「全球首家 AI 全科醫院在上海誕生」？事件真相與深入解析

近日，網路熱傳一則令人振奮的新聞——號稱「全球第一家AI全科醫院」在中國上海正式上線，主打24小時營業、不需排隊、不需掛號專家號，每天最多能接診3,000名病患，還能涵蓋21個科別、超過300種病症，由「42位AI醫生」提供服務。這項消息迅速在社群平台引爆，甚至一度引起醫療界與科技圈的強烈關注。

但事實是否真如標題所說？經查證後發現，這項「AI醫院」的概念確有其真實原型，但目前並非一家實體醫院，而是一項仍處於內部測試階段的AI醫療系統，名為 「紫荊 AI醫生」系統（Agent Hospital），由清華大學旗下的AIR（人工智能研究院）研發。紫荊AI醫生是一套以大型語言模型為基礎，結合龐大醫學知識庫與病例資料訓練而成的虛擬診療系統。它透過模擬醫院環境，建立出21個虛擬科別，並配置42位AI醫師角色，涵蓋內科、外科、神經科、婦科、腫瘤科、小兒科等領域，聲稱能夠處理超過 300種常見與複雜疾病。

根據團隊公開的資料，在封閉測試中，該系統可在短時間內完成超過1 名「虛擬病人」的完整診療流程，包括問診、檢查、診斷、處方與隨訪，平均每位病患的診療時間不到15分鐘。其診斷正確率在某些標準測試資料集（如MedQA）中達到93%左右，效率遠超人類醫師，並展示出其潛在的應用價值。然而，這些「病患」與「醫師」目前全部都存在於模擬與模型環境中，並未真正應用於現實世界的臨床醫療流程之中。換句話說，這是一套「AI醫療系統雛形」，而非具備醫療執照與病床的實體醫院。儘管「紫荊AI醫生」展現令人印象深刻的演算與決策能力，但其應用尚存多項技術與倫理上的限制。

首先，在臨床場景中，診療往往涉及複雜的人體檢查、病史交叉驗證、患者主觀陳述的解讀與非語言訊號的感知，這些人類醫師的經驗與直覺難以完全由 AI 模型複製。其次，目前醫療規範（如中國的《執業醫師法》與《醫療器械條例》）對於 AI 系統作為主診斷工具仍持審慎態度，多數 AI 工具只能作為醫師輔助系統使用。若未經註冊核准即上線診療人類病患，可能涉及非法執業與醫療事故風險。因此，雖然 AI 醫院的願景美好，但短期內尚難全面落地應用，仍須依賴人類醫師把關診療品質。

根據媒體報導與Reddit、微博、知乎等社群討論，民眾對這項技術的態度呈現兩極。一方面，有人認為這是醫療公平與效率的重大突破，特別是對於醫療資源匱乏地區來說，AI醫生的普及可望緩解「看病難、看病貴」的問題，提升整體公共健康水準。另一方面，也有不少人擔憂AI系統會出現誤診、資料偏誤或演算法歧視等問題，甚至會引發醫患責任歸屬混亂等法律爭議。

更有專業醫師指出，醫療行為是高度人性化與倫理性的行業，單靠數據驅動的決策在很多緊急情況下可能無法妥善應對。截至目前為止，「紫荊AI醫生」仍處於封閉式模擬測試階段，尚未對社會大眾開放實際掛號或診療功能。那些聲稱「9.9元掛號費、5分鐘問診、10分鐘報告、15分鐘治療」的說法，實際上是基於內部模型測試的理想效率數據，尚未在真實病患中驗證其可靠性。因此，這並不是真正意義上的「醫療機構上線」，而是一項值得關注、但仍處於早期發展的技術原型。

從長遠來看，人工智慧在醫療領域的應用潛力無庸置疑。無論是影像判讀、初步分診、用藥建議，甚至是慢性病管理與健康諮詢，AI都能顯著提升效率與覆蓋面。然而，在現階段，應理性看待這些技術成果。AI或許能做到大量基礎工作自動化，但「看病」從來不是單純的資訊對應遊戲，它涉及的是生命、感受與判斷，仍需醫師、病人與制度三方之間的信任與協調。「AI醫院」不是科幻，而是科學正在邁向未來的腳步聲。它可能在未來幾年真正問世，但那一天的到來，仍需人類的耐心、制度的跟進與技術的謹慎推進。

“World’s First AI General Hospital in Shanghai”? The Truth Behind the Viral Claim and In-Depth Analysis

Recently, a sensational piece of news has gone viral across social media: the world’s first AI-powered general hospital has officially opened in Shanghai, China. The hospital allegedly operates 24/7, requires no queues or specialist appointments, and claims to serve up to 3,000 patients per day across 21 departments, with diagnoses and treatments handled by 42 AI “doctors”. This announcement sparked a wave of excitement and intense discussion, particularly within the medical and tech communities.

