but "what has changed in this network's response mode."

For example, facing chronic psychological stress, the key question is not whether the NF-κB node exists (of course it does), but what has changed in the response mode of the NF-κB network — whether its activation threshold has decreased, whether its negative feedback loops have failed, whether its coupling strength with other signaling pathways has changed.

Step Four: Arguing the Necessity of Response Mapping.

If we want to describe "response modes," we must map "responses" into an appropriate description space.

The impact of the environment on the human body is comprehensive — from gene expression, protein modifications, and metabolites to cell behavior, organ function, and overall manifestations — information is distributed across multiple scales and multiple dimensions. Directly processing this high-dimensional, sparse, and noisy raw data is infeasible. Therefore, "responses" must be mapped into a state space of appropriate dimensionality with biological causal constraints. This mapping process itself is one of compression, refinement, and meaning-assignment of the original information.

Step Five: Validating the Deduction with Examples.

Theoretical deduction needs to be tested with concrete examples. Let us return to the confusion about "Wind."

In 2025, Professor Yang Zifeng's team, in collaboration with Southeast University, translated the TCM concept of "Wind-Cold" into quantifiable environmental variables — combinations of temperature and airflow parameters ("artificial Wind-Cold") — and systematically explored the effects of Wind-Cold exposure on respiratory immune defense through animal models. The study found that under sustained cold airflow exposure conditions, the body's "Wei Qi" defense system exhibited a three-stage pathological evolution from "suppressed defensive Yang → damaged defensive Qi → simultaneous blaze of Qi and Blood" (Yang et al., 2025).

The significance of this study lies in: it proves that "Wind-Cold" is indeed not naming an entity, but naming the staged response patterns of the human body facing coupled perturbations of temperature and airflow. Scholars have further pointed out that "Wind pathogen" has significant associations with immune molecules such as tumor necrosis factor-α (TNF-α) and interleukin-6 (IL-6) (Liao Yan et al., 2022), providing preliminary evidence for mapping the response pattern of "Wind" to molecular networks.

Let us consider another entirely different example: fever. Fever is not a "damage," but a response pattern. Facing a bacterial infection, the hypothalamic temperature set-point elevates, peripheral blood vessels constrict to reduce heat dissipation, muscles shiver to increase heat production, and immune cells release inflammatory cytokines. The core judgment a physician makes when facing fever is not "whether to reduce the fever," but "is this fever response pattern appropriate (a normal defense against infection), or is it out of control (such as the collapse of thermoregulation in septic shock)?"

Starting from the confusion about "Wind" and proceeding through first-principles deduction, we arrive at the same conclusion: medicine needs a dedicated terminology system to describe "responses," not merely to describe "objects."

III. The Limitations of Human Concepts: The Lesson of mTOR's Naming

Having established the necessity of response description, we need to pursue a deeper question: Even if we recognize the importance of response description, can we complete the construction of this terminology system through human experiential induction alone?

The naming history of the mTOR gene provides a cautionary answer.

TOR originally stood for "Target of Rapamycin," as it was discovered in yeast as the target of the immunosuppressant rapamycin. Later, a homolog of this gene was found in mammals, and it was named mTOR (mammalian Target of Rapamycin).

But as research deepened, it became clear that this protein's functions extended far beyond "being inhibited by rapamycin" — it is a core regulatory hub for cell growth, metabolism, autophagy, and aging, involved in dozens of disease processes including cancer, diabetes, and neurodegeneration. Calling it "Target of Rapamycin" is like calling Einstein "the patent office clerk" — while this was the clue that led to his discovery, it falls far short of capturing his essence.

In 2009, the HUGO Gene Nomenclature Committee attempted to correct this issue, changing mTOR's official name to "mechanistic Target of Rapamycin," but the initialism remained mTOR. In 2013, Michael Hall, one of the founding figures of the TOR field, specifically wrote about this naming issue, pointing out that the mixed use of old and new names was causing unnecessary confusion (Hall, 2013).

This is the typical dilemma of human concepts: concepts are empirically named at the time of discovery; as understanding deepens, the old name increasingly fails to reflect the object's essence, but it has already been written into textbooks, databases, and diagnostic standards — it cannot be changed.

> Re-examining mTOR from the perspective of response mapping yields a deeper insight: mTOR is essentially an extreme case of response mapping.

mTOR was discovered and named because it was the "target" of rapamycin — that is, the system's response to this chemical substance was anchored to this single gene. What researchers focused on was not the full range of functions of the mTOR gene itself, but its role as the bearer of the "rapamycin response." However, this response was extremely compressed onto a single gene node, rather than being described as a change in network state.

