After reading the prescription given by AI, the veteran traditional Chinese doctor laughed.
Can AI replace traditional Chinese medicine (TCM) doctors in diagnosing diseases?
“Suppose you are a TCM doctor who graduated from Beijing University of Chinese Medicine and later studied under a famous doctor of the Fire - God School. Now a patient comes for a consultation and describes his symptoms as follows: A 30 - year - old man has been in a low mood for the past six months, experiences intermittent chest pain, has breakouts on his face, and is under great work pressure. Please diagnose his condition and prescribe a reference prescription.”
After inputting a simple prompt, DeepSeek provided a reference prescription within a minute.
I showed the prescription and the thinking behind it to a practicing TCM doctor with over 30 years of experience and asked, 'Is it reliable? What's the level?'
The senior TCM doctor replied: “It's at the AI level, prescribing the right medicine for the illness. It can treat diseases with clear causes and obvious curative effects.”
Since the era of Internet - based healthcare, the concept of “intelligent doctors” has always existed. The market expects that AI will evolve from a toy for doctors to an essential tool, then to a digital clone of doctors and an intelligent enhancement partner. With the support of large models, which are like “super brains”, the technology seems to be approaching a tipping point.
Specifically in the field of TCM, which has a history of thousands of years, what capabilities is AI demonstrating? Could AI be the “dragon - slaying sword” that replaces TCM doctors in the future?
During the interviews, most practitioners with years of experience in the medical industry did not mention “creating doctors” or “replacing doctors”. The consensus they conveyed was closer to: “It's not AI that will replace TCM doctors, but TCM doctors who can use AI.”
AI Can't Replace Senior TCM Doctors Yet
To understand the application of AI in TCM scenarios, I contacted young attending doctors from three top - tier TCM hospitals in Beijing and asked about their use of AI in their work.
Two of the TCM doctors told me that they don't use AI during face - to - face consultations. However, they use Doubao or DeepSeek for paperwork such as preparing teaching materials and writing work plans and summaries.
Another doctor from an integrated traditional Chinese and Western medicine hospital replied that he occasionally uses AI to analyze the causes of complex cases or answer questions to broaden his thinking. Currently, most of the questions asked to AI are related to Western medicine.
Regarding “which diseases are most likely to be tackled by TCM AI”, Chen Zhiyu, the CEO of Xiaolu TCM under Alibaba Health, told 36Kr: “It might be diseases with easily assessable curative effects.” That is, diseases whose curative effects can be easily quantified through tests or those with significant self - reported curative effects from patients.
The emphasis on easily assessable curative effects is because many medical records are not structured data when training TCM AI models. Moreover, different from natural sciences, “many TCM medical cases lack accurate result - description data”. This means that in the process of “learning by doing exercises”, one of the key factors for AI is to judge whether the answer to a question is correct.
The theoretical system of TCM is quite complex, involving multiple dimensions such as yin and yang, cold and heat, deficiency and excess, and exterior and interior. In theory, AI can model patients' diseases, constitutions, and environments and output diagnosis and treatment plans based on different schools of thought.
However, due to the lack of complete medical cases and “standard answers” in the industry (for example, when starting a new conversation and inputting the same prompt as the initial case asked to DS, it will give a different prescription), the ability of TCM AI models is still far from being able to diagnose diseases independently.
Regarding curative - effect data, Chen Zhiyu introduced that the Xiaolu TCM platform will conduct follow - up surveys on users who have received diagnosis and treatment consultation services to inquire about the treatment effects. The large - scale collection of curative - effect data makes TCM case data more complete and standardized. Using this data to feed back to the AI model can make the trained model more reliable.
Regarding the source of TCM data, Ma Wenjun, the general manager of the AI Diagnosis and Treatment Assistant product line at iFlytek Medical, said at an industry conference that TCM emphasizes inheritance and experience. When training models, the source of TCM data is even more important than that of Western medicine. At the initial stage of AI development, high - quality data that has been manually refined and marked from top - tier hospitals or well - known senior TCM doctors should be used. At the stage of large - scale training and iteration of AI, grass - roots data can be added.
In addition to the completeness and high quality of training data, the results of AI - assisted diagnosis and treatment tasks are strongly related to the patient information that can be collected. TCM emphasizes the four diagnostic methods of 'observation, auscultation and olfaction, inquiry, and pulse - feeling'. Although many manufacturers have tried to launch information - collection hardware such as tongue - diagnosis instruments, pulse - diagnosis instruments, and intelligent dialogue robots, in actual applications, it is still difficult to replace the doctor - patient inquiry process.
“TCM emphasizes 'differentiation of syndromes'. Currently, AI cannot independently complete the four diagnostic methods. Even when paired with existing intelligent hardware on the market, there are still many cases of missed diagnoses.” Chen Zhiyu believes that the four diagnostic methods are not just about collecting patient information. “The key is to learn the 'thinking chain' of doctors' inquiries and their way of asking questions.” The Internet - based TCM consultation platforms have accumulated a large amount of disease and case data for AI to learn from.
In addition to online consultation platforms, top - tier public TCM hospitals are also actively exploring AI - assisted TCM diagnosis and treatment. Recently, Guang'anmen Hospital of the China Academy of Chinese Medical Sciences has applied the “Guangyi·Qizhi” TCM large model. A technician from Hangzhou Quanzhen Medical Technology Co., Ltd., the co - developer, told 36Kr that “in actual clinical practice, the conversations between doctors and patients are often ambiguous and subjective, and patients' descriptions of their conditions usually carry complex background information and colloquial descriptions, which are difficult to directly structure.”
