Tairex, an artificial intelligence company in the medical field that was created by the Institute for Industrial AI Research at Tsinghua University (AIR), has introduced a new system called Multidisciplinary Human-Agent Collaboration (MHAC) and updated its Agent Hospital system. The presentation took place on October 21st at the China National Congress Center in Beijing.
The Agent Hospital system is not entirely new; its research roots lie in the work of Professor Liu Yan's team from AIR. This team developed a simulated hospital where AI-based doctor agents treat millions of synthetic patients, analyze errors, and derive applicable rules. The novelty lies in transitioning from using individual AI assistants to coordinating avatars of various specialists during routine polyclinic appointments.
System Goal: Integration of Expert Consultations
The main objective is to implement multidisciplinary team consultations, where specialists from different departments review one complex case. Tairex claims that such consultations are often conducted too late, after the patient has visited several departments; they are limited to large hospitals and require offline meetings, making them costly. The MHAC system aims to integrate this expertise directly into the routine work of clinics.
In practice, the doctor treats patients as usual. When new information arrives, the system determines whether a consultation with another specialist is relevant and displays the advice of that specialist avatar as a prompt. For example, if a patient being monitored in the endocrinology department for hyperthyroidism also has a history of asthma, the respiratory avatar will warn about risks associated with medications. In cases requiring multiple specialties, several avatars join the consultation review, offering recommendations on diagnosis and medication. The human doctor sees the source of each recommendation and makes the final decision.
The avatars are designed to learn based on the experience of real doctors. One of the company's leaders gave an example: the AI recommended the most effective drug, but the doctor chose a cheaper option covered by health insurance, and such adjustments are used to retrain the system. Tairex also states that with the agreement of both hospitals, external experts can join consultations.
According to the company, by July, the system included 54 AI doctors and millions of simulated patients with thousands of diseases, and the virtual clinic covers 27 departments. At Tsinghua Changgong Hospital in Beijing, 18 clinical departments regularly use the system with over 40 doctors for outpatient examinations. The company and the hospital report an efficiency increase of approximately 30% and a MedQA accuracy above 96%, although these figures have not yet undergone independent verification, and the AI does not diagnose independently.