If the concept of "target gene" is expanded from a single node to a gene network module, describing how rapamycin selectively remodels the topological structure and information transmission efficiency of this module — this is response mapping. The naming history of mTOR tells us that even in the purest object description paradigm (molecular targets), the logic of response has never truly been absent — it has merely been over-compressed.

The lesson of mTOR also reveals two systematic limitations of human experiential induction in defining concepts. First, naming path dependency — once a concept is named, it becomes locked in, even if it is later proven inaccurate. Second, the boundaries of inductive capacity — facing the massive data of the multi-omics era, the human brain cannot identify all meaningful pattern combinations through experience. We can recognize macro-level patterns like "fever," but we cannot identify the countless possible "response mode variants" among thousands of nodes in the PPI network.

This is precisely the fundamental reason why AI for Science needs to intervene.

IV. The SEMO Algorithm: An Engineering Example of AI-Defined Concepts

If the lesson of mTOR demonstrates the limitations of human concept generation, then the SEMO Algorithm demonstrates an engineering pathway for AI's intervention in concept generation.

The core logic of SEMO (Selective Remodeling of Protein Networks by Chemicals) is: the impact of a chemical substance on a biological system is not a single-point action, but a selective remodeling of specific sub-networks of the protein-protein interaction (PPI) network. The related methods have been disclosed in preprints and patents (Xiong, 2023; Xiong Jianghui et al., 2023).

The key operation is: SEMO does not merely compute the global target set of a chemical substance, but intersects the global targets with different PPI sub-networks. The significance of this "intersection" operation is that it resolves the issue of concept granularity.

Take curcumin as an example. Curcumin is known to have over a hundred targets, involving multiple pathways such as NF-κB, COX-2, STAT3, and p53. If only the global target set is used to describe "curcumin's response characteristics," the resulting concept is overly broad — anti-inflammatory, antioxidant, pro-apoptotic, autophagy-regulating, and all other possible effects are mixed together, making it impossible to distinguish what curcumin is actually doing in a specific context.

But through SEMO's "intersection" operation, we can generate a series of fine-grained local concepts:

- SEMO-Curcumin-NFκB Core Module: The intersection of curcumin targets with the NF-κB core signal transduction sub-network. This local concept describes "curcumin's capacity for direct intervention in the NF-κB canonical pathway."

- SEMO-Curcumin-Inflammasome Module: The intersection of curcumin targets with the NLRP3 inflammasome-related sub-network. This local concept describes "curcumin's capacity to regulate inflammasome activation."

- SEMO-Curcumin-Oxidative Stress Module: The intersection of curcumin targets with the NRF2-ARE antioxidant pathway. This local concept describes "curcumin's capacity to enhance endogenous antioxidant defense."

These three local concepts, although all derived from the same compound, correspond to different biological sub-networks and can therefore be independently measured, verified, and matched to different individual states. If Patient A's primary problem is NF-κB overactivation, then a high score for SEMO-Curcumin-NFκB Core Module may indicate that curcumin will be effective for them; if Patient B's problem is redox imbalance, then a high score for SEMO-Curcumin-Oxidative Stress Module is the key metric.

The essence of the SEMO Algorithm is a form of "concept refinement engineering." If human-defined "inflammation" is a 1:1,000,000 map, SEMO generates 1:100,000 or even 1:10,000 high-precision local maps. These local maps do not negate the global map, but have higher practical value in specific application scenarios.

More importantly, the SEMO Algorithm points to a methodology for AI-generated new concepts: not creating from nothing, but performing "intersection" and "combination" at the boundaries of existing concepts, thereby generating concept variants with finer granularity and stronger contextualization. Human experts can define "curcumin" and "inflammation," but cannot manually enumerate the intersection of curcumin with every biological sub-network and evaluate the relative importance of these intersections across different disease contexts — this requires AI's computational capability.

V. The Significance of "Hallmarks Engineering": From Generating Quantifiable Hallmarks to Concept Generation

If SEMO demonstrates how AI can generate finer concept variants within existing conceptual frameworks, then "Hallmarks Engineering" takes a step toward AI autonomously discovering new concepts.

In 2024, Xiong Jianghui's team published a preprint on bioRxiv titled "AI-Generated Hallmarks of Aging and Cancer," proposing the computational method of "Hallmarks Engineering" (Xiong, 2024).