However, is this headline really as groundbreaking as it seems? Upon further investigation, it turns out that while the concept of this “AI hospital” has a real basis, it is not yet a physical hospital, but rather an AI medical system prototype still under internal testing. The project is called “Agent Hospital”, also known as the Zijing AI Doctor System, developed by Tsinghua University’s Institute for AI Research (AIR).

What Is the “Zijing AI Doctor” System?

The Zijing AI Doctor is a virtual medical system built on large language models, trained on a vast repository of medical knowledge and patient case data. It simulates a hospital environment with 21 virtual departments, each staffed by AI doctor agents, covering a wide range of fields such as internal medicine, surgery, neurology, gynecology, oncology, and pediatrics. The system claims to handle over 300 common and complex illnesses.

According to the development team, in closed testing environments, the system can diagnose, examine, prescribe, and follow up on virtual patients in less than 15 minutes per case. On standardized benchmark datasets like MedQA, the system achieved a diagnostic accuracy of around 93%, greatly exceeding the average efficiency of human doctors and suggesting considerable future potential.

Virtual, Not Real: A Prototype, Not a Licensed Medical Institution

However, both the patients and doctors in this system are entirely virtual, existing only within simulated and model-based environments. It is not a real hospital, nor is it legally authorized to provide clinical diagnosis or treatment in the real world. In essence, this is a conceptual prototype for an AI-based healthcare delivery system—not a functioning, licensed medical facility with hospital beds or practicing physicians.

Despite its impressive capabilities, there are multiple technical, legal, and ethical hurdles preventing such systems from being deployed in actual clinical practice at this stage.

The Current Challenges: Technology and Ethics

First, real-world clinical diagnosis is far more complex than pattern matching or textbook responses. It involves physical examinations, cross-checking of medical history, interpreting patient narratives, and even non-verbal cues—aspects of human intuition and experience that AI models cannot yet replicate.

Second, medical regulations, such as China’s Law on Practicing Physicians and Medical Device Regulations, take a cautious approach to AI in healthcare. AI systems are currently permitted only as decision-support tools—not as primary diagnostic authorities. Using an unregistered AI model to diagnose real patients could be deemed unauthorized practice of medicine, with potential legal consequences.

Thus, although the vision of an AI-powered hospital is inspiring, its widespread application is unlikely in the near term, and human doctors must continue to oversee and ensure the quality of diagnosis and treatment.

Public Reaction: Between Hope and Skepticism

On platforms like Reddit, Weibo, and Zhihu, public reactions have been mixed. On one hand, many people see this as a breakthrough in medical accessibility, especially for under-resourced regions. AI doctors could alleviate long-standing issues such as “difficult and expensive access to care”, and help improve public health equity.

On the other hand, critics are concerned about risks such as misdiagnosis, data bias, algorithmic discrimination, and accountability in the event of medical errors. Legal ambiguities around liability and consent are also widely debated.Some medical professionals have voiced concerns that healthcare is inherently human and ethical in nature, and that AI, while powerful in handling data, may fail in emergency or emotionally sensitive scenarios that require moral judgment and empathy.

What the “9.9 Yuan Registration” Claims Actually Mean

Claims such as “9.9 RMB registration, 5-minute consultation, 10-minute diagnosis, and 15-minute treatment” are based on internal model testing simulations, not on interactions with real patients. The numbers reflect ideal performance benchmarks within a controlled environment—not proven clinical outcomes.

Therefore, this is not yet a functioning AI hospital, but a technological prototype with promising early results, still undergoing research and testing.

Looking Forward: Not Sci-Fi, But Still Far From Reality

In the long term, the potential of AI in healthcare is undeniable. From medical imaging, triage, drug recommendations, to chronic disease management and wellness counseling, AI can significantly improve efficiency and coverage.

But at this stage, these technologies must be viewed rationally and realistically. Treating illness is not a simple matter of matching symptoms to data. It involves life, human emotion, and moral judgment—which require trust and collaboration among doctors, patients, and systems.

The “AI Hospital” is not a scene from science fiction, but a scientific milestone under cautious development. While we may one day see such systems in widespread clinical use, that future depends on careful technical progress, regulatory adaptation, and continued human oversight.

In short: AI may assist medicine—but cannot replace its human heart.