Nested Path Modeling Analysis of TCM Diagnostic AI

Nested Path Modeling Analysis of TCM Diagnostic AI System Based on DIKWP Model

Yucong Duan

International Standardization Committee of Networked DIKWPfor Artificial Intelligence Evaluation(DIKWP-SC)

World Academy for Artificial Consciousness(WAAC)

World Artificial Consciousness CIC(WAC)

World Conference on Artificial Consciousness(WCAC)

(Email: duanyucong@hotmail.com)

1. Introduction: Background of TCM AI Diagnosis and the Necessity of Famous Doctor Avatars

In recent years, the application of artificial intelligence in the medical field has developed rapidly, but it still faces many challenges in Traditional Chinese Medicine (TCM) diagnostic scenarios. Traditional medical AI systems are often like "black boxes," making their decision-making processes difficult to understand and lacking transparent semantic understanding mechanisms. This not only leads to the model's insufficient grasp of complex TCM theories but also makes it difficult for the diagnoses and suggestions provided by AI to gain the trust of doctors and patients. On one hand, TCM diagnosis emphasizes Syndrome Differentiation and Treatment, involving multi-dimensional concepts such as Yin and Yang, Exterior and Interior, Cold and Heat, and Deficiency and Excess. It requires complex reasoning based on the synthesis of the patient's symptoms and signs. If the internal reasoning link of the AI model is invisible, its differentiation process remains a black box, and physicians and patients cannot understand why the AI reached a certain diagnosis, significantly discounting trust.

On the other hand, clinical experience in TCM relies heavily on the knowledge inheritance of famous veteran TCM doctors. Many famous TCM doctors have formed unique diagnosis and treatment styles and valuable experience through decades of clinical practice. However, the traditional apprenticeship mode is inefficient, and the experience of famous doctors is easily lost. The TCM industry faces problems such as a shortage of famous doctor resources and long doctor training cycles. If AI technology can be used to precipitate and reproduce the diagnostic thinking of famous doctors, it will greatly promote TCM inheritance and medical inclusivity. Therefore, the concept of "Digital Avatars of Famous Doctors" has emerged, which involves using artificial intelligence to create a personalized "Smart Brain" for each famous doctor, allowing their experience and style to continue and assist more patients.

Currently, some internet medical companies have begun to explore the practice of AI implementation for famous doctors. For example, Ant Group launched the AI health assistant AQ, which features nearly 200 AI avatars of famous doctors. By imitating real doctors to gradually inquire about the condition and guiding users to provide information, it gives health suggestions. It is reported that to train the AI avatar of a famous TCM doctor in Hangzhou, the research team used up to 100 hours of doctor-patient dialogue data and incorporated the doctor's own refined clinical knowledge, references, and diagnosis and treatment guidelines for training. Such AI avatars can provide consultation and advice in the style of the famous doctor online, realizing functions such as pre-diagnosis screening and post-diagnosis follow-up, while seamlessly docking with real doctors when necessary to provide continuous medical services for patients. It can be seen that the industry consensus is not to let AI replace doctors, but to let TCM doctors who can use AI expand their service capabilities.

However, to truly realize "AI Famous Doctors," it is necessary to solve the problems of explainability and personalized modeling of AI for the TCM diagnostic process. The diagnosis and treatment ideas of famous doctors from different schools vary widely—TCM has many schools such as the Typhoid School (Shanghan), the Warm Disease School (Wenbing), and the Fire Spirit School (Huoshen), each with a relatively independent and complete thinking system. Current general TCM AI models struggle to cater to all schools. Lacking sufficient high-quality, structured TCM case data and unified evaluation standards, there is still a gap between model capabilities and independent syndrome differentiation and treatment. An industry expert pointed out: "TCM emphasizes 'differentiation.' Currently, AI cannot independently complete the Four Examinations. Even with intelligent hardware, there are still many missed diagnoses. The Four Examinations are not just information collection; the key is to learn the doctor's 'chain of thought' during inquiry, learning their way of asking questions." That is to say, AI not only needs to acquire data such as the patient's tongue image and pulse image but also needs to master the process of how famous doctors reason, follow up with questions, and make decisions based on symptom clues. This is the core difficulty of modeling "Digital Avatars" of famous doctors and is also the focus of this paper: How to use the DIKWP Model to intelligently model the cognitive path of the TCM diagnostic link, enabling AI to think dialectically like a famous doctor while demonstrating the reasoning chain in a transparent and explainable manner.

2. Overview of the DIKWP Model and Introduction to Nested Mechanisms

To solve the "black box" problem in AI diagnosis, Professor Yucong Duan proposed the DIKWP Semantic Cognitive Model, which divides the cognitive process of intelligent agents into five levels: Data (Data), Information (Information), Knowledge (Knowledge), Wisdom (Wisdom), and Purpose (Purpose). The DIKWP model is an extension of the classic DIKW (pyramid) model, adding the Purpose dimension at the top of Data-Information-Knowledge-Wisdom to fully express the goals and motivations behind decisions. These five levels form a cognitive chain from perception to decision-making, with each level performing its own function:

Data Layer (D): Raw input objective facts, unprocessed symbols, or signals. In TCM diagnosis, this corresponds to the chief complaints stated by the patient and raw signs obtained from inspection, listening, smelling, and palpation. For example, temperature readings, tongue coating images, pulse waveforms, or direct descriptions like "I have been coughing for three days and have a sore throat." The data layer provides material for subsequent analysis.

Information Layer (I): Meaningful details refined after organizing and processing data. This layer answers "what kind of phenomenon occurred." In the TCM scenario, the information layer corresponds to the doctor's preliminary classification and description of symptoms and signs. For example, organizing the patient's symptoms into attributes of cold/heat, disease location, and symptom characteristics. If a patient says "phlegm is yellowish and sticky," for the doctor, "yellow" implies the existence of internal heat, which is a pathogenesis clue extracted from the data. However, it should be noted that much information is still embedded in the data and requires the doctor's professional knowledge to interpret it to rise to the information level.

Knowledge Layer (K): Regular understanding, medical knowledge, and pathological models condensed on top of information. This layer answers "what mechanism/syndrome do these symptoms reflect." The "Zheng" (Syndrome) in TCM diagnosis is located in the knowledge layer: the doctor synthesizes symptom information and associates it with existing TCM theories, such as the Eight Principles and etiology/pathogenesis, to determine a Syndrome Category. For example, through a series of manifestations, the doctor summarizes that the patient belongs to a pathogenesis pattern of "Wind-Heat invading the Lung, Lung loss of diffusing and descending," and names it "Wind-Heat invading Lung Syndrome." The knowledge layer involves raising specific information to abstract concepts to explain the internal laws of the disease. Many TCM technical terms (such as "Wind-Heat invading Lung," "Lung Heat Accumulation") are products of the knowledge layer; they are highly condensed but contain multi-layer meanings of etiology, location, and nature. It must be emphasized that these knowledge concepts themselves nest the underlying data and information: if a knowledge layer concept is given directly without expanding the underlying information, it is difficult for laypeople to understand its meaning. Therefore, when modeling, we must be able to link knowledge layer concepts back to the symptoms and mechanism information that support them, achieving a correspondence of upper and lower semantics.

Wisdom Layer (W): The ability to make comprehensive judgments and decisions based on knowledge. This layer answers "how to solve the problem, and what is the best decision." For TCM doctors, the wisdom layer is embodied in diagnostic thinking and treatment decisions. That is, the doctor uses mastered knowledge to make a plan for syndrome differentiation and treatment for a specific patient. For example, after confirming that a patient fits the "Wind-Heat invading Lung" syndrome (Knowledge Layer), the doctor will further judge what treatment plan should be adopted, such as "diffusing the Lung and clearing heat, resolving phlegm to stop coughing." This process applies abstract differentiation knowledge to specific practice and is the transformation of knowledge into action. The wisdom layer also includes the doctor's ability to adjust and optimize the plan according to the patient's specific situation to ensure the decision is most appropriate.

Purpose Layer (P): The purpose and motivation behind the cognitive process, the highest level of driving force. It answers "what goal is to be achieved ultimately, and for what." In medical scenarios, the purpose layer is often embodied as clinical goals and the purpose of practicing medicine. For example, the doctor's intention in diagnosing and treating a disease may be "to find the cause accurately and explain it to the patient as soon as possible to guide treatment." Or in a teaching scenario, the AI's purpose might be "to let middle school students understand TCM diagnostic concepts," thereby determining its expression strategy. During clinical diagnosis, the purpose layer influences the application of the underlying wisdom and knowledge, ensuring that decisions align with the patient's health goals. It can be said that the purpose layer sets the tone for the entire cognitive chain; for example, if the doctor's purpose is "to clear external pathogens to avoid the disease going internal," their treatment decision in the wisdom layer will tend towards timely resolving the exterior.

The DIKWP model organically connects the above levels, structurally representing the entire process of intelligent diagnosis from low-level perception to high-level decision-making and even goal orientation. The layers are closely linked: Data provides the perceptual basis, Information organizes detailed clues, Knowledge refines underlying pathological laws, Wisdom applies laws to formulate plans, and Purpose calibrates the direction from the top level. Because of this layered mapping, the DIKWP model is considered to achieve a "white-box" presentation of traditional cognitive processes that are difficult to observe. This is especially important for scenarios requiring explanation and communication like medical diagnosis: we can clearly locate which level a concept or reasoning step belongs to, thereby adjusting the method of explanation for different audiences. For example, for professional doctors, communication can focus on the knowledge and wisdom layers, while for patients, medical terms in the knowledge layer need to be broken down into the information layer, explaining the correspondence between causes and symptoms in plain language.

It is worth noting that the DIKWP model is not limited to a linear pyramid structure but has the characteristic of Networked Nesting. In the cognitive process of a single subject, high-level concepts often contain combinations of lower-level elements. For example, the knowledge layer syndrome "Wind-Heat invading Lung" semantically spans the information layer and the knowledge layer: the concept of "Wind-Heat" refines the nature of the cause (Information I), and the concept of "invading Lung" explains the visceral mechanism of the pathogen's action (Knowledge K). Another example is "Lung Heat" in "Lung Heat Accumulation Syndrome," which specifies the disease location and nature (heat pathogen in the lung, belonging to Information I), while "Accumulation" describes the intense disease trend causing Lung Qi blockage (Knowledge K). In other words, every TCM diagnostic term is actually a summary of deeper symptoms and mechanisms: the patient may have a series of raw symptom data (D), the doctor summarizes them into certain symptom pattern information (I), then uses existing knowledge concepts to explain and classify them (K), and finally expresses them in concise terms. This reflects the Nested Semantic Mechanism of DIKWP: nodes in the knowledge layer actually connect to rich content in the information and data layers, and we can unfold high-level concepts back to low-level elements to trace their roots.

In addition, in multi-subject interaction (such as doctor-patient dialogue), the DIKWP model can also map the situation where information is transmitted and transformed between different subjects bidirectionally. This is called the DIKWP×DIKWP interaction model. Simply put, it can be understood as a 5×5 matrix describing how the output of a certain layer of one party becomes the input of a certain layer of another party, with a total of 25 possible semantic conversion forms. For example, "Knowledge" (K, a professional diagnostic concept) output by a doctor may be just "Information" (I, some medical explanation needing understanding) for the patient, or "Data" (D, raw symptoms) provided by the patient is transformed into "Information" (I, pathogenesis clues) recognized by the doctor after interpretation. This matrix-style interaction depicts the mechanism of cross-layer transmission of semantic content between doctors and patients, providing a model basis for analyzing cognitive alignment and deviation. in the diagnostic scenarios focused on in this paper, we will see similar phenomena: raw symptoms described by the patient (Patient D) become important information in the doctor's eyes (Doctor I), and professional judgments proposed by the doctor (Doctor K) need to be translated into information that the patient can understand (Patient I) for communication. This mapping and nesting between DIKWP levels allow us to model the semantic process of TCM diagnosis more comprehensively, including both the layered evolution of the doctor's internal reasoning chain and the extraction and interpretation of information between doctor and patient.

In summary, the DIKWP model provides a multi-level, explainable theoretical framework for TCM AI diagnosis. It allows us to map the diagnostic ideas of famous doctors into a series of semantic levels from data to purpose, and to further subdivide and nest the internal structure of each concept. In the following sections, we will reconstruct the diagnostic path of famous doctors based on the DIKWP model to realize the layered nested mapping and semantic unfolding of the diagnostic reasoning chain.

3. Collection of Famous Doctor Diagnosis Paths and Abstraction of Concept Structures

To build a diagnostic intelligent model of famous TCM doctors, the primary task is to collect and organize the diagnostic paths of famous doctors and abstract the conceptual structures behind them. This includes obtaining various data from the famous doctors' diagnosis and treatment processes, as well as refining the key knowledge and logic that embody their diagnostic thinking. Specifically, we can proceed from the following aspects:

(1) Clinical Cases and Multimodal Data Collection:

The diagnostic experience of famous doctors is mainly reflected in their rich clinical cases. Therefore, we need to collect a large number of high-quality medical cases and diagnosis records of famous doctors. This can include accumulated medical case texts over the years, electronic medical records from outpatient clinics, and typical case analyses published by famous doctors. According to technical reports, existing data platforms have integrated 100,000+ diagnosis records and medical cases from ancient TCM books, laying the foundation for knowledge graph construction. In addition, unstructured data from famous doctors' diagnoses should also be collected, such as video recordings of outpatient inquiries, voice records, scans of handwritten prescriptions, etc., to enrich the model's multimodal perception of the diagnosis process. Using OCR and voice recognition/NLP technologies, these unstructured data are converted into analyzable text and parameters. At the same time, attention should be paid to the collection of physiological data such as Inspection, Listening, Smelling, Inquiry, and Palpation: equipped with high-definition cameras to collect face and tongue images, electronic pulse diagnostic instruments to obtain pulse waveforms, recording equipment to record doctor-patient dialogues, etc. By fusing multimodal data and constructing a comprehensive "digital portrait" of the patient, the basis for the famous doctor's differentiation can be more accurately restored.