The most striking finding of this work is: hallmark-level features generated through causal emergence analysis have up to 9.7 orders of magnitude stronger disease association than individual genes. This means that characterizing diseases at the system-level "hallmark" dimension has overwhelming advantages over characterization at the individual gene micro-level. This is because macro-level "hallmarks" filter out the vast molecular stochastic noise at the lower level, directly capturing the true dynamical trajectories of system homeostatic shifts. Just as in the stock market, looking at macro-level market indices often reveals more about economic reality than staring at the millisecond-level fluctuations of a single stock.

In terms of specific research methodology, "Hallmarks Engineering" adopted a top-down modeling strategy — utilizing genome-wide DNA methylation data to quantitatively reconstruct existing aging and cancer hallmark features (such as genomic instability, telomere attrition, and epigenetic alterations), and evaluating their predictive performance across ten age-related disease datasets. The study found that hallmark-level features possess stronger signal intensity in disease prediction, and hallmark-based models achieved comparable predictive performance with fewer predictor variables compared to conventional pathway-based models.

It should be carefully noted that the current focus of this research is on generating quantifiable versions of existing aging and cancer hallmarks, and comparing them with known biological pathways and ontologies. It is a "proof of concept" — demonstrating that AI can generate feature representations with strong predictive power at the system level, and establishing quantitative mappings between TCM and modern medicine concepts. It does not yet claim to have produced entirely new hallmarks never before named by humans. But from a methodological perspective, it has demonstrated a clear pathway: as data scale grows and algorithmic capabilities advance, AI is entirely capable of autonomously discovering response patterns in massive data that humans have not yet named but that possess strong causal explanatory power, and solidifying them into new terms, new concepts, and new ontology nodes.

The "macro causal emergence" theory proposed by Hoel et al. provides deep theoretical foundations for this direction: in complex systems, macro-scale features — such as system-level response patterns and higher-order state labels — often possess stronger causal explanatory power than micro-scale isolated points (Hoel et al., 2013). This forms a theoretical and empirical resonance with the findings in "Hallmarks Engineering" research.

This theoretical resonance also leads to a deeper inference: since the advantage of micro-level features lies in granularity and the advantage of macro-level features lies in causal explanatory power, there must exist a vast "middle ground" between micro and macro — and this middle ground is precisely where the Response Mapping Hypothesis is most applicable. We can systematically discover a series of new concepts in this middle ground: more macro than individual genes (and therefore more explanatory), yet more refined than classical Hallmarks (and therefore more actionable). The ultimate form of these new concepts may well be new Ontology Terms — neither patching up old concepts nor tearing everything down to start over, but finding those "conceptual habitats" with the most biological significance and clinical value in the vast space between micro and macro.

This allows us to classify AI's capacity for generating concepts into three levels:

Level One: Concept Mapping. Mapping natural language descriptions to existing standardized terms. This is the current mainstream application of medical AI.

Level Two: Concept Generation. Automatically generating new leaf nodes within existing ontology frameworks. For example, automatically naming new gene functions within the Gene Ontology framework.

Level Three: Conceptual Framework Reconstruction. Discovering blind spots in existing ontology systems, establishing new conceptual dimensions, and changing the fundamental way we classify diseases and health states. This is the true target of the Response Mapping Hypothesis — not merely adding new nodes within an old framework, but discovering an entirely new conceptual dimension (response description) and systematically constructing it into a computable Ontology.

In the broader AI for Science field, this direction has already seen exciting progress. The Spacer system proposed in 2025, through a "deliberate decontextualization" method, automatically extracts keyword combinations with high innovation potential from keyword maps constructed from 180,000 biology papers, and converts them into original scientific concepts through a validation pipeline (Lee et al., 2025). In the same year, the AI-Newton system demonstrated another pathway — autonomously deriving physical laws from raw experimental data, with one of its core innovations being "proposing interpretable physical concepts to construct laws" (Fang et al., 2025). These works indicate that AI autonomously discovering new concepts from data has moved from theoretical envisioning to engineering implementation.

VI. Proposing the Response Mapping Hypothesis: Naming a Cognitive Framework

At this point, we have completed the full deduction from the starting point of confusion to the construction of a framework:

1. Starting from the confusion about "Wind," we discovered that the core terms of TCM are essentially response description language, not object description language.

2. Deducing from first principles, we demonstrated that disease is essentially a deviation in system response patterns, and that modern medicine has a systematic blind spot in the response description dimension.

3. From the lesson of mTOR's naming, we observed the structural limitations of human experiential induction in concept generation — and mTOR itself is an extreme case of response mapping being over-compressed to a single gene node.

