China’s tech scene is buzzing, and this time, it’s not just about the latest smartphone or e-commerce innovation. The spotlight is on DeepSeek, a powerful AI model that’s making waves across various sectors, from coding to, believe it or not, healthcare. For those of us living here, it feels like every week there’s a new headline about AI changing the game. But is this hype or reality, especially when it comes to something as critical as our health?
Recent reports suggest DeepSeek isn’t just another tech novelty; it’s being deployed in hospitals across China, potentially marking a significant turning point for “AI healthcare,” a field that has seen its share of booms and busts over the past decade. After years of promise and some setbacks, could DeepSeek be the breakthrough that finally brings AI-powered medical assistance into the mainstream?
To understand the buzz, let’s dive into what’s happening on the ground. Imagine a bustling emergency room or a packed Intensive Care Unit (ICU) in a major Chinese hospital. These are places where every second counts, and doctors are constantly juggling complex cases, making critical decisions under immense pressure. Traditionally, in these high-stakes environments, patient outcomes hinge almost entirely on the expertise and rapid response of human doctors. But now, picture this: doctors equipped with an AI assistant powered by DeepSeek, capable of instantly analyzing vast amounts of patient data, sifting through intricate medical histories, and proposing treatment strategies in real-time.
This isn’t science fiction anymore. Hospitals like Shenzhen University-affiliated Huainan Hospital, Kunshan First People’s Hospital, and the Second Affiliated Hospital of Army Medical University have already started implementing DeepSeek models. Med-Go, a medical model integrated with the massive DeepSeek-R1 (boasting a staggering 671 billion parameters), is being clinically tested at Shanghai Eastern Hospital and other institutions, with initial validations focusing on its capabilities within the demanding ICU setting.
For Americans accustomed to long wait times for specialist appointments and navigating a complex healthcare system, the idea of AI streamlining medical processes is incredibly appealing. China, with its massive population and uneven distribution of medical resources, faces even greater challenges in ensuring quality healthcare for everyone. For years, AI in healthcare has been touted as a potential solution to these very problems – aiming to bridge the gap in access to experienced doctors, improve diagnostic accuracy, and ultimately, make healthcare more efficient and affordable.
From AI-powered medical imaging analysis to drug discovery, surgical robots, and now, large language models in medicine, numerous companies have ventured into this space, hoping to leverage AI to solve these systemic issues. This initial wave of enthusiasm from 2016 to 2023 even birthed several AI healthcare unicorns, companies valued at over a billion dollars. However, many of these early AI medical products, despite securing regulatory approvals as medical devices, hit roadblocks when it came to widespread adoption in hospitals and securing patient or insurance payments. As one healthcare investment partner put it, the hurdles were twofold: “insufficient technological barriers leading to industry competition, and a lack of strong clinical demand due to unclear payment models.” In simpler terms, the tech wasn’t quite good enough to stand out, and nobody was really sure who was going to pay for it.
But the emergence of DeepSeek and similar large language models may be changing the narrative. The capacity of these models to understand and reason in medical contexts has ignited a new wave of excitement. The question now is: can these advancements in AI technology finally overcome the previous obstacles and propel AI healthcare into a phase of meaningful impact? Can the “silicon-based life” of AI genuinely alleviate the “carbon-based life” challenges of “difficult and expensive healthcare,” as the common Chinese saying goes?
DeepSeek’s Deep Dive into Medical Applications
This February, a flurry of Chinese healthcare and pharmaceutical companies, including names like Yidu Tech, Airdoc, and Zhiyun Health, announced their integration of DeepSeek into their operations. The goal? To enhance their existing services in medical data analysis, AI-driven image diagnostics, and chronic disease management. One Hong Kong-listed medical company, currently integrating DeepSeek, reported that their tech teams are swamped, focusing on “landing” projects as quickly as