(2) Key Information Annotation and Structuring:

With raw data, it is also necessary to clean and annotate it to extract key information contained in the diagnosis process of famous doctors. For example, extract fields such as Chief Complaint, History of Present Illness, Findings from Four Examinations, Past History from medical case texts, and annotate the TCM attributes corresponding to each symptom or sign (Cold/Heat, Deficiency/Excess, Associated Organs, etc.). The questions asked by the famous doctor during the inquiry and the details observed must also be recorded and attached with semantic labels (e.g., Is this confirming an Eight Principles element? Verifying a syndrome hypothesis?). This process can invite TCM experts to manually proofread and annotate the data to ensure accuracy. Through structuring and annotation, originally messy text records are transformed into standardized data representations, laying the foundation for subsequent modeling. For example, the annotation points out that "coughing for three days, sore throat, fever" belongs to patient Chief Complaint Data (D); "red tongue, yellow greasy coating" belongs to Inspection results (Data D) and implies "internal excess heat" (Information I); in the inquiry, the doctor asks "Are you afraid of cold? Any sweating?" to distinguish Cold/Heat (Purpose P, invoking Knowledge K)... Marking these points captures the information flow in every step of the famous doctor's diagnostic thinking.

(3) Concept System and Knowledge Graph Construction:

The reason why famous doctors are brilliant lies in the systematic medical knowledge and empirical rules in their minds. We need to make this implicit knowledge explicit and establish a TCM Knowledge Graph or concept network exclusive to the famous doctor. Specific practices include: organizing the Syndromes, Therapeutic Methods, Prescriptions, and their relationships commonly used by the famous doctor. In this regard, existing TCM knowledge standards can be utilized. For example, based on the semantic framework of the TCM Clinical Terminology System (TCMCTS), normalize the levels and term definitions of concepts such as syndromes, diseases, and formulas. Then, align the syndromes appearing in the famous doctor's medical cases with these standard terms and map the relationship graph between them. Nodes in the knowledge graph can be entities such as syndromes, symptoms, drugs, prescriptions, etc., and edges represent relationships (such as "Syndrome A contains Symptom X," "Prescription B treats Syndrome A," "Doctor prefers using Drug C under Syndrome A," etc.). Special attention should be paid to the personalized knowledge of famous doctors, such as the disease fields they excel in, unique differentiation points, and medication characteristics, which should be reflected in the graph. For example, National Medical Master Zhang Lei has rich experience in treating internal diseases. He advocates the concept of "Achieving Harmony (Zhi Zhong He)" and emphasizes the regulation of the patient's "Heart Disease (Mental aspect)," even forming a unique "Prescription without Medicine" style. Therefore, in the knowledge graph, nodes representing the treatment method of "Psychological Counseling" and its connection with relevant syndromes (such as Depression/Yu Zheng) can be added. Another example is the late famous doctor Yue Meizhong, who proposed the principle of combining specialized formulas for specialized diseases with syndrome differentiation and treatment, and was good at using classical prescriptions (Jingfang) to treat acute and severe diseases. The graph can reflect the classical prescriptions he often selected for certain specialized diseases (such as high fever convulsions) and the corresponding differentiation points. Through the knowledge graph, we transform the empirical laws implied by famous doctors into a network structure understandable by programs.

(4) Extraction of Famous Doctor Differentiation Logic:

After possessing data and a knowledge graph, it is also necessary to refine the logical flow and decision rules in the famous doctor's diagnostic thinking. This can be seen as superimposing a Reasoning Engine on top of the knowledge graph. For example, summarize a set of Differentiation Flowcharts or Decision Trees used by the famous doctor when diagnosing a certain type of disease. Technical solutions mention that combined with expert feedback, Clinical Decision Trees can be constructed to restore the famous doctor's differentiation path, thereby simulating their diagnostic thinking. For instance, for a case of recurrent cough, a famous doctor might follow this reasoning logic: First determine Exterior Syndrome or Interior Syndrome? --> If there is aversion to cold and fever, etc., it belongs to Exterior Syndrome, then determine Wind-Cold or Wind-Heat? --> Judge whether it is Wind-Heat (sweating, yellow phlegm) or Wind-Cold (no sweating, clear watery phlegm) through the presence of sweat and phlegm color; if there is no Exterior Syndrome, consider Interior Syndrome cough, and further distinguish Phlegm-Dampness, Yin Deficiency, etc. This Differentiation Process can be represented by a tree structure, where each node is a judgment condition, each branch represents a different differentiation direction, and the final leaf node corresponds to a specific syndrome type and treatment plan. By analyzing a large number of famous doctors' medical cases, we can statistically determine the differentiation nodes and decision conditions commonly used by them. For example, when Master Zhang Lei treats difficult and miscellaneous diseases, he "focuses on treating the pathogenesis." Whenever encountering a complex case, he must seek its Main Pathogenesis and then treat it specifically. This suggests to us that in his decision tree, every step closely follows the main line of "pathogenesis": First distinguish if the pathogenesis is Blood Stasis? Phlegm-Fluid? Qi Stagnation? Then select formulas and medicines to directly attack the pathogenesis. After these decision rules are refined, they can be transformed into Reasoning Statements or Algorithms in the reasoning engine (such as rule-based reasoning systems, or training a decision tree/random forest model). In addition, Uncertainty Reasoning mechanisms can be introduced to handle cases where symptom manifestations are vague or atypical in clinical practice. For example, for cases where symptoms do not completely fit a certain syndrome type, Bayesian inference or fuzzy logic is introduced to assign a certain degree of confidence to multiple syndrome types, and then the Wisdom layer decides to select the most suitable one. This flexibility can improve the model's ability to differentiate complex cases.

(5) Modeling Famous Doctor Diagnosis Style Characteristics:

Every famous doctor has their unique Diagnosis and Treatment Style and Academic School background, and we should fully consider these personalized characteristics when modeling. For example, Taoist doctor Zhang Zhishun, as a Taoist physician, proposed the concept of "Zhenyuan Medicine," emphasizing treating the person before the disease, prioritizing prevention over treatment, and focusing on health preservation methods such as diet and breathing. His inquiry style might pay special attention to information such as the patient's daily life and emotional regulation. When AI simulates his diagnosis, more questions related to emotions and lifestyle should be set in the "Inquiry" link, and Taoist health preservation theories should be added to the knowledge base. Another example is National Medical Master Deng Tietao, who dedicated his life to the integration of Traditional Chinese and Western Medicine. He conducted in-depth research and innovation on the TCM diagnostic system. Elder Deng proposed some academic propositions regarding diagnostic methods, such as attaching great importance to the role of pulse diagnosis in differentiation and advocating "Reference of Pulse and Symptoms," etc. If simulating Elder Deng's diagnosis, the AI system needs to strengthen the information weight of the "Palpation" link and introduce his special rules for the correspondence between pulse images and pathogenesis. The Lingnan TCM school characteristics he inherited should also be reflected in the knowledge graph. Furthermore, Professor Yue Meizhong advocated applying Classical Prescriptions (Jingfang) to modern difficult and severe diseases, often "treating major diseases with specialized formulas." This means that when AI learns Elder Yue's cases, it should pay special attention to his ideas on prescription selection: when encountering certain critical syndromes, directly associate them with ancient famous formulas instead of rigidly sticking to complex prescriptions of later generations. Therefore, in the reasoning engine, a priority can be given to the syndromes matched by these classic prescriptions. When patient symptoms match the original text of a certain classic prescription, the AI boldly recommends it (this is exactly Elder Yue's style of "lifting major diseases with classical prescriptions"). Through the above methods, the personal style of famous doctors is integrated into the model, making AI decisions closer to the thinking mode of specific famous doctors.

Through Data Level Collection and Knowledge Level Abstraction of famous doctors' diagnostic paths, we have laid the foundation for subsequent modeling based on DIKWP. Simply put, it is to comprehensively depict the process of "how a famous doctor asks, sees, thinks, and decides" using data and knowledge. Next, we will build the DIKWP*DIKWP nested path model framework on this basis to specifically demonstrate the layered mapping of famous doctor diagnostic cognition.

4. Construction of DIKWP*DIKWP Diagnostic Path Modeling Framework (Four Examinations + Chief Complaint + Conditioning Ideas)

In this section, we build a TCM diagnostic path model integrating the DIKWP model to simulate the entire process of a famous doctor from obtaining patient information (Chief Complaint + Four Examinations) to forming a diagnosis and formulating conditioning ideas (treatment plan) in actual diagnosis and treatment. Since the TCM "Four Examinations"—Inspection, Listening/Smelling, Inquiry, and Palpation—are the basic links of diagnosis, we will focus on the information acquisition and processing of these four examinations, combined with DIKWP levels, to describe how AI reproduces the reasoning chain of famous doctors. At the same time, the DIKWP×DIKWP interaction perspective is introduced to reflect the correspondence of information in different semantic layers during doctor-patient dialogue.

Framework Summary: After the patient enters the consultation room, they first provide the Chief Complaint (main symptoms and discomfort). The doctor collects comprehensive information through Inspection (Observation), Listening/Smelling (listening to voice, smelling odor), Inquiry (asking about symptoms and history), and Palpation (pulse taking, touching). The AI famous doctor system takes this as input, processes it through multiple DIKWP layers, and outputs Differentiation Results and Conditioning Ideas (treatment plans). The whole process includes an outer Diagnostic Process DIKWP Chain and multiple inner Sub-chains (for example, the information acquisition of each diagnostic method itself can be viewed as a DIKWP process). Details are as follows in order:

4.1 Chief Complaint Reception and Preliminary Information Classification (Data → Information)

The patient's chief complaint is often the first-hand data (D) contacted by the AI system. For example: "Coughing for three days, sore throat, accompanied by low fever." The AI parses this into a structured symptom list and marks preliminary attributes: Time Length (three days), Location (sore throat involves the Lung system), Nature (fever indicates heat sign, pain nature implies excess syndrome), etc. This step is equivalent to converting the patient's subjective description into Diagnostic Information (I) usable by the model. Just like the previous example, "cough," "sore throat," and "fever" are inherently data layer content, but after organization, we refine information clues such as "External sensation symptoms for 3 days, suspected Exterior Heat Syndrome." The AI famous doctor will quickly compare these preliminary pieces of information with disease and syndrome patterns in its knowledge base (e.g., knowing that cough + sore throat + fever is common in Wind-Heat invading Lung, Wind-Cold binding Lung, etc.) as a basis for the subsequent inquiry direction.

4.2 Inspection Information Processing (Data → Information/Knowledge)

The doctor next obtains data on the patient's external signs through Inspection, including complexion, tongue image, body posture, etc. At this time, the AI system may receive image data from cameras and body surface detection data. For example, the doctor sees that the patient has "yellow thick tongue coating, red tongue body," which is first recorded as perceptual input in the data layer. After analyzing the image, the AI extracts Information Layer points such as Red Tongue Body (indicating heat sign) and Yellow Thick Greasy Coating (indicating Phlegm-Heat or Damp-Heat). This information maps directly to TCM knowledge: red tongue and yellow coating are important features of Heat Syndrome. Combined with cough and sore throat, it points to heat pathogen in the Lung Meridian. Therefore, the AI can form a Knowledge Layer Hypothesis as early as the inspection stage: "It may be a syndrome like Wind-Heat invading Lung or Lung Heat Accumulation." This reflects the process where inspection data is refined by Layer I and activates Layer K knowledge. At the same time, the AI also checks the patient's facial expression and form, such as whether the complexion is dull (Qi and Blood Deficiency?), or if there are rashes on the skin (possible outward eruption of Wind-Heat?). These are stored as supplementary information. Note: Information obtained from inspection is often directly used for subsequent differentiation reasoning. Because it is objectively visible, many famous doctors are accustomed to "knowing by seeing." AI can mark the information extracted from inspection with high confidence and give it greater weight when synthesizing evidence.

4.3 Listening/Smelling Information Processing (Data → Information)

Listening and Smelling includes listening to sounds and smelling odors. Modern AI can collect the patient's cough sound and speaking voice through microphones, and detect the patient's breath and body odor through sensors. Ping An Good Doctor's "Smart Listening/Smelling" system follows a similar idea, quickly collecting user voices and analyzing them via AI to determine if they belong to TCM constitutions like Qi Stagnation, Qi Deficiency, or Yang Deficiency, achieving "diagnosing disease by listening to sound." In our framework, the AI analyzes the patient's voice characteristics and odor characteristics data (D): for example, whether the cough sound is muffled or hoarse (indicating Phlegm-Dampness or Lung Dryness), whether speech is weak (Qi Deficiency?), or if there is a special odor like putrid smell (possible Lung Abscess), etc. These perceptions are converted into text-described Information (I). If the patient's cough is loud and accompanied by a hoarse throat and odor in the mouth, the AI can extract information like "Cough sound is coarse and harsh (belonging to Excess Heat)" and "Breath is hot and foul (Stomach/Intestine accumulated heat)." At this point in the Knowledge Layer K, the AI finds increased evidence associated with "Lung Heat Blazing or Phlegm-Heat Obstructing Lung." Conversely, if the listening diagnosis reveals the patient's voice is low and faint, breath is short, and there is no obvious heat odor, it leans towards Deficiency Cold Syndrome. It can be seen that data provided by listening/smelling, processed by Layer I, also directly supports certain Knowledge Layer judgments. It needs to be pointed out that currently, many manufacturers have launched tongue diagnostic instruments, pulse diagnostic instruments, inquiry robots, etc., for collection hardware, but the effect in practical application is limited. The reason lies precisely in the fact that data obtained from the four examinations need the doctor's chain of thought to interpret, and hardware itself cannot replace reasoning. Therefore, in our model, every piece of listening/smelling data does not exist in isolation; it must be matched to meaning by the knowledge base combined with the overall context to enter subsequent decision-making.