4. From the SEMO Algorithm, we saw that AI can generate finer-grained concept variants at the boundaries of existing concepts through "intersection" operations.

5. From "Hallmarks Engineering" and frontier research on AI concept generation, we observed that the technical pathway for AI to autonomously discover new concepts is maturing.

All these deductions converge into a unified cognitive framework. We name it the "Response Mapping Hypothesis."

The reason for using "hypothesis" rather than "theory" is based on scientific philosophical caution. Response mapping is currently in the stage of logical construction from first principles and preliminary data validation. Calling it a "hypothesis" clarifies that the current work is posing a new scientific question — namely, "Does there exist a universal response mapping space?" — rather than declaring that the final answer has been found.

The core claim of the Response Mapping Hypothesis is: medicine needs to establish a dedicated, computable terminology system to describe "how systems respond to environmental perturbations," not merely to describe "what damage the system has sustained." The construction of this terminology system cannot rely solely on human experiential induction, but requires leveraging AI's computational capabilities to autonomously discover and define new response pattern concepts in multi-omics data.

VII. The True Mission of AI for Science: Re-inventing the Language of Medicine

If the Response Mapping Hypothesis holds, then its implications for AI for Science are profound.

Today, the vast majority of AI applications in biomedicine operate within existing conceptual systems. AI predicts protein structures using concept systems already defined by humans; AI assists in diagnosis by learning disease categories labeled by human physicians; AI screens drug targets using existing knowledge graphs.

This is certainly valuable. But it is essentially performing computation within existing conceptual systems, not reconstructing the conceptual systems themselves.

For complex problems like aging, chronic diseases, and sub-health, the existing conceptual system itself has structural deficiencies — it is too heavy on object description and too light on response description. Without fixing this structural deficiency, no matter how powerful AI becomes, it merely runs faster on old maps without being able to draw new ones.

Therefore, the true mission of AI for Science is not only to use AI for optimization within old frameworks, but should also include: using AI to discover new concepts, establish new frameworks, and reconstruct how we understand health and disease.

The Response Mapping Hypothesis is a specific entry point in this direction. It identifies the systematic absence of the "response description" dimension in the current medical conceptual system, and proposes a computable framework that enables AI to autonomously discover and define new response pattern terminology in state spaces such as PPI networks, metabolic fluxes, and epigenetic markers. This is not merely "using AI to analyze data," but using AI to re-invent the language of medicine itself.

VIII. Prospects: AI-Driven "Feature Engineering" and "Concept Generation"

If the Response Mapping Hypothesis holds, then a natural question follows: What specifically can AI do? What kind of new concepts will it produce? The following discussion unfolds in a Q&A format.

Q: Will AI propose entirely new Aging Hallmarks? A: Yes. This is not about replacing the consensus process of human experts, but about opening up an entirely new, data-driven pathway.

Currently, AI can already identify hidden patterns related to aging from massive datasets. For example, AI models have predicted that activation of specific genes causes cells to exhibit aging markers such as arrhythmia and inflammation-related gene activity. This proves AI has the capacity to "see" new features from data.

In the future, AI could autonomously define new system-level features representing aging that humans have not yet named, through unsupervised learning on large-scale aging cohorts (combining genomic, epigenomic, proteomic, and even wearable device data). For example, AI might discover that "mitochondria-endoplasmic reticulum communication efficacy decline" is an independent aging hallmark — involving reduced physical contact sites between the two organelles, decreased calcium signal transmission efficiency, and disordered lipid exchange. This hallmark cannot be simply classified into any existing Hallmark (it is neither purely "mitochondrial dysfunction" nor purely "loss of proteostasis"), but it has unique causal explanatory power for the aging trajectories of specific tissues (such as skeletal muscle and neurons). Another possible example is "loss of tissue microenvironment homeostatic resilience" — AI might discover from single-cell sequencing data that with aging, the robustness of paracrine signaling networks between different cell types in tissues declines, manifesting as flattened signal gradients and sluggish feedback loops. This is a highly system-level feature that human experts would have difficulty identifying through experiential induction, but AI can automatically recognize it by analyzing topological changes in intercellular communication networks.

Q: Can AI "upgrade" and "merge similar items" of existing Aging Hallmarks? A: Yes, and this is even AI's strongest current capability.