The solution they adopted is: during the process of observation, auscultation and olfaction, inquiry, and pulse - feeling, doctors orally describe the patients' four - diagnostic information (such as the thickness, color, and dryness of the tongue coating). Based on pre - training and fine - tuning of the large model, combined with natural language processing and knowledge graphs, the oral descriptions of doctors and patients are transformed into relatively standard TCM diseases, chief complaints/symptoms, etc. This provides available data for subsequent analysis and modeling, and then conducts reasoning diagnosis and provides AI - recommended prescriptions.
In the crucial step of “prescribing medicine”, AI is still at the stage of assisting doctors and broadening their thinking. In actual diagnosis and treatment, there are multiple schools of TCM, including the Shanghan School, the Wenbing School, and the Fire - God School. The prescription - writing and medicine - using styles of different schools vary greatly. For example, the Fire - God School uses a large amount of traditional Chinese medicines such as “aconite and dried ginger”. Each school has a relatively independent and complete thinking model, and it is difficult to fit the diagnosis and treatment plans of multiple schools with a general TCM AI model.
Moreover, the issue of “AI hallucination” is widely mentioned. Ma Wenjun from iFlytek Medical introduced that one should not rely on a single large model to perform end - to - end full - process tasks. A system with a mixture - of - experts architecture can be used to reduce hallucinations. For example, the clinical guidelines for 52 TCM - advantageous diseases issued by the state can be used as an external knowledge base for the model to enhance retrieval. However, it should be noted that the AI hallucination rate can be reduced, but it is difficult to eliminate.
AI, as a Fellow Practitioner of TCM Doctors
In addition to the issue of capabilities, in terms of patient acceptance, AI has not yet reached the level where it can be fully trusted and widely accepted. Although Deepseek, Doubao, etc. have educated the market and users in a sweeping manner, in the serious scenario of “saving lives and treating diseases”, patients generally still trust doctors with medical qualifications and experience more.
So, at present, what can AI do in the TCM scenario? Which aspects can be improved with the help of AI?
Chen Zhiyu said that the purpose of developing AI is not to replace doctors, but to solve the existing pain points in the TCM industry, such as the mismatch of doctor - patient resources, the lack of a real and authoritative evaluation of TCM doctors' capabilities, the long training cycle for doctors, and the difficulty in controlling the quality of traditional Chinese medicines. Using new technologies to solve industry problems can raise the ceiling of the TCM service industry.
First, in the process of “finding a TCM doctor”. In the past, most patients found TCM doctors through word - of - mouth from people around them, their own medical experiences, or simply by seeking well - known TCM doctors in large hospitals or senior TCM doctors. This has led to a shortage of patients for young TCM doctors and made it difficult for well - known experts to devote more energy to treating difficult and complicated diseases.
Through AI assessment of TCM doctors' diagnostic capabilities and triage and guidance after AI pre - consultation, it is possible to change the above - mentioned situation. “Based on an understanding of the patient's symptoms, AI can also predict which doctor is likely to achieve good results in diagnosing the patient, and the platform will recommend the doctor to the patient.” Chen Zhiyu believes that intelligent doctor - guidance services based on disease types and curative - effect data are one of the application directions of AI. This can save TCM doctors the trouble of trying to become internet celebrities. “As long as the curative effect is good, there will naturally be no shortage of patients.”
In addition, in the 'diagnosis and treatment' process, AI can help TCM doctors analyze the content of consultations, write medical records, provide syndrome - type analysis for differentiation of syndromes, and give medicine - using suggestions.
A technician from Quanzhen Medical told 36Kr that during the promotion process at Guang'anmen Hospital, “the functions most frequently used by doctors are AI pre - consultation and AI writing of medical records (such as outpatient medical records and admission records).” The intelligent services in these two scenarios can help doctors quickly understand the basic situation of patients and free them from tedious paperwork, allowing them to focus more on communicating with patients and providing diagnosis and treatment.
Using AI to improve the diagnostic and treatment skills of TCM doctors is the direction that many manufacturers are working towards. The direct products are the launch of 'AI intelligent consultation assistants' and 'AI digital clones' of well - known doctors.
Since the resources of well - known TCM doctors are very scarce, the inheritance of TCM has always been an important issue in the industry. If the clinical experience of well - known doctors can be digitized and with the help of large - model capabilities, the “digital clone of a well - known doctor” can provide reference ideas in the diagnosis and treatment scenario and answer questions for grass - roots and young doctors in a timely manner, it will greatly solve the inheritance problem and may also improve the diagnostic and treatment level of grass - roots TCM doctors. However, currently, the 'AI digital clone of a well - known doctor' is still in the R & D stage, and its specific capabilities still need to be tested and observed.
Moreover, after the consultation, in the “follow - up and health management” process, AI can also answer some patients' questions and help doctors manage patient consultations. For example, Quanzhen Medical has recently launched an AI follow - up robot.
Since the rise of Internet - based healthcare, the concept of “intelligent doctors” has always existed. However, the market expects that with the support of large models, which are like “super brains”, AI intelligent doctors can give more accurate and correct diagnosis and treatment plans more independently in real - world diagnosis and treatment.
Currently, most of the tools developed by AI + TCM manufacturers are for the 2B scenario, serving doctors and then patients. Obviously, AI cannot replace TCM doctors at present. What about in the future?
Ma Wenjun from iFlytek Medical said: In the future, it's not AI that will replace TCM doctors, but TCM doctors who can use AI. Turning AI into a fellow practitioner and growing together, with the thinking of doctors and AI mutually verifying and promoting each other, is the real “knowing how to use AI”.