4.4 Inquiry Interaction and Information Completion (Data ↔ Knowledge → Wisdom)

Inquiry is the most flexible and critical link in TCM diagnosis. In the AI famous doctor system, the inquiry process is embodied as dialogue interaction between AI and the patient; that is, the DIKWP×DIKWP interaction is most frequent here. The AI doctor will actively ask questions based on the currently obtained information and knowledge layer hypotheses to supplement data or verify inferences. This is actually a bidirectional process from Knowledge Layer/W Layer down (Doctor side) and from Data Layer up (Patient side):

AI Questioning (Doctor side Wisdom W Layer behavior, with Purpose P): When the AI forms a preliminary differentiation idea based on chief complaint and inspection/listening/smelling information, it will ask targeted questions to obtain further data. For example, in the case, the doctor asks the patient: "Are you afraid of cold these days? How is the sweating?" This question is not random but stems from a clear diagnostic Purpose (P): To distinguish whether the patient belongs to Wind-Cold or Wind-Heat. The doctor mobilized Experience in Knowledge Layer K: "Those with external Wind-Cold often have aversion to cold and no sweat, while those with Wind-Heat often have fever, slight aversion to cold, and sweating or little sweat." Based on this knowledge, he formulated a strategy to inquire about cold/heat symptoms in Wisdom Layer W to obtain corresponding Data D, thereby verifying his judgment on the syndrome type. Therefore, it can be said that behind this question connects the doctor's Knowledge (K: "Differentiation points of Wind-Heat/Wind-Cold") and Purpose (P: "Clarify Cold/Heat attributes"), belonging to a Diagnostic Inquiry of the Wisdom Layer. Similarly, the AI might continue to ask: "Is there much phlegm? What color is the phlegm?" This is also to obtain the nature of the phlegm. The presence and color of phlegm are raw data, but they hide pathogenesis information; for example, yellow phlegm indicates internal heat. The doctor realizes this using knowledge, so they obtain this data through inquiry. It can be seen that every question asked by the AI has a corresponding knowledge basis and diagnostic purpose behind it, reflecting the mechanism of high-level guiding low-level in the DIKWP model.

Patient Response (Patient side Data D provision): Facing the AI's question, the patient gives an answer, which is new data input. For example, the patient answers: "I was a bit afraid of cold at first, now it's just fever, not much sweat." This provides Data D regarding the evolution of cold and heat: initial aversion to cold, followed by mainly fever, and little sweating. The patient may not have summarized the meaning of these changes, but the AI doctor will interpret them as Information I: persistent aversion to cold turning into high fever with little sweat conforms to the typical manifestation of Wind-Heat Exterior Syndrome. Thus, the AI integrates this answer into its syndrome judgment. From the DIKWP level perspective, the patient's answer itself stays at the Data Layer (only describing phenomena, without analyzing causes), while the Doctor's brain elevates it from Data to Information/Knowledge Layer: interpreting "initial aversion to cold, followed by fever" as a clue of pathogenesis evolution "Exterior pathogen transforming from Wind-Cold to Heat and entering the interior." This process is completed inside the AI and not necessarily stated to the patient, but for the model, it is a transformation of D→I→K. Similarly, the patient answering "There is phlegm, yellowish and sticky" provides qualitative data about phlegm. The AI records the data point "Phlegm is yellow and sticky," then relates it to the rule "Yellow sticky phlegm = Lung Heat or Damp-Heat" in the knowledge base, transforming it into information: "Patient's phlegm belongs to heat sign." Thus, whenever the patient responds, the AI interprets and elevates the data, thereby constantly perfecting the cognitive structure of the syndrome.

Through such multi-round Q&A interactions, the AI famous doctor gradually perfects the chain of evidence and approaches the final diagnostic conclusion. It is worth mentioning that in this process, the AI can adjust its questioning strategy at any time based on new information. For example, if the patient presents unexpected symptoms, the AI may trigger another set of inquiry flows (e.g., originally thinking it was Wind-Heat, but the patient says no fever but severe aversion to cold, the AI turns to ask for more detailed manifestations of cold syndrome). This dynamic adjustment is the embodiment of the Wisdom Layer. In summary, the DIKWP Chain in the Inquiry Stage often presents a cyclic iteration: Doctor W Layer sets question based on K Layer knowledge -> Patient D Layer replies -> Doctor elevates D to I/K -> If needed, W Layer decides next question, until information is sufficient.

4.5 Palpation Information Processing and Verification (Data → Information)

Palpation mainly refers to Pulse Diagnosis, and also includes Touch Diagnosis (Pressing Diagnosis). The AI system can obtain pulse waveform data from the patient's wrists via intelligent pulse diagnostic instruments, and measure some body surface indicators (such as skin temperature, pressure points) via wearable sensors. These physical signals belong to Data Layer D, but need to be converted into Information I such as TCM Pulse Images and findings from touch diagnosis. For example, after pulse waveform analysis, the AI might determine "Pulse is Slippery, Rapid and Forceful," and touch diagnosis finds tonsil enlargement and tenderness. These are organized as Information Layer content: Slippery Rapid Pulse with force usually corresponds to Interior Heat Excess Syndrome, and tonsil swelling and pain confirm Heat Toxin in the Lung Meridian. Actually, in our example, a similar scene appeared: The doctor muttered a summary while taking the pulse: "Sore throat and cough, phlegm is yellow and sticky, slight aversion to cold and fever...". This indicates that the doctor already knew the score during palpation, restating and integrating the information collected from inspection, listening, smelling, and inquiry earlier, preparing for the final differentiation judgment. The pulse result serves as a verification basis. If the pulse shows "Floating Rapid" (Exterior Heat Syndrome) or "Slippery Rapid" (Interior Heat with Phlegm Abundance), it matches the previous inference, confirming the diagnosis. If the pulse does not match the previous information, the AI might re-examine the previous inference. For instance, if "Deep Slow Forceless Pulse" appears, it must consider whether Cold Syndrome or Deficiency Syndrome was missed. Therefore, Palpation data is often used for Cross-Verification. In the model, logic can be set: Convert the pulse diagnosis result into Eight Principles pulse information and compare it with the Eight Principles attributes of the AI's current differentiation result. If there is a conflict, trigger Wisdom Layer adjustment (e.g., add inquiry details). In general, the role of Palpation in the DIKWP chain is, firstly, to provide the last batch of objective Data D, and secondly, to process this data into Information I significant for syndrome typing, thereby acting as corroboration or correction for the Knowledge K Layer judgment.

4.6 Comprehensive Analysis and Syndrome Differentiation (Information → Knowledge)

When the "Four Examinations" information collection is completed, the AI famous doctor system enters the Comprehensive Differentiation stage, which is identifying the patient's Syndrome (Diagnosis in TCM sense) from the complex information. Under the DIKWP framework, this corresponds to activities mainly performed in the Knowledge Layer (K). The AI aggregates the key information items obtained previously, such as: "Symptoms: Cough, throat redness, swelling and pain, expectorating yellow phlegm, fever, slight aversion to cold, little sweat; Tongue red, coating yellow and greasy; Pulse slippery rapid." Then, the AI calls the knowledge base for Pattern Matching and Reasoning: finding the pathogenesis and syndrome name that best fit based on these symptom patterns. For example, matching the rule: "Wind-Heat invading Lung Syndrome: Manifestations include heavy fever, slight aversion to cold, sore throat, coughing yellow sticky phlegm, red tongue, yellow coating, floating rapid pulse." The AI finds that the patient's information basically matches the "Wind-Heat invading Lung" syndrome type, so it sets it as the primary diagnosis. At the same time, note that the disease course may have internal transmission: if the patient has persistent high fever and incessant coughing/panting, it may evolve into "Lung Heat Accumulation" syndrome. The AI will also list this as a differential diagnosis. For instance, the case in Professor Yucong Duan's report shows that in the development process of Patient Cold-Pharyngitis-Bronchitis, the corresponding TCM diagnosis changed from initial Wind-Heat invading Lung Syndrome to Lung Heat Accumulation Syndrome. The AI should be able to judge the current stage based on symptom changes. If the system design allows compound syndromes, the AI can also output a combination of multiple syndromes (e.g., "Wind-Heat invading Lung combined with Phlegm-Dampness"), but there must be a basis. The advantage of the DIKWP Model is reflected here: Every determined Knowledge Layer conclusion (Syndrome Name) is not popped out of a black box but can be traced back to the underlying Information and Data it condenses. For example, if the AI determines "Wind-Heat invading Lung," it can simultaneously list the key information-data basis supporting this syndrome: External Wind-Heat pathogen (Etiology I), Lung Wei affected (Disease Location I), Symptoms include fever, sore throat, coughing yellow phlegm, etc. (Data D summary). This is exactly the meaning of the "Evidence" part in traditional medical cases. Through the DIKWP model, we can let the AI automatically bind the corresponding Information-Data basis when establishing the Knowledge Layer diagnosis, thereby achieving Explainability of the diagnostic conclusion.

4.7 Treatment Decision and Conditioning Idea Generation (Knowledge → Wisdom → Purpose)

Once the syndrome is identified, the AI enters the Treatment Decision stage, which is formulating conditioning ideas and specific prescription plans. According to the DIKWP model, this mainly happens in the Wisdom Layer (W) and is guided by the Purpose Layer (P). Specifically:

First, the AI retrieves the general therapeutic principles and common formulas for the syndrome from the knowledge base (Options provided by Knowledge Layer K). For example, for "Wind-Heat invading Lung" syndrome, the typical treatment method is Dispersing Wind and Clearing Heat, Diffusing Lung to Stop Cough; common formulas include Sang Ju Yin, Yin Qiao San, etc. For "Lung Heat Accumulation" syndrome, the treatment method should be Clearing Lung and Draining Heat, Resolving Phlegm and Soothing Throat, commonly using Ma Xing Shi Gan Tang and other lung-clearing heat-draining formulas. These belong to the Knowledge Layer Plans reserved in the knowledge base. The AI takes them as candidates.

Next, the AI makes choices and adjustments in the Wisdom Layer based on the patient's individual situation. This involves answering "What is the best plan." For example, if the patient has obvious abundant and sticky phlegm besides typical Wind-Heat symptoms, measures to resolve phlegm and stop coughing should be added on top of dispersing wind and clearing heat. The AI might decide to use Sang Ju Yin as a base and add lung-clearing phlegm-resolving drugs to enhance the cough-stopping effect. If the patient has a weak constitution, the Righteous Qi (Zheng Qi) must be cared for, so it is not suitable to use excessively bitter and cold drugs; perhaps a little Qi-supplementing and Yin-nourishing medicine is added to the formula. These subtle adjustments are the play of the Wisdom Layer, equivalent to the famous doctor's modification according to symptoms. As the report points out: "The Wisdom Layer is the application and practice of knowledge, ensuring appropriate decisions through reflection and adjustment." Therefore, the AI will not mechanically apply textbook prescriptions but simulate the famous doctor's thinking in prescription composition, flexibly adding or subtracting within the safety range to formulate personalized conditioning ideas.

When finally determining the plan, the AI also considers the guidance of the Purpose/Intention Layer (P). For example, for acute Wind-Heat invading Lung, the AI's treatment intent might be "Clear and resolve the exterior pathogen as soon as possible to avoid the pathogen transmitting internally"; for the stage of Lung Heat Accumulation, the intent lies in "Clearing Lung and Draining Fire to prevent lung damage." These intentions will affect plan details; for instance, the former seeks rapid resolution (perhaps heavier dosage, short course), while the latter might need to consider protecting Lung Yin. In human-computer interaction, the AI's purpose also includes explaining the treatment purpose to the patient: "Why use this formula, and what effect is expected." This is important for establishing patient trust. Famous doctors often explain "what is being treated" to patients while prescribing. The AI can similarly attach an explanation when outputting the prescription, such as "I am using modified Sang Ju Yin for you, aiming to disperse wind heat and clear lung to stop coughing. This can quickly bring down the fever and relieve sore throat, preventing inflammation from moving down to the lungs." This is actually the AI's elucidation of its own decision in the Purpose Layer.

Through the above steps, the AI generates a complete Conditioning Idea: including Diagnosis (Syndrome Type), Treatment Method, Prescription, and Expected Goal. Taking the case as an example, finally, the AI doctor says to the patient: "It is just that you caught Wind-Heat pathogen, causing a cold and sore throat. The heat pathogen invaded the Lung Meridian, so you cough. For treatment, we disperse wind and clear heat, diffuse lung and stop cough. I will first write a prescription called Sang Ju Yin...". This paragraph condenses all information from Knowledge Layer (Wind-Heat invading Lung diagnosis) to Wisdom Layer (Dispersing Wind Clearing Heat method and Sang Ju Yin prescription) and then to Purpose Layer (explaining etiology and mechanism). Inside the AI system, these contents are represented in clear layers, while when outputting to the patient, they are fused into natural language explanations and plans.