First, upgrades (concept refinement). Existing Hallmarks are highly condensed macro-level descriptions. AI can "dimensionally reduce" a macro-level hallmark to a more precise molecular level. Take the hallmark of "altered intercellular communication" — currently a relatively general description. Using single-cell sequencing data, AI can refine it into "functional imbalance of specific signaling axes between specific immune cell subsets and stromal cells." For example, AI might discover that the CXCL12-CXCR4 signaling axis between CD8+ T cells and fibroblasts in aging tissues has undergone specific alterations, and this change is highly correlated with fibrosis degree. This upgrades a macro hallmark into a targetable, cell-type-specific local response pattern.

Now, merging (concept association). Existing Hallmarks are not entirely independent; complex coupling relationships exist among them. AI can discover synergistic change patterns between different Hallmarks. For example, through multi-omics data analysis, AI might discover that "genomic instability" and "telomere attrition" are strongly correlated under specific branches of the DNA damage response pathway — they are essentially two downstream manifestations of the same upstream mechanism (nuclear lamina structural disruption). Based on this discovery, AI could group them under a more superordinate concept, such as "nuclear structural homeostasis disruption." This "merging of similar items" is not simple semantic induction, but data-driven causal emergence analysis that discovers feature combinations with unified explanatory power at higher levels.

Q: Will AI produce Hallmarks that fuse TCM and Western medicine? A: Not only is this possible, but it is the application scenario with the most potential for AI-driven "response mapping."

The greatest obstacle to TCM-Western medicine integration is the ontological difference between the two conceptual systems — one emphasizing object description, the other emphasizing response description. AI can establish a quantifiable mapping bridge between the two.

Specifically, AI can establish associations between TCM's "syndromes" (证候) and Western medicine's molecular networks. By mapping macro-level syndrome descriptions onto specific gene expression or metabolic pathway changes, AI can create fusion concepts like "Qi Deficiency — Energy Metabolism Network Low-Dynamic Mode." This is akin to AI precisely delineating the "sphere of influence" (Attractor Basin) of each traditional TCM syndrome in three-dimensional molecular network space. This fusion concept preserves the response description connotation of TCM's "Qi Deficiency" (insufficient system dynamism, declining recovery capacity) while endowing it with a computable molecular definition (declining mitochondrial oxidative phosphorylation efficiency, altered AMPK signaling pathway activity, reduced specific metabolite fluxes).

In practice, research teams are already conducting similar work. Professor Li Shao's team at Tsinghua University is using multimodal AI to integrate clinical phenotypes, TCM syndromes, and multi-omics data for identifying key biomarkers in the earliest stages of cancer. Professor Cheng Haibo's team at Nanjing University of Chinese Medicine has also proposed that one of the key directions for future TCM-Western medicine integration is using AI to transform traditional TCM diagnosis and treatment into digital indicators.

One can envision that future AI systems may autonomously discover a fusion Hallmark such as: "Liver Stagnation — Chronic Inflammation Coupled State." It describes the response pattern of specific inflammatory networks (chronic low-grade NF-κB activation, decreased glucocorticoid receptor sensitivity, weakened vagal anti-inflammatory reflex) resulting from long-term psychological stress (Liver Stagnation). This fusion concept both explains why "Liver-soothing and Depression-resolving" formulas are effective for certain chronic inflammatory diseases and provides a target network for developing new drugs targeting this specific response pattern.

IX. From Parts View to Ecology View: A Paradigm Shift in Medical Cognition

The Response Mapping Hypothesis is not a negation of modern medicine. The achievements of modern medicine in object description — from the discovery of pathogens to gene localization, from targeted drugs to gene editing — are treasures of human civilization and an indispensable foundation for future medicine.

What the Response Mapping Hypothesis attempts to do is to complete the long-missing piece of the medical cognition puzzle. It originates from a simple first principle — life is about response. If we acknowledge this, then the center of gravity of medical cognition must shift from "what damage has occurred" to "whether the system can still respond correctly."

This shift will bring cognitive elevation at three levels:

First, re-understanding complex diseases. Metabolic syndrome, multimorbidity, and aging are no longer understood as simple superpositions of multiple independent diseases, but as different manifestations of the same system response network imbalance. The intervention target is no longer a single "part," but the reshaping of the entire "response network."

Second, the mathematization of sub-health. Traditional physical examinations look at whether indicators "exceed standards"; response mapping looks at the covariant structure among multiple indicators — whether network connections are loosening, whether regulatory loops are becoming sluggish. This "network health" measurement can issue warnings years in advance during the compensatory phase, transforming "treating diseases before they arise" from an empirical concept into a quantifiable scientific practice.