4.8 Summary

Through the above framework, we have realized a DIKWP*DIKWP Nested Path Model aiming at "Inspection, Listening/Smelling, Inquiry, Palpation + Chief Complaint + Conditioning Ideas." It includes both the vertical diagnostic reasoning chain (DIKWP chain from patient data to doctor decision) and the horizontal interaction mapping (correspondence of DIKWP levels between patient and doctor). Information obtained from each link of the four examinations is processed separately and integrated into the overall differentiation: Chief Complaint provides initial data, Inspection/Listening/Smelling/Palpation provide objective signs, and Inquiry fills in subjective symptoms and dynamic changes. Information from these different sources fuses in the Knowledge Layer to help AI identify syndromes; subsequently, conditioning plans are formulated in the Wisdom Layer; finally, the Purpose Layer ensures clear treatment goals consistent with patient needs. Throughout the process, every step of reasoning has a basis, and decisions at every layer can trace back to the basis in the layer below, achieving Layered Mapping and Semantic Unfolding of the diagnostic path. As shown in the next section, we can use Information Flow Charts and Semantic Unfolding Charts to visually demonstrate the operation details of this framework.

5. Nested Hierarchy Information Flow Chart and Semantic Unfolding Chart

To understand the application of the DIKWP model in the TCM diagnostic path more intuitively, we can draw corresponding Information Flow Charts and Semantic Unfolding Charts. Although graphics cannot be truly displayed in a pure text report, we describe the key points contained in these charts here.

5.1 Information Flow Chart: Hierarchical Transfer of the Diagnostic Process

The Information Flow Chart depicts the full link process from the patient providing information to the AI outputting diagnostic decisions, advancing step by step according to DIKWP levels, with arrows indicating the transfer relationship at each stage. Based on the previous case, this flowchart can be described as follows:

Data Inflow: Patient chief complaint and four examinations data enter the system (Input at Layer D in the chart). For example, input nodes include: "Chief Complaint Symptoms," "Tongue Coating Image," "Pulse Waveform," "Inquiry Answers," etc. Each input is annotated with its content, such as "Cough for 3 days, sore throat, fever" (Chief Complaint), "Red tongue, yellow coating" (Inspection), "Fear of cold turning to fever, little sweat" (Inquiry), etc.

Information Extraction: These data are processed by the Information Layer and converted into meaningful labels and features. Beside the D→I arrow in the flowchart, extraction rules or results can be written, such as "Fever + Sore Throat => External Heat Pathogen (I)," "Yellow Coating => Internal Heat (I)," "Little sweat and aversion to cold => Exterior Heat Syndrome (I)," etc. A group of information nodes gathers at Layer I of the chart, representing the patient's symptom patterns and pathogenesis clues.

Knowledge Matching: Information from Layer I is integrated into medical concepts via the Knowledge Layer. The I→K arrow marks knowledge application, for example, "Combine symptoms and tongue/pulse, match Wind-Heat invading Lung Syndrome (K)." Nodes at Layer K display possible syndrome diagnoses, such as "Wind-Heat invading Lung," or differential items like "Lung Heat Accumulation." If there are multiple candidate syndromes, multiple nodes can be paralleled at Layer K in the chart, followed by decision selection.

Wisdom Decision: The Layer K syndrome is sent to the Wisdom Layer for treatment decision formulation. The K→W arrow can mark the decision basis, such as "Based on Wind-Heat invading Lung Syndrome, determine treatment method Dispersing Wind and Clearing Heat (W)." Layer W nodes include specific "Conditioning Ideas" or "Treatment Plans." In the chart, this can be subdivided into "Treatment Method" nodes and "Prescription" nodes. For example, "Method: Disperse wind and clear heat, diffuse lung and stop cough," "Prescription: Modified Sang Ju Yin." If the AI performs formula modification, Layer W can also show it, such as which drugs were added to correspond to which symptoms.

Purpose Calibration: The Wisdom Layer plan is submitted to the Purpose Layer for review. The W→P arrow explains checking whether the plan meets the goal. Layer P nodes state the purpose, such as "Treatment Purpose: Clear and resolve exterior pathogen early, prevent pathogen from entering interior." If the plan meets the purpose, it passes; if not (e.g., patient has other demands like improving sleep), the AI might supplement corresponding measures in Layer W to align with the purpose. The Purpose Layer also guides explanation and communication; for example, an instruction "Explain diagnosis and plan to patient" might issue from Layer P in the chart.

Output: Finally, information is passed down from high levels to the patient side. The AI outputs the diagnostic conclusion and conditioning plan to the patient in natural language form. On the far right of the Information Flow Chart, the patient can be represented as another subject, where AI output is received as information on the patient side. For example, the doctor's Knowledge Layer conclusion "Wind-Heat invading Lung" becomes information received by the patient "Discomfort in lung caused by external wind-heat"; the doctor's Wisdom Layer plan becomes instructions heard by the patient "Take prescription to clear heat and resolve exterior." This corresponds to the process of Doctor K/W transforming into Patient I in the DIKWP×DIKWP model, using arrows of different colors to represent cross-subject semantic transmission.

Through such an Information Flow Chart, we can see the source and destination of every step of AI diagnostic decision-making at a glance: from patient data input to internal processing at each layer, and then to result output and feedback. Arrows between every layer can be traced and annotated, making the model's reasoning chain displayed transparently, solving the pain point where traditional AI models only show input and output without being able to understand the intermediate process.

5.2 Semantic Unfolding Chart: Hierarchical Breakdown of Professional Terms

The Semantic Unfolding Chart is used to demonstrate how a high-level medical concept is composed of lower-level semantic elements. In TCM diagnosis, this usually refers to breaking down Syndrome Names or TCM Terms into specific symptoms and pathogenesis components they contain. We take "Wind-Heat invading Lung" and "Lung Heat Accumulation" as examples to construct Semantic Unfolding Charts (Professor Duan's report gave similar tables and diagrams):

The center of the chart is the term itself (Knowledge Layer K): e.g., a round box writing "Wind-Heat invading Lung," another writing "Lung Heat Accumulation."

Unfolding downwards one layer are its Information Layer (I) elements:

For "Wind-Heat invading Lung," the information layer includes two key components: Wind-Heat Pathogen (External Heat Pathogen) and Lung Wei Affected (Pathogen invading Lung Exterior). These two elements correspond to the refinement of etiology and disease location respectively, represented by rectangles, with arrows pointing to the "Wind-Heat invading Lung" node above, indicating they combine to summarize this syndrome name.

For "Lung Heat Accumulation," information layer elements are Internal Excess Heat in Lung and Heat Pathogen Obstructing and Not Descending. The former describes disease location and nature: heat pathogen inside the lung; the latter describes disease trend/result: heat pathogen obstructing the lung, causing Lung Qi to fail in diffusing.

Further down, each information element can be further associated with more specific Data Layer (D) manifestations:

"Wind-Heat Pathogen" corresponds to clinical data of the patient suffering from external wind-heat signs, such as Heavy fever, slight aversion to cold, sore throat, etc.

"Lung Wei Affected" corresponds to data such as Cough, expectorating yellow thick phlegm, turbid nasal discharge, red tongue, floating rapid pulse, specific manifestations implying dysfunction of lung diffusion and descending, typical symptoms of exterior heat.

For "Lung Heat Accumulation," its information "Internal Excess Heat in Lung" corresponds to data including Persistent high fever, incessant coughing and panting, expectorating yellow thick phlegm, red tongue with yellow thick coating, slippery rapid pulse, manifestations of blazing internal heat; "Obstructing and Not Descending" corresponds to data manifestations like Cough and panting, abundant sticky phlegm, aggravated sore throat, symptoms of Lung Qi failing to diffuse.

Unfolding upwards, the Knowledge Layer term can be associated with Wisdom Layer (W) and Purpose Layer (P):

Although semantic unfolding usually focuses on breaking down lower layers, TCM concepts also contain implied treatment ideas. For example, for "Wind-Heat invading Lung," famous doctors seeing this syndrome often decide immediately to adopt the method of Dispersing Wind and Clearing Heat, Diffusing Lung and Stopping Cough. We can draw a small box above the Wind-Heat invading Lung node indicating "Therapeutic Principle: Disperse Wind and Clear Heat," representing the transition from Knowledge to Wisdom. Corresponding to "Lung Heat Accumulation," the principle is Clearing Lung and Draining Fire, Resolving Phlegm and Soothing Throat.

Further up in the Purpose Layer, the purpose of treating these syndromes can be noted. E.g., Wind-Heat invading Lung intends to clear and resolve exterior pathogen early to avoid entering the interior; Lung Heat Accumulation intends to clear lung and drain fire to avoid pathogen toxin damaging the lung. The Purpose Layer is usually not drawn directly in the term unfolding chart, but can be explained in text in the report.

Through such a Semantic Unfolding Chart, we can see: A profound TCM term actually contains semantics at multiple levels. Wind-Heat invading Lung is not just a name; it contains the fact that the patient received Wind-Heat pathogen (I) and the mechanism of lung dysfunction (K), and condenses a series of manifestations like fever, sore throat, coughing yellow phlegm (D). In teaching, we precisely hope AI can explain this complex concept to students by breaking it apart, rather than throwing out a term directly. Through the unfolding chart, students can clearly see the correspondence between terms and symptoms. For example, the chart marks "Wind-Heat invading Lung corresponds to Fever, Sore Throat, Coughing Yellow Phlegm, so Disperse Wind and Clear Heat"; "Lung Heat Accumulation corresponds to High Fever, Cough and Panting, Abundant Phlegm, so Clear Lung and Drain Fire." This translates obscure concepts into plain language. For the AI system itself, this unfolding means it stores a "component list" and "action chain" for each knowledge concept internally, thus enabling hierarchical unfolding during reasoning and explanation. For example, when the AI diagnoses Lung Heat Accumulation Syndrome, it can automatically retrieve the symptom set and treatment points associated with this concept, ensuring its advice covers all key points and does not miss the treatment of important symptoms.

In summary, the Semantic Unfolding Chart provides a tool to check and demonstrate AI semantic understanding: on one hand verifying whether the AI has grasped the core connotation of TCM concepts (e.g., did it miss any layer of meaning of "Wind" or "Heat" in "Wind-Heat invading Lung"); on the other hand, it is used to explain the AI's diagnosis to the user, establishing a connection between professional concepts and specific feelings. In the famous doctor diagnostic AI system we build, every diagnostic conclusion and every professional term can be unfolded into a small DIKWP semantic package using similar methods behind it. This guarantees the transparency of the system: giving both the conclusion and the layered reasons.

5.3 Verification Role of Diagrams on Models

Finally, it is worth mentioning that drawing these information flow and semantic unfolding charts is not only for display purposes. For developer and physician expert teams, it is also a process of verifying the correctness of the model. If a logic step in the chart is unreasonable (e.g., an arrow points in the wrong direction, or a necessary node is missing), it prompts us that there is a loophole in the model reasoning chain, and corresponding knowledge or rules need to be supplemented. Professor Yucong Duan's team emphasized detailed semantic level annotation and reasoning chain display for key concepts when building the AI TCM diagnosis and treatment system to make the internal reasoning of the model transparent. Such white-box analysis can discover similarities and differences between AI and human expert thinking, thereby continuously optimizing AI's cognitive ability. Our DIKWP Nested Path Model, through Information Flow Charts and Semantic Unfolding Charts, achieves Visual Representation of the Entire Diagnostic Process, making the reasoning process of AI "visible" from data to purpose.

6. Sample Data Annotation and Path Visualization

To further illustrate how the DIKWP model reconstructs famous doctor diagnostic thinking in specific cases, this section demonstrates word-by-word semantic annotation of data and visualization of the diagnostic path through a simplified Diagnosis and Treatment Dialogue Example.

Case Background: A young male patient. Symptoms: Cough for three days, sore throat with swelling, slight fever. The AI famous doctor inquires and diagnoses him. Below is an excerpt of the simulated doctor-patient dialogue, where every sentence is attached with DIKWP level semantic annotation (indicating which level the sentence or word belongs to in parentheses), and explains the internal cognitive state of the AI when hearing/saying this sentence:

Patient: "Doctor, I have been coughing for three days, my throat hurts a lot, and I have a bit of a fever." (Data D: Chief Complaint Symptoms: Cough 3 days, Sore throat, Fever)

Analysis: The patient provided the duration (three days) and three symptoms. These are unanalyzed raw data, directly reflecting the patient's subjective feelings. At this moment, the AI has not yet performed causal inference, and the semantics mainly stay at the data layer. Next, the AI will extract information from it to understand.

AI Doctor: "Coughing for three days, sore throat, fever..." (AI internally restates the patient's words and records them, listing points on paper or in memory) – Here AI is already performing Information Layer I organization, listing symptoms as facts to be analyzed, and noting that "three days" indicates the disease course is not long, belonging to new external disease. This step is equivalent to a human doctor writing the chief complaint in the medical record. (This self-talk record is not necessarily spoken out, but processed internally by AI).

AI Doctor: "Is there phlegm? What color is the phlegm?" (Wisdom W Layer behavior, with Purpose P: Guide patient to provide new Data D)

Analysis: Based on existing symptoms, the AI realizes the need to understand the phlegm situation to distinguish the nature of the pathogen (This is the Doctor's Diagnostic Purpose P: Identify Heat Pathogen or not). The question itself invokes TCM Knowledge K: "Yellow phlegm is mostly internal heat, white clear phlegm is mostly cold." Therefore, this question semantically belongs to Wisdom Layer W operation (using knowledge to solve problems), reflecting a clear goal. The patient can directly understand the literal meaning of the question (This is Information I on the patient side), but the medical theory behind the question is only known to the AI.

Patient: "There is phlegm, yellowish and sticky." (Providing new Data D)

Analysis: The patient answered two data points: having phlegm (qualitative data), phlegm color is yellow and quality is sticky (qualitative data). From the patient's perspective, this is just describing what is seen truthfully, still belonging to Data Layer content. But the AI immediately performed semantic interpretation on "Yellow and Sticky," considering it implies information of "Heat Sign." However, as far as the dialogue is concerned, "Yellow and Sticky" is still a perceptual description given by the patient, not directly saying "I have heat in my body." So in annotation, "Yellow and Sticky" is classified as Data D as the patient's words, but the AI internally views it as important Pathogenesis Clue Information I.

AI Doctor (Thinking for a moment then asks): "Are you afraid of cold these days? How is the sweating?" (Wisdom W Layer behavior, Purpose P: Further distinguish Cold/Heat attributes)

Analysis: The AI is inclined to judge as Wind-Heat at this moment, but needs final confirmation on the Cold/Heat characteristics of the Exterior Syndrome. Based on Knowledge K (Wind-Cold Syndrome: Aversion to cold, no sweat vs. Wind-Heat Syndrome: Fever, slight aversion to cold, sweating), the AI selected the question pair asking about "Aversion to Cold and Sweat" in Layer W. This logic is similar to asking about phlegm previously, also a question with a specific purpose. Therefore, annotated as W Layer, noting the intention P behind it (Distinguish Cold/Heat) and used Knowledge K (Wind-Cold/Wind-Heat differentiation experience). To the patient, it sounds like an ordinary question (Still I Layer on patient side: Symptom inquiry).

Patient: "I was a bit afraid of cold at first, now it's just fever, not much sweat." (Data D)

Analysis: The patient provided data regarding time evolution: afraid of cold at the beginning, now mainly fever, and not much sweating. These descriptions are crucial to the AI, belonging to Dynamic Data. The patient did not summarize the meaning, just stated facts, still classified as Layer D. The information refined by AI based on this is: "Pathogen entering from exterior to interior, transforming from Wind-Cold to Wind-Heat, currently in the stage of Exterior Heat" (External syndrome transforming into Heat Syndrome, pathogen still partially at exterior). But this inference belongs to the activity of AI's internal Knowledge Layer K, not explicitly stated in the dialogue.

AI Doctor: (Muttering while concentrating on pulse taking) "Throat pain and cough, phlegm yellow and sticky, slight aversion to cold and fever... (Recording these symptom clues)" (AI internal integration Data → Information)

Analysis: This scene simulates the process of a doctor organizing thoughts while feeling the pulse. The AI repeats the patient's main symptoms and signs, actually representing that it has internally integrated all the aforementioned Data D into Information I of "Wind-Heat Exterior Syndrome." The ellipsis after the self-talk actually contains the AI's subtext: "...This suggests Wind-Heat invading Lung." So although this sentence looks like restating data, it actually reflects the AI internally forming results from D to I and then to K. In annotation, we can view this self-talk as an "Internal Prompt" of the AI, noting in parentheses "(Here AI has integrated the content of the previous sentence into clue information I for Wind-Heat invading Lung Syndrome, and began to form a judgment in Knowledge Layer K)". This reflects the conversion of internal data to information in the AI assistant: the doctor begins to form a preliminary impression of the syndrome.

Patient: "Yes, sore throat with swelling, it hurts even when swallowing saliva." (Supplementing Data D)

Analysis: The patient confirmed and supplemented the severity of the sore throat. This response provides new data again (pain obvious on swallowing). For the AI, this strengthens the information of vigorous heat toxin in the Lung Meridian, but at this time the general situation is set, this data will not change the diagnosis, only affecting the treatment plan (suggesting need for heat-clearing and detoxifying drugs). Semantically still marked as D Layer.

AI Doctor: "Hmm, tonsils are inflamed. Is there any pus point?" (Wisdom W Layer question, Purpose P: Understand if the condition has developed to the suppuration stage)

Analysis: This did not appear in the previous text. We assume the AI asks about the tonsil situation to judge the severity of inflammation (judge if there is heat toxin). Still a W Layer question with intent. The patient might answer whether there are pus points, which is also data.

Patient: "Looking in the mirror, there seem to be some light yellow points." (Data D: Light yellow points on tonsils)

Analysis: Patient describes object resembling light yellow points on tonsils. This data indicates slight suppuration. AI interprets as heat toxin not severe, moderate inflammation.

AI Doctor: "Okay. Let me check the tongue coating and pulse." (Preparing to acquire more Data D: Inspection of Tongue and Palpation of Pulse)

Analysis: This sentence explains the AI is about to perform the final tongue inspection and pulse taking check. Semantically, this can be viewed as the AI's action description, not content of doctor-patient dialogue. But it can be annotated as W Layer behavior (Executing diagnostic operation).

(After viewing tongue and feeling pulse)

AI Internal Data Acquisition: Tongue body red, coating thin yellow, pulse floating rapid forceful. (Data D)

AI Analysis: Tongue and pulse findings suggest Wind-Heat at the exterior, pathogen not deep.

Patient: "What does it mean, doctor?" (Patient puzzled, seeking Information I)

Analysis: Seeing the doctor pondering silently after checking, the patient might ask questions. This indicates the patient's information need at this moment, i.e., hoping the doctor tells him the diagnostic result in language the patient can understand (Layer I).

AI Doctor: "It is just that you caught Wind-Heat pathogen, causing a cold and sore throat. The heat pathogen invaded the Lung Meridian, so you cough." (Explaining to patient: Doctor K → Patient I)

Analysis: This sentence is the Diagnostic Explanation given by the AI, describing the etiology and mechanism in plain language, basically equivalent to telling the patient: "You have Wind-Heat invading Lung Syndrome." Words like "Wind-Heat Pathogen" and "Lung Meridian" still carry TCM color, but the patient can roughly understand as "You have a cold with lung heat caused by external wind-heat." Semantically, this is the doctor's side Knowledge K expression transformed into Patient side Information I. When annotating, note "(Doctor translates professional diagnosis into layman explanation, patient obtains Information I)". It is worth noting that each phrase here actually corresponds to elements in our semantic unfolding chart earlier: Wind-Heat Pathogen, Lung, a bit of inflammation/disorder = Fever Cough. The AI is precisely following the DIKWP layer-by-layer translation method to break down knowledge layer concepts for the patient.

Patient: "I understand. So how to treat it?" (Patient asks for next step, seeking Plan Information I)

Analysis: The patient accepted the doctor's explanation (agreed with Information), now asks for treatment plan. This suggests the AI needs to output Wisdom Layer W and Purpose P content in the form of information to let the patient understand.

AI Doctor: "For treatment, we Disperse Wind and Clear Heat, Diffuse Lung and Stop Cough. (Treatment Method W) I will first write a prescription called Sang Ju Yin (Prescription K), mainly composed of Mulberry Leaf, Chrysanthemum, Forsythia, Almond, Platycodon, etc. (Prescription Explanation K->I) This prescription can help you clear heat and resolve the exterior, and relieve cough. (Purpose P)"

Analysis: In this paragraph, the Bold part "Disperse Wind and Clear Heat, Diffuse Lung and Stop Cough" is a typical TCM therapeutic method term, belonging to the treatment strategy formulated by the doctor in Wisdom Layer W, and can also be viewed as Knowledge-ized content (TCM professional vocabulary). But immediately the doctor mentions specific formula "Sang Ju Yin," which is a Knowledge Layer K entity (Classic Formula). The doctor then lists the drug composition in the formula and explains efficacy, which is breaking down formula knowledge into information to tell the patient, letting the patient understand these drugs are used for clearing heat and resolving exterior. The last sentence summarizing efficacy "Clear heat resolve exterior, stop cough" points out the Treatment Purpose P (Clear heat pathogen, relieve symptoms) and the effect the patient cares about. These sentences convey contents of Wisdom Layer and Purpose Layer to the patient, equivalent to a complete account of the conditioning idea. Annotation can be subdivided: Treatment Method belongs to W Layer, Formula Name belongs to K Layer (but is a new noun for patient, equivalent to Information I, needing explanation), Drug List is breaking down K Layer formula into D Layer drug composition, Efficacy Explanation belongs to information derived from P Layer, etc. This answer fully reflects the application of the DIKWP model: AI not only gives the plan but also explains the principle and purpose of the plan, making the patient clear "Why treat like this" and "What will happen after treatment."

Patient: "Okay, thank you doctor!" (Patient accepts plan)

From the above sentence-by-sentence analysis, it can be seen that in the dialogue, every understanding and response of the AI assistant can be annotated and explained using DIKWP levels: which words are Data D provided by the patient, which sentences contain Knowledge K applied by the doctor, and which behaviors stem from the doctor's Wisdom Decision W and Purpose P. Professor Yucong Duan's research has similarly performed word-by-word semantic analysis and state evolution demonstration on an AI doctor inquiry dialogue, showing how AI extracts data and information from patient's words, ascends to knowledge, and then forms a closed loop of wisdom decision. For example, the report tags every word of the patient (D/I/K/W/P) and maps the semantic network of AI understanding and response. Through such word-by-word annotation, we verified the consistency of the AI's reasoning chain with the human doctor's thinking: Every question and answer conforms to TCM theory, and every step of reasoning has a source. At the same time, these annotations also make the visualization of the diagnostic path possible—we can almost draw a flowchart or network graph based on annotations, mapping the dialogue into a path of semantic level transformation. If the patient's words are viewed as input nodes, AI's decisions as intermediate nodes, and AI's explanation plans as output nodes, then this graph clearly describes the Vein of AI diagnostic thinking: Patient Symptoms (D) → Pathogenesis Information Extraction (I) → Syndrome Identification (K) → Questioning for New Data (D) → ... (Loop) ... → Syndrome Establishment (K) → Therapeutic Method Prescription Decision (W) → Treatment Purpose (P) → Output Explanation (I). This is exactly the complete reconstruction of "Famous Doctor Diagnostic Cognitive Path" we pursue.

Through this instance, we verified the effectiveness of DIKWP Nested Path Modeling in TCM diagnostic scenarios. AI can, like a famous doctor, perform semantic classification annotation on every sentence and every symptom of the patient, understanding its meaning level; meanwhile, AI's own every question and every decision can also explain the motivation using semantic levels. This Word-Level Fine-Grained Annotation and Visualization allows the AI's diagnostic process to reach unprecedented transparency. For developers, they can check if every annotation is correct to adjust model understanding. For medical professionals, they can evaluate if AI's thinking matches real famous doctors through these annotation results. For patients and the public, these annotations provide a window to understand AI diagnosis—it is no longer a mysterious black box but a logical process that can be broken down and explained step by step.

7. Program Design Pseudo-code and Reasoning Module Logic Diagram

In the process of implementing the above DIKWP diagnostic model, we need to transform the theoretical framework into a concrete software system. Below we provide Pseudo-code and Reasoning Module Logic Diagram to describe the workflow of the AI famous doctor diagnostic system. This will help technical implementation personnel understand how to embed the DIKWP multi-layer reasoning mechanism into the program.

7.1 Diagnostic Process Pseudo-code

The following pseudo-code describes the main flow of AI famous doctor diagnosis in a Python-like style, with each step corresponding to DIKWP level processing:

function TCM_Diagnosis_AI(input_data):

    # Input: input_data contains patient basic info and chief complaint

    # Initialize empty containers to store information of each level

    Data_layer = {}      # Data Layer Dictionary

    Info_layer = {}      # Information Layer Dictionary

    Knowledge_layer = {} # Knowledge Layer Candidates

    Wisdom_layer = {}    # Wisdom Layer Decision Plans

    Intent_layer = {}    # Purpose Layer Goals

    # 1. Data Layer acquires chief complaint and basic information

    Data_layer = input_data  # E.g., "Cough for three days, sore throat, fever"

    Data_layer = input_data    # Such as age, gender, etc.

    # Extract preliminary symptom information from chief complaint

    Info_layer.update( extract_symptom_info(Data_layer) )

    # E.g., extracted {"Cough": 3 days, "Sore Throat": severe, "Fever": low grade}

    # 2. Four Examinations Information Collection

    # 2.1 Inspection

    vision_data = perform_visual_diagnosis()   # Acquire tongue image, complexion image, etc.