Third, paradigm transformation in drug development. No longer asking only "whether this molecule hits that target," but asking "whether this intervention can pull the abnormal response sub-network in the disease state back into the healthy state's attractor basin." This provides an entirely new computational framework for natural product development, drug repurposing, and individualized intervention.

This is the evolution from a parts view to an ecology view, and the necessary path from precision medicine toward precision medicine (from precisely locating targets to precisely regulating network states).

X. Conclusion: Finding the Coordinates for "Wind"

Returning to the confusion at the beginning of this article: What exactly is "Wind" in TCM?

After this journey of deduction, we can perhaps offer a clearer answer: "Wind" is not an object of physics, nor an entity of etiology. It is the abstraction and naming of a category of human body response patterns from two millennia of TCM clinical practice. Today, we can attempt to find computable molecular coordinates for this ancient concept using modern state spaces such as PPI networks, metabolic fluxes, and epigenetic markers.

The significance of this answer extends far beyond the single concept of "Wind." It points to a larger possibility: in the era of AI for Science, we can not only use AI to optimize existing diagnostic and treatment processes, but also use AI to discover and define entirely new medical concepts — response pattern concepts that humans, due to the limitations of experiential induction, have never discovered, yet are crucial for understanding complex diseases.

From the extreme compression of mTOR, to the network expansion of SEMO, to the macro-level emergence of Hallmarks Engineering — the logic of response mapping has always been present, merely lacking a unified name and framework. The Response Mapping Hypothesis provides precisely a cognitive framework and theoretical starting point for this logic.

Starting from the confusion about "Wind," we have arrived at a new understanding of the nature of medical language. And from this understanding, we are only just beginning to move toward the true reconstruction of medicine's conceptual system in the AI era. In the future medical landscape, human experts will be responsible for providing first principles and ethical boundaries, while AI will truly evolve from the Copilot of assisted computation into the Co-Scientist who reconstructs human knowledge systems.

References

1. Xiong, J. AI-Generated Hallmarks of Aging and Cancer: A Computational Approach Using Causal Emergence and Dependency Networks. bioRxiv 2024.08.28.610200 (2024).

2. Hoel, E. P., Albantakis, L., & Tononi, G. Quantifying causal emergence shows that macro can beat micro. Proceedings of the National Academy of Sciences, 110(49), 19790-19795 (2013).

3. Barabási, A. L., Gulbahce, N., & Loscalzo, J. Network medicine: a network-based approach to human disease. Nature Reviews Genetics, 12(1), 56-68 (2011).

4. Gan, X., Shu, Z., Wang, X., Yan, D., Li, J., et al. Network medicine framework reveals generic herb-symptom effectiveness of traditional Chinese medicine. Science Advances, 9(43), eadh0215 (2023).

5. Horvath, S., & Raj, K. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nature Reviews Genetics, 19(6), 371-384 (2018).

6. Noble, D. The principles of systems biology. Experimental Physiology, 102(12), 1599-1606 (2017).

7. Yang, Z. F., Qian, H., et al. Impact of airflow stimulation: mild cold airflow is more sensitive to influenza virus infection. Journal of Thoracic Disease, 17(6) (2025).

8. Hall, M. N. On mTOR nomenclature. Biochemical Society Transactions, 41(4), 887-888 (2013).

9. Lee, M., et al. Spacer: Towards Engineered Scientific Inspiration. arXiv:2508.17661 (2025).

10. Fang, Y. L., et al. AI-Newton: A Concept-Driven Physical Law Discovery System without Prior Physical Knowledge. arXiv:2504.01538 (2025).

11. Liao Yan, Xu Haodong, et al. The influence of Wind-Dampness-Phlegm-Stasis on immune homeostasis from the perspective of Orthodox and Pathogenic Qi. China Medical Herald, 19(28), 120-123 (2022).

12. Xiong, J. Utilizing Pre-trained Network Medicine Models for Generating Biomarkers, Targets, Re-purposing Drugs, and Personalized Therapeutic Regimes: COVID-19 Applications. Cold Spring Harbor Laboratory (2023).

13. Xiong Jianghui, Liang Fengji, Chang Suhua. DNA methylation-based quantitative methods for TCM concepts and syndromes and their applications in depression and cancer screening. Chinese Journal of Pharmacology and Toxicology, 37(S1) (2023).

14. Ruiz, C., Zitnik, M., & Leskovec, J. Identification of disease treatment mechanisms through the multiscale interactome. Nature Communications, 12, 1796 (2021).