    Data_layer = vision_data

    Info_layer.update( analyze_visual_signs(vision_data) )

    # Such as: {"Tongue Body": Red, "Tongue Coating": Yellow Greasy, "Complexion": Ruddy}

    # 2.2 Listening/Smelling

    audio_data, smell_data = perform_listen_smell()  # Acquire voice, odor

    Data_layer = audio_data

    Data_layer = smell_data

    Info_layer.update( analyze_audio(audio_data) )

    Info_layer.update( analyze_odor(smell_data) )

    # Such as: {"Cough Sound": Loud, "Breath": Foul odor}

    # 2.3 Dynamic Inquiry Dialogue

    while True:

        question = plan_next_question(Info_layer, Knowledge_layer)

        if question is None:

            break  # Break if no questions needed

        purpose = identify_question_intent(question)  # Identify Question Intent (P)

        print("AI Doctor asks:", question, "(Purpose:", purpose, ")")

        answer = get_patient_answer(question)        # Get patient answer

        print("Patient answers:", answer)

        # Treat answer as new data

        Data_layer = answer

        # Extract valid information from answer and update Information Layer

        Info_layer.update( extract_symptom_info(answer) )

        # Update possible syndrome judgments in Knowledge Layer

        Knowledge_layer = match_possible_patterns(Info_layer)

        # If there are new uncertain syndromes, loop to ask more questions to clarify

    # Loop ends, inquiry complete

    # 2.4 Palpation

    pulse_data = measure_pulse()      # Acquire pulse data

    palpation_data = perform_palpation()  # Other touch diagnosis data

    Data_layer = pulse_data

    Data_layer = palpation_data

    Info_layer.update( analyze_pulse(pulse_data) )

    Info_layer.update( analyze_palpation(palpation_data) )

    # Such as: {"Pulse": Floating Rapid, "Tenderness": Tonsil Enlargement}

    # 3. Knowledge Layer: Comprehensive Differentiation Analysis

    Knowledge_layer = match_possible_patterns(Info_layer)

    # Match the most fitting syndrome based on all information

    primary_pattern = select_best_pattern(Knowledge_layer)

    # E.g., primary_pattern = "Wind-Heat invading Lung"

    alt_patterns = get_secondary_patterns(Knowledge_layer, exclude=primary_pattern)

    # E.g., alt_patterns = if backup syndromes exist

    # 4. Wisdom Layer: Formulate Treatment Plan

    # Retrieve standard treatment method and formula for the syndrome from knowledge base

    base_plan = retrieve_base_plan(primary_pattern)

    # E.g., base_plan = {"Method": "Disperse Wind Clear Heat, Diffuse Lung Stop Cough", "Recommended Formula": "Sang Ju Yin"}

    # If there are concurrent syndromes or secondary syndromes, consider combined treatment

    for pat in alt_patterns:

        base_plan = adjust_plan_for_pattern(base_plan, pat)

    # Adjust plan according to patient specific situation

    Wisdom_layer = modify_prescription(base_plan, Info_layer, patient=Data_layer)

    Wisdom_layer = base_plan

    # 5. Purpose Layer: Set treatment intent and explanation strategy

    Intent_layer = set_clinical_goal(primary_pattern, Info_layer)

    # E.g., Intent_layer = "Clear and resolve exterior pathogen, prevent entering interior"

    Intent_layer = "Explain diagnosis and plan in language patient can understand"

    # 6. Output Results and Closed Loop

    # Output diagnostic results and treatment plan (with explanation)

    diagnosis_result = primary_pattern  # TCM Syndrome Name

    explanation = generate_explanation(primary_pattern, Info_layer, Intent_layer)

    treatment_plan = Wisdom_layer

    advice = generate_followup_advice(primary_pattern)

    print("AI Diagnosis Result:", diagnosis_result)

    print("Explanation:", explanation)

    print("Proposed Method:", Wisdom_layer, "; Prescription:", treatment_plan)

    print("Conditioning Advice:", advice)

    # Follow-up can be simulated later, feeding feedback back as input, closed-loop adjustment (omitted here)

    return {"Syndrome": diagnosis_result, "Prescription": treatment_plan, "Explanation": explanation}

The pseudo-code flow corresponds to the behavior of each level of the DIKWP model:

Steps 1-2: Collect patient data and extract information (D→I). Includes data acquisition functions for Chief Complaint and Four Examinations, and corresponding information extraction functions. [Functions like extract_symptom_info, analyze_visual_signs in the code simulate conversion from Data to Information]

Step 2.3 While Loop: Dynamic inquiry, here AI decides the next question (Wisdom Layer decision) and questioning intent (Purpose Layer guidance) based on existing Information Layer and Knowledge Layer status. Patient answers update Data Layer and Information Layer, subsequently refreshing Knowledge Layer judgment on syndromes. Loop until no new questions need to be asked (meaning Knowledge Layer uncertainty reduces to acceptable level).

Step 3: Comprehensive Differentiation, calling match_possible_patterns to match Information Layer with syndrome patterns in knowledge base, finding possible syndrome lists, and deciding the most fitting one as primary diagnosis via select_best_pattern. Optionally retain secondary suspicious syndromes for subsequent concurrent syndrome processing.

Step 4: Formulate treatment plan. Obtain basic plan (Treatment Method + Representative Formula) via primary syndrome from knowledge base. If there are concurrent conditions, adjust plan via adjust_plan_for_pattern (e.g., combining formulas or adding drugs if Lung Heat Accumulation co-exists). Then call modify_prescription to refine prescription according to patient details (e.g., dosage, addition/subtraction). Store final prescription and method in Wisdom Layer.

Step 5: Determine Purpose Layer content, including setting clinical treatment goals (e.g., focus on expelling pathogen/supporting righteous) and communication explanation strategies. The latter determines the generation of explanation language for the patient.

Step 6: Output results, simultaneously forming doctor-patient communication content. AI prints/feeds back Syndrome Diagnosis Result, Treatment Plan, and Explanatory Note to user. Here generate_explanation function will utilize previously prepared intent and information to explain diagnosis in plain words. generate_followup_advice can give some conditioning suggestions and follow-up prompts. Finally return structured results for system recording or downstream processing.

From the pseudo-code, it can be seen that the DIKWP framework equips the AI diagnostic program with Clear Module Division: Data Acquisition Module, Information Processing Module, Knowledge Reasoning Module, Decision Generation Module, Purpose Calibration Module, etc. Each module corresponds to a different level of the cognitive process. This design makes the system more interpretable and maintainable—if the diagnostic result is wrong, we can trace whether it is a data extraction error (D→I problem), knowledge matching deviation (I→K problem), or improper decision (K→W problem), and improve targetedly. This is much more transparent than traditional end-to-end black box models.

7.2 Reasoning Module Logic Diagram

Besides pseudo-code, we can draw a schematic diagram of system architecture and reasoning logic. In the diagram, main modules are represented by boxes, data/control flows between modules are represented by arrows, and key contents are annotated:

Input Layer Module: Responsible for collecting patient input, including Natural Language Processing Unit (converting patient text/voice to structured symptoms) and Sensor Interfaces (processing image, audio, pulse signals). Output enters "Data Buffer."

Data Layer Processing Module: Reads raw data from Data Buffer, calls multiple parsers to extract symptoms and features. Parsers include: Symptom Extraction NLP, Tongue Image Analysis CNN, Pulse Analysis Algorithms, etc. Results stored in "Information Library."

Information/Knowledge Layer Reasoning Module: The core reasoning engine. Composed of Knowledge Graph and Reasoning Machine. Knowledge Graph nodes include symptoms, syndromes, methods, formulas, etc. Reasoning Machine can be a rule engine or knowledge graph query system. Its logic: Based on activation of symptom nodes in Information Library, search matching syndrome nodes in graph (can be obtained by calculating match degree of symptom-syndrome association edge weights). Send match results to Decision Unit. If multiple syndromes match, Decision Unit selects primary syndrome based on priority rules (or heuristic scoring), while recording secondary possible syndromes. [On logic diagram, this part can be drawn as a Syndrome Decision Tree or Multi-Candidate Voting Mechanism icon].

Dialogue Management Module (Wisdom Layer Interaction Sub-module): This is the control module for the inquiry process. It decides whether more information is needed based on current Knowledge Layer uncertainty, and generates questions. When patient answer enters Data Layer, it notifies Reasoning Module to update judgment. Strategies in Dialogue Management Module are implemented by Diagnostic Decision Tree or Reinforcement Learning Agent, following famous doctor questioning logic (previously mentioned inquiry relationship network in Knowledge Graph can also be used here). In the logic diagram, a loop arrow can be drawn from Dialogue Module to Patient Input, representing dynamic interaction feedback.

Wisdom Layer Decision Module: After determining syndrome, enters treatment plan generation. This module calls Treatment Knowledge Base, including treatment methods and classical formulas for each syndrome, and famous doctor experience prescription library. Logic: Get Base Formula -> Adjust based on Secondary Syndromes and Patient Condition -> Output Proposed Prescription. A flowchart can represent the process of "Syndrome -> Method -> Formula -> Addition/Subtraction."

Purpose Layer Module: Includes two sub-functions: 1) Goal Setting: Derive clinical intent from diagnostic conclusion, e.g., need to resolve exterior or attack down, how long to see effect, etc.; 2) Explanation Generation: Call Semantic Conversion Unit, convert professional nouns into expressions easy for patients to understand. Semantic conversion can use aforementioned term semantic breakdown rules (e.g., Wind-Heat invading Lung = Cough and Fever caused by external wind-heat attacking lung...). On logic diagram, this module receives results from Knowledge/Wisdom Layers, outputs to next Communication Module.

Output Communication Module: Assemble diagnostic results, prescription, and explanation into report or dialogue answer to send to user. This part can use Natural Language Generation technology, outputting according to certain templates or fine-tuned generation models. But since we have rich structured information, generated content accuracy and professionalism can be controlled.

Follow-up Feedback Module (Optional): Patient feeds back efficacy data after treatment, again processed via Data->Information, entering Knowledge Graph for comparison. If efficacy is poor, might indicate initial diagnosis differentiation was incomplete, triggering adjustment. This closed-loop feedback is shown as a dashed arrow from Output back to Input in the logic diagram, representing Active Learning mechanism.

The above modules are arranged from bottom to top according to DIKWP hierarchy in the diagram, connected by arrows in between. Two loops are particularly significant: Small Loop is data-information-knowledge repeated iteration during inquiry interaction (continue asking if information insufficient); Big Loop is feedback re-closed loop after treatment (realizing active learning, continuously optimizing model).

The final output end of the logic diagram can also interface with hospital HIS systems or electronic medical records, providing AI suggestions for human doctors to reference, or directly generating standardized electronic case content. Such integration already has prototypes in some applications, like Ping An Good Doctor's One-Minute Clinic, which can structuredly transmit information collected by AI to backend doctors, reducing communication costs.

Through pseudo-code and logic diagram, we have clearly expressed the operating mechanism of the DIKWP TCM Diagnostic AI System. For development, key modules (NLP, Knowledge Graph, Decision Tree, NLG, etc.) all correspond to mature technologies, which can be implemented separately and interact through interfaces, conforming to High Cohesion Low Coupling principles of software engineering. The DIKWP framework provides the "Contract" and "Language" for these modules' collaboration, i.e., the content and form of information transmitted by each layer, making the whole system operate in good order and easy to debug. More importantly, this architecture naturally possesses Explainability: developers or medical experts can monitor and debug at the output of every module, for example checking which syndrome matched in Knowledge Layer, what formula chosen in Wisdom Layer, how interpreted in Purpose Layer. Compared to an end-to-end neural network, it is obviously much more transparent, controllable, and reliable.

8. Comparative Analysis with Ali Health and Ping An Good Doctor TCM AI Systems

Currently, there are multiple TCM AI inquiry products in practice in the industry, such as Ping An Good Doctor's Intelligent TCM System and Ali Health (Deer TCM)'s AI Diagnosis and Treatment Assistant. They have similarities with the DIKWP model framework proposed in this paper but also have core differences. Below we conduct a comparative analysis of the DIKWP model and the above actual systems from two aspects: Diagnostic Path Structure and Explainability.

8.1 Diagnostic Path Structure Comparison:

Ping An Good Doctor TCM AI (Modern Hua Tuo System):

Ping An Good Doctor launched the landmark "Modern Hua Tuo Plan" around 2018. Its core is to introduce AI technology into the Four Examinations (Inspection, Listening/Smelling, Inquiry, Palpation) to realize the standardization and digitization of the four examinations process. Specifically, they developed four modules: "Intelligent Tongue Diagnosis," "Intelligent Listening/Smelling," "Intelligent Inquiry," and "Intelligent Pulse Diagnosis": analyzing tongue image via tongue surface image, recognizing constitution breath via sound collection analysis, inquiring symptoms via inquiry robot, and acquiring pulse image via electronic pulse diagnostic instrument. These modules correspond to data collection of the four examinations respectively and output digitized results. For example, Intelligent Tongue Diagnosis gives structured description of tongue color and coating; Intelligent Listening/Smelling maps user voice to TCM constitution categories like Qi Stagnation, Qi Deficiency, etc. Then, these data undergo Comprehensive Analysis by AI, pushed to backend TCM doctors, and generate TCM Prescription Recommendations for doctor's reference. It can be seen that Ping An system's diagnostic path is modularized according to traditional four examinations order, each module outputs standardized parameters, then summarized to get conclusion, equivalent to an assembly line structure. Its advantage lies in realizing Objective Quantification and Standardized Collection of Four Examinations Information, improving consistency and efficiency of diagnosis. Especially after introducing hardware in links like Tongue Diagnosis and Pulse Diagnosis which lacked quantification means in the past, human error is reduced, beneficial for data accumulation and model training.

However, the Ping An system is relatively weak in Reasoning Chain Presentation. It emphasizes front-end data standardization more, and details on how backend "Comprehensive Analysis" works are not fully disclosed. It is speculated that behind it, there may also be differentiation of syndrome types based on certain rules or models, and then prescription recommendation. For instance, news reports mentioned it can give personalized health preservation suggestions and list "16 conditioning plans, each with sources from ancient books and famous doctors." This indicates it has a built-in knowledge base matching certain symptoms/constitutions to classic famous formulas and famous doctor experiences, and citing these sources to enhance credibility. E.g., "Coordinate with a certain 16 conditioning plans, and all have sources from ancient books and famous doctors." This provides explanation to a certain extent—justifying the recommended plan by citing authoritative literature. But this belongs to Corroboration of Result, not explanation of the process. So in terms of diagnostic path structure, Ping An Good Doctor's system tends towards an architecture of "Perception Front-loaded, Decision Back-loaded," where data collected by various diagnostic methods enter a comprehensive analysis engine like a black box, finally outputting a plan. The intermediate differentiation reasoning is not visible to users and doctors. Therefore, although its diagnostic path fully implements the inspection, listening, smelling, inquiry, and palpation process, the Reasoning Chain remains Opaque (belonging to product trade secret or complex model, invisible at user end). In contrast, the path structure of the DIKWP model, based on realizing dataization of four examinations, emphasizes step-by-step mapping of five layers of cognition more, even allowing users/doctors to view intermediate reasoning stage results (such as corresponding syndrome points, reasoning scores, etc.). In other words, the Ping An system is more like an AI Assistant Tool, accelerating pre-diagnosis information collection through standardized input, while the DIKWP system is more like an AI White-box Doctor, demonstrating its own thinking process while providing results.

Ali Health Deer TCM AI Diagnosis and Treatment System:

Relying on DAMO Academy and the acquired Deer TCM team, Ali Health launched the TCM AI Assisted Diagnosis and Treatment System. Its feature lies in integrating Large Models (e.g., DeepSeek) with Knowledge Graph/Retrieval-Augmented Generation (RAG) technology. According to reports, Deer TCM AI, through Pre-trained Large Model + TCM Knowledge Fine-tuning, can achieve high efficiency improvement in scenarios like doctor assisted inquiry. Specifically, they built an exclusive TCM Knowledge Graph, including relationship networks of Chinese medicines, formulas, syndromes, and trained the model by inputting million-level TCM case data. On the diagnostic path, the Ali system seems closer to the process of traditional human doctors. One example is: In a cooperation project with China Academy of Chinese Medical Sciences, they let doctors dictate four examinations information, AI converts dialogue into structured TCM disease/syndrome, chief complaint symptoms, etc. via large model and NLP, and then performs reasoning diagnosis to give AI recommended prescriptions. That is to say, it lets AI "understand" the doctor's inquiry process record and come up with a diagnosis and prescription suggestion. Here the large model (like DeepSeek) plays a big role; it possesses the ability to understand natural language and certain reasoning capabilities. DeepSeek was once tested to prescribe medicine directly based on prompts, and the results were quite "to the point," evaluated by senior TCM doctors as "able to treat the cause, with obvious efficacy." This shows that the large model itself has stored a lot of TCM knowledge and can perform syndrome differentiation and treatment to some extent. In addition, doctors also reflected that in actual use, they would use AI to help analyze disease causes and treatment plans for complex cases to broaden their thinking. This implies that Ali system's diagnostic path allows doctors to freely communicate cases with AI, and AI provides analytical suggestions, the form of which might not be fixed four examinations module order like Ping An, but Flexible Human-Machine Synergy. In general, Ali's system architecture is based on Large Language Model (LLM), so the diagnostic path has end-to-end characteristics: from condition description input to diagnosis and plan output, the middle relies on model internal parameters and knowledge. To improve accuracy, it introduces Knowledge Graph Retrieval Augmentation, enabling the model to query authoritative clinical guidelines or famous doctor cases, thereby reducing the risk of model generating nonsense. This point has similarities with the DIKWP framework idea, both emphasizing Fusing External Knowledge to constrain AI decision. However, the difference is that the Ali system mainly uses LLM's powerful generation and understanding capabilities to connect the diagnostic path, and its reasoning chain is mostly implicit inside the LLM; while the DIKWP system breaks down the reasoning chain into explicit five layers, LLM can be used in a certain layer (e.g., explanation generation) but does not replace the whole process. Therefore, structurally, the Ali system reflects "Large Model + Knowledge," while the DIKWP system is "Small Model Combination under Knowledge Framework." Both have pros and cons: Large Model solution is fast to develop, dialogue is natural, but capturing style differences of different schools of famous doctors requires extra fine-tuning; Knowledge Framework solution requires carefully building knowledge and rules, but once built, can be precisely controlled, and one knowledge base per famous doctor can realize style customization.

8.2 Explainability Comparison:

Transparency of DIKWP Model:

As mentioned earlier, the DIKWP model makes AI diagnosis possess high explainability through modeling the white-box cognitive chain. The entire reasoning process is mapped to five semantic levels, and the output of each stage can be manually checked and understood. In clinical use, this means AI can explain to doctors or patients: "I collected what data (D), derived what symptom information (I), associated with what syndrome knowledge (K) based on this, decided to adopt what treatment plan (W), to achieve what goal (P)." For example, when AI recommends a certain prescription, it can point out: "This plan references the treatment method for XX disease in 'National Clinical Guidelines' (Knowledge Basis), uses such and such classic formula with modifications (Wisdom Decision), aiming to quickly relieve patient chief complaint symptoms and treat the root (Purpose)." It can even give confidence for every step. For instance, Professor Duan's team proposed using DIKWP Difference Graph to annotate differences between different model outputs, thereby improving interaction transparency. This is equally applicable in medicine: If AI disagrees with human doctor, respective DIKWP chains can be compared to find where the divergence lies in which layer—maybe different interpretation of a symptom (Layer I difference), or different knowledge selected (Layer K difference), etc. This fine-grained white-box evaluation is a unique advantage of the DIKWP model. And the improvement of explainability directly brings improvement of Credibility and Controllability. Doctors are more willing to trust an AI that can state its reasoning process, and regulators can more easily audit whether its decision basis conforms to norms. In summary, under the DIKWP framework, the Transparency and Explainability of the AI diagnostic path are high, which is a major improvement over the traditional black-box mode.

Explainability of Ping An Good Doctor System:

The Ping An system focuses on using AI assistance to improve efficiency, but in terms of explainability, public information is limited. It can be speculated that on the patient side, they provide direct constitution reports, nursing suggestions, and preliminary diagnoses, etc., without explaining in depth how the model reached these results. For example, after a patient finishes an AI inquiry, they might receive a "TCM Constitution Identification Report" stating constitution type and nursing plan. These results are relatively direct and easy to understand for patients, but if asking "Why did AI judge me as this constitution," the system may not be able to explain in detail (unless the report attaches various indicators). For doctors, Ping An AI mainly plays the role of collecting information and recommending prescriptions. Recommended prescriptions come with ancient books and famous doctor sources, which is a way of explanation, indicating "AI's suggestion has basis." This basis citation is common in clinical decision support systems and can increase trust to some extent. However, it cannot dynamically answer deeper whys—such as why choose this formula instead of that one, how AI weighed symptoms internally to arrive at this syndrome type, etc. If the doctor needs further information, they may only be able to refer to the structured medical history data provided by AI and then judge by themselves. By comparison, if using the DIKWP system, the doctor can query AI's reasoning at every step, such as "What are the basis symptoms for AI identifying Wind-Heat invading Lung," or "Why did AI ask a certain question"—because all have label annotations. Therefore, Ping An system's explainability belongs to Static Literature Corroboration Type, while DIKWP system's explainability belongs to Dynamic Process Transparent Type, the latter obviously being richer and more flexible.

Explainability of Ali Deer TCM System:

For Large Model based solutions, explainability depends on the form of model output. If designed properly, LLM can be prompted to output Thinking Process, or even imitate TCM doctor "narrating" diagnostic ideas. For example, DeepSeek gave formula and prescribing ideas simultaneously during media testing. This indicates that through prompt engineering, LLM can include in the answer: "Differentiated as certain syndrome, treatment method is what, use formula so-and-so, reason is...". This looks like providing explanation, but notice whether this explanation is the reproduction of the model's real reasoning? Likely not, but it Reproduces a Reasonable Explanation based on its own knowledge. LLM itself is still a black box; the ideas it speaks may not be the trajectory it really reasoned step by step when deriving the formula. This involves the issue of "Aligned but Not True," i.e., LLM can learn to say explanations that satisfy us, but inside it might be hallucinating fabrication. To reduce this unreliability, Ali's solution adopts "Mixture of Experts Architecture," e.g., accessing 52 TCM dominant disease clinical guidelines into model retrieval to enhance answer reliability. Thus LLM cites authoritative materials more when answering, reducing nonsense. However, LLM may still Produce Hallucinations, and the industry recognizes that "hallucination rate can be reduced but difficult to eliminate." The DIKWP model, due to its logic chain based on symbols and knowledge networks, will almost not produce baseless content out of thin air; every step is verifiable, naturally eliminating baseless hallucinations. This is the reliability improvement brought by Combination of Symbolism and Connectionism.

In addition, the Ali system emphasizes personalized adaptation to famous doctors, e.g., fine-tuning each famous doctor's diagnostic features via LoRA. This means they can train a model avatar specifically imitating a famous doctor's answering style and medication habits. But the explanation output by this method is still speech patterns learned based on samples. In contrast, if DIKWP builds exclusive knowledge graph and rule engine for each famous doctor, then the explanation given by AI is actually reproducing the famous doctor's thinking logic back then (because the knowledge graph may come from famous doctor's works and medical cases, encoded by experts). For example, if a famous doctor often says "Treatment should protect Spleen and Stomach," AI will add this clause in the explanation, because there is this experience in the famous doctor knowledge base. It can be seen that the DIKWP model can reflect famous doctor thoughts in a more Controllable and Real way, rather than just imitating their language style.

8.3 Comparison Summary:

Ping An Good Doctor's system has advantages in Standardized Collection and Completeness of Process; Ali Health's system performs prominently in Data Volume Drive and Intelligent Dialogue; while the DIKWP Model's system has unique advantages in Reasoning Transparency and Personalization Depth. More specifically:

Diagnostic Path Structure: Ping An is modular linear flow, Ali is large model coordinated flow, DIKWP is multi-layer nested flow. DIKWP flow is relatively complex but can meticulously depict cognitive process. If combined with Ping An and Ali, they might complement each other. E.g., first use Ping An hardware to collect data, then use DIKWP reasoning chain to analyze, use Ali large model to generate natural dialogue, this might be future trend.

Explainability: DIKWP provides White-box level explainability, able to explain "How it thought"; Ping An system mainly Cites Basis, explaining "Based on what classic" to increase credibility; Ali system relies on LLM Generating Explanation, explaining "Speaking like a famous doctor." In terms of credibility, White-box > Basis Citation > Generated Explanation. DIKWP belongs to the first kind, thus having greatest potential in enhancing patient and doctor trust.

Famous Doctor Personalization: Ping An system currently does not emphasize famous doctor school differences, relatively standardized; Ali realizes famous doctor personality simulation via fine-tuning models; DIKWP reflects famous doctor schools via customized knowledge graphs and rules. The latter two can both achieve "Famous Doctor Avatar" to some extent. DIKWP's advantage lies in being able to Explicitly Maintain Knowledge Bases of Different Famous Doctors, avoiding style crosstalk; Ali model if using one large model to learn multiple famous doctors might confuse, needing strict isolation of training data or one model per famous doctor.

Standardization and Scalability: DIKWP model is very suitable for combining with TCM Standardization work. Because it requires making knowledge explicit into structure, which exactly promotes TCM knowledge standardized coding, such as coding syndromes and formulas into knowledge graph. Ali and Ping An also realized importance of standardization, just different technical routes: one starts from data "Structure first then Reason," one starts from model "Reason first then Structure/Refine Result." DIKWP belongs to the former (Symbolism route), relatively conforming to the demand of building TCM standard system in scientific research. For example, Professor Yucong Duan has reports specifically discussing how to build TCM standardization system based on DIKWP. It is foreseeable that in the future, evaluating a TCM AI will not only look at accuracy but also likely look at its DIKWP layer performance, i.e., so-called White-box Evaluation. DIKWP framework exactly provides such an evaluation standard and grip.

In summary, our DIKWP TCM Diagnostic AI System reflects innovation in diagnostic path and explainability, compared with existing systems like Ping An Good Doctor and Ali Deer TCM, it can bring higher transparency and personalization level for TCM AI. Below enters summary, generalizing the value of DIKWP model in improving TCM AI diagnosis transparency, personalization, and standardization.

9. Summary: Application Value of DIKWP Model in TCM AI Diagnosis Transparency, Personalization, and Standardization

This paper has discussed in depth the "Nested Path Modeling of TCM Diagnostic AI System Based on DIKWP Model," focusing on famous doctor diagnostic style modeling, elucidating how to reconstruct TCM diagnostic cognitive path using DIKWP*DIKWP framework, and realizing explainable AI diagnosis through layer-by-layer nested reasoning chain and semantic unfolding. Synthesizing the full text, the DIKWP model demonstrates significant application value in the following three key aspects:

(1) Enhancing Transparency (Explainability) of AI Diagnosis Process:

The DIKWP model breaks down the internal reasoning process of TCM diagnosis into five levels: Data, Information, Knowledge, Wisdom, and Purpose, making every step of cognitive transformation traceable. Compared to black-box models that only give results, the AI doctor based on DIKWP can "Narrate" its own diagnostic thinking like a human doctor. This solves the long-standing trust pain point in medical AI field: patients and doctors often dare not fully trust its conclusions because AI is unexplainable. Now, with the DIKWP framework, AI can explain: "Because the patient has fever, sore throat, coughing yellow phlegm (D), I judge he belongs to Wind-Heat invading Lung Syndrome (K), so I used Sang Ju Yin to treat Wind-Heat (W), aiming to clear heat and resolve exterior as soon as possible (P)." Such a clear logic chain allows professionals to understand its rationality and lets patients understand "Why AI treats like this." This not only improves patient acceptance of AI but also provides doctors with basis to review AI decisions, thereby in Human + AI Collaboration mode, doctors can supervise and correct AI plans, forming a safe and reliable diagnosis and treatment closed loop. Just as the industry view: "What replaces TCM doctors is not AI, but TCM doctors who can use AI." The DIKWP model enhances AI explainability, helping TCM doctors safely use AI as a capable assistant to serve more patients without worrying about AI giving blind suggestions that cannot be regulated. It is foreseeable that with the promotion of DIKWP white-box evaluation framework, AI model internal cognitive chains can be quantitatively evaluated, and performance at each stage can be measured and compared, which will become an important path for Trustworthy and Controllable AI development. In a field concerning life like medicine, transparency is the lifeline, and DIKWP provides a practical solution for this.

(2) Realizing Personalized Reproduction and Inheritance of Famous Doctor Diagnosis and Treatment Experience:

The flexible nesting mechanism of the DIKWP model allows it to accommodate the diagnostic thinking modes of famous doctors from different schools. By constructing Exclusive Knowledge Graphs and Differentiation Logic Rules for each famous doctor, the AI system can embody the style of the corresponding famous doctor. This is equivalent to creating a "Digital Brain" for each famous doctor, realizing the personalized precipitation and innovative application of their experience and academic thoughts. For example, when AI imitates Master Zhang Lei in consultation, it will pay more attention to patient spirit and emotion (because the knowledge base loads Zhang Lei's concept of "Treating Heart First"); when simulating Professor Yue Meizhong, it tends to quickly match classical formulas (due to strong Classic Formula-Special Disease correspondence in his graph); for Professor Deng Tietao, it might increase Pulse Diagnosis weight and often synthesize Chinese and Western indicators (reflecting his integrated Chinese-Western Medicine thinking). In this way, Valuable experience of famous old TCM doctors no longer exists only in books or disciples' memories, but is active in the AI diagnosis and treatment system. Young TCM doctors can get guidance from famous doctors' ideas with AI assistance, as if accompanied by a mentor. This is significant for TCM inheritance: in the past, famous doctor experience inheritance faced problems of limited teachers and random teaching; now AI famous doctor assistants can provide consistent guidance to thousands of medical students. Even better, AI can absorb group wisdom through continuous learning of new cases and efficacy data, drawing inferences from famous doctor experience, thereby promoting the advancement of TCM theory and clinical practice with the times. Professor Yucong Duan and others proposed "Active Learning" and "Incremental Learning" mechanisms, letting AI continuously improve diagnostic accuracy and knowledge base richness from new data. It can be seen that the DIKWP model not only replicates famous doctor thinking but also injects AI's continuous learning ability into it, realizing "Dynamic Inheritance and Development of Famous Doctor Experience." Every famous doctor digital avatar can become "wiser" with data accumulation, and may even derive a new generation of "AI Famous Doctors" in the future, becoming the crystallization of the fusion of real famous doctor experience and big data wisdom. This is the unique contribution of artificial intelligence to TCM personalization and innovative inheritance.

(3) Promoting the Fusion of Standardization, Normalization, and Intelligence in TCM Diagnosis:

TCM diagnostics has long had certain differences in terminology and logic due to different apprenticeship schools, and standardization has always been a difficulty. The DIKWP model requires clear semantic definitions and hierarchical division of various links in TCM diagnosis, which is itself a kind of Standardization Work. For example, what counts as data, what counts as information, what is knowledge (syndrome) must have a unified understanding; also, to establish a layered semantic table of TCM diagnostic terms, it is necessary to standardize the description of etiology and pathogenesis elements of each syndrome. These efforts will promote the formation of consensus Concept Models and Knowledge Graphs in the TCM field. In fact, China is already formulating standards for TCM clinical terminology, syndrome classification, etc. The DIKWP framework can directly apply these standards to model building, realizing the implementation application of standards. A set of standardized TCM knowledge systems plus active AI reasoning will undoubtedly improve the objectivity and repeatability of diagnosis, making the differentiation of the same case by doctors and AI from different regions and backgrounds tend to be consistent. This is helpful for improving TCM clinical efficacy and scientific research exchange. In addition, the DIKWP model naturally supports Integration of Traditional Chinese and Western Medicine and Multi-disciplinary Fusion. Because it uses semantic levels as links, it can include Western medicine test data into the Data Layer, and modern medical knowledge as part of the Knowledge Layer, thereby comprehensively considering under a unified framework. For example, AI can look at TCM syndrome and Western disease simultaneously, converting CT imaging indicators into information to consider together with tongue and pulse information, and then making decisions containing both Chinese medicine and necessary Western medicine. This aligns with the concept of Chinese-Western semantic fusion proposed in the "Active Health" strategy. Professor Yucong Duan is also exploring the role of DIKWP in Chinese-Western integrated active medicine, emphasizing the use of mapping from concept space to semantic space to realize full-dimensional health decision-making. It is foreseeable that future TCM AI diagnostic systems will break through multi-source data based on standardization, integrate diagnosis and treatment plans, and provide Individualized Global Health Solutions. The DIKWP model is exactly the ideal architecture for such complex systems: it acts like a multi-level language, allowing different types of information (whether TCM, Western medicine, subjective, objective) to play roles at their appropriate levels, then deciding uniformly in the Wisdom Layer, and aligning overall health goals in the Purpose Layer. Therefore, the DIKWP model is expected to become a bridge connecting traditional TCM wisdom and modern AI technology, leading the upgrade of the TCM diagnostic paradigm in the fusion of standardization and intelligence.

Conclusion:

TCM diagnostics is a learning balancing art and science. How to let cold AI master the essence of this ancient medical art has always been a fascinating challenge. The DIKWP model gives a brand-new solution idea: through the layer-by-layer nesting of Data-Information-Knowledge-Wisdom-Purpose, we can let AI both "See" the ins and outs of the TCM diagnostic process and "Think Clearly" about the mystery of famous doctor differentiation and treatment. From black box to white box, from imitation to inheritance, from fragmented data to standardized knowledge, the DIKWP framework injects a credible soul into the TCM AI diagnostic system. It makes AI no longer a cold algorithm, but closer to a "Digital Healer" with reason, evidence, and reverence. Imagine in the near future, every grassroots doctor has an AI famous doctor assistant, giving high-level diagnostic advice and detailing reasons at any time; every patient can also understand their condition mechanism and way of nursing through AI, instead of just getting a prescription without explanation. This will be the continuation and promotion of TCM wisdom in the digital age: AI becomes the "Avatar" of famous doctors, TCM is rejuvenated by technology, and the precision, personalization, and accessibility of TCM diagnosis will all step up to a new level.

References

Yucong Duan, et al. Vernacular Deconstruction and Cognitive Adaptation of TCM Diagnosis from DIKWP Semantic Perspective. April 2025.

Yucong Duan, et al. Semantic Simulation and Modeling of the Whole Process of Diagnosis and Treatment of TCM Cold-Pharyngitis-Bronchitis Based on DIKWP Artificial Consciousness. April 2025.

Ping An Good Doctor Releases TCM Intelligent Listening/Smelling System, Launching Full Process AI TCM Platform of Inspection, Listening/Smelling, Inquiry and Palpation. Shanghai Securities News, 2018.

Hairuo Jing. Reading the Prescription Prescribed by AI, the Old TCM Doctor Smiled. 36Kr Pro, April 2025.

Ali Health Deer TCM AI Assisted Diagnosis and Treatment System Integrates Large Model DeepSeek and TCM Knowledge Graph to Improve Diagnosis and Treatment Efficiency. 2023.

Detailed Explanation of TCM Famous Doctor AI Agent Technical Solution. CSDN, November 2025.

Reports on Diagnosis and Treatment Experience of Master Zhang Lei, Academic Thought Materials of Professor Yue Meizhong, and other Famous Doctor Interviews. 2021-2023.

Reference Sources:

ScienceNet—Review of Application of DIKWP Semantic Model in Medical AI - Yucong Duan's Blog, https://blog.sciencenet.cn/blog-3429562-1490177.html

(PDF) Semantic Simulation and Modeling of the Whole Process of Diagnosis and Treatment of TCM Cold-Pharyngitis-Bronchitis Based on DIKWP Artificial Consciousness, https://www.researchgate.net/publication/390670721_jiyuDIKWPrengongyishidezhongyiganmao-yanyan-zhiqiguanyanzhenliaoquanguochengyuyimoniyujianmo

Detailed Explanation of TCM Famous Doctor AI Agent (LLM) Technical Solution_TCM Rehabilitation AI Foundation Database Construction Design Scheme - CSDN Blog, https://blog.csdn.net/2401_84204207/article/details/151070836

Ant Releases AI Health Steward AQ, Famous Doctor "AI Avatars" Online | AI_Sina Finance_Sina.com, https://finance.sina.com.cn/roll/2025-06-26/doc-infcmkhf3373967.shtml

Reading the Prescription Prescribed by AI, the Old TCM Doctor Smiled | AI_Sina Finance_Sina.com, https://finance.sina.com.cn/roll/2025-04-17/doc-inetnfmx3555325.shtml

Shocking! Reading the Prescription Prescribed by AI, the Old TCM Doctor Couldn't Help Smiling, https://eu.36kr.com/zh/p/3253687914983172

(PDF) Vernacular Deconstruction and Cognitive Adaptation of TCM Diagnosis from DIKWP Semantic Perspective, https://www.researchgate.net/publication/390670626_DIKWP_yuyishijiaoxiazhongyizhenduanbaihuahuajiegouyurenzhishipei

(PDF) Research on Digital Space "Safety" Reconstruction and Cognitive Manipulation Deception Mechanism Based on DIKWP×DIKWP Interaction Model, https://www.researchgate.net/publication/396389756_jiyuDIKWPDIKWPjiaohumoxingdeshuzikongjiananquanzhonggouyurenzhicaokongxingqipianjizhiyanjiu

Research on Meaning of Life and Artificial Life in Digital World Based on DIKWP, https://zhuanlan.zhihu.com/p/32024051085

Masters Have Formulas but Don't Prescribe Medicine (Approaching National Medical Masters (24)), http://www.natcm.gov.cn/xinxifabu/meitibaodao/2019-10-11/11290.html

The 3rd National Medical Master Zhang Lei: Sincere Great Doctor, Plain Master Style - China Medical and Health Talent Network, http://www.chinaylws.com/index.php/index/masterDetail?id=86

Professor Yue Meizhong's Experience in Lifting Major Diseases with Classical Prescriptions - 360Doc, http://www.360doc.com/content/22/1005/16/11604731_1050576732.shtml

TCM Academic Tyrant Yue Meizhong: Medical Books TCM Students Should Read (Very Comprehensive), https://zhuanlan.zhihu.com/p/499786887

National Medical Master Zhang Lei's Experience in Treating Difficult and Miscellaneous Diseases with Xuefu Zhuyu Decoction, https://www.sinomed.ac.cn/article.do?ui=2022451727

Buddhist Doctor Yang Ning, Taoist Doctor Zhang Zhishun, TCM Doctor Ni Haisha: Who is the Health Navigator_Life - Sohu, https://www.sohu.com/a/854827290_504192

Research on Inheritance of National Medical Master Deng Tietao's Academic Experience - WeChat Reading, https://weread.qq.com/web/bookDetail/f3432800811e1fafag0148fc

Ping An Good Doctor Launches Domestic First TCM "Intelligent Listening/Smelling" System, https://news.cnstock.com/news,jg-201810-4287760.htm

ScienceNet—Artificial Consciousness Driven by DIKWP Model: Theoretical Framework, White-box Evaluation and Application ..., https://blog.sciencenet.cn/blog-3429562-1493392.html

Meta-analysis of Capability Mapping between DIKWP White-box Evaluation and LLM Black-box Benchmark - Zhihu Column, https://zhuanlan.zhihu.com/p/26428209440

Research on Construction Mechanism and Evolution Path of "Trust" in DIKWP Semantic Model - Zhihu Column, https://zhuanlan.zhihu.com/p/1936030246206969280

Fusing DIKWP White-box Evaluation and LLM Black-box Evaluation to Promote Trustworthy and Controllable Development of Large Models - Zhihu, https://zhuanlan.zhihu.com/p/23657986354

AI "Pulse Taking" VS Old TCM Doctor Inquiry, Which is Stronger? Tide News Reporter Experienced It, https://tidenews.com.cn/news.html?id=2985154

Ali Health: Undervalued Medicine Selling Business and Overvalued AI Medical? - DoNews, https://www.donews.com/article/detail/5331/85651.html

Research on Semantic Fusion of Traditional Chinese and Western Medicine in Active Medicine Based on DIKWP Model - Zhihu Column, https://zhuanlan.zhihu.com/p/1919818190864753944

Is AI TCM Reliable - China News Service, https://www.chinanews.com.cn/sh/2025/02-24/10373326.shtml

Vernacular Deconstruction and Cognitive Adaptation of TCM Diagnosis from DIKWP Semantic Perspective, https://zhuanlan.zhihu.com/p/1893237302307836273

Technical Report on Construction of TCM Standardization System Based on DIKWP Model - Zhihu Column, https://zhuanlan.zhihu.com/p/1928092662382179649

Semantic Mathematical Analysis Report on DIKWP Model and Artificial Consciousness White-box Evaluation by Professor Yucong Duan, https://blog.sciencenet.cn/blog-3429562-1493394.html

Research on Semantic Fusion of Traditional Chinese and Western Medicine and Active Medicine Strategy Based on DIKWP Model, https://xb.ynucm.edu.cn/ynzyydxxb/article/html/20250501

Simulation Case of Self-Construction and Ethical Game of Active AI TCM Diagnosis and Treatment - Zhihu Column, https://zhuanlan.zhihu.com/p/1929897087996261605

Application Analysis of DIKWP Information Field and Energy Field in Neuroscience and Consciousness Research, https://zhuanlan.zhihu.com/p/31585241218

人工意识与人类意识

人工意识日记

玩透DeepSeek：认知解构+技术解析+实践落地

人工意识概论：以DIKWP模型剖析智能差异，借“BUG”理论揭示意识局限

人工智能通识 2025新版 段玉聪 朱绵茂 编著 党建读物出版社

主动医学概论 初级版

世界人工意识大会主席 | 段玉聪

邮箱|duanyucong@hotmail.com

世界人工意识科学院

邮箱 | contact@waac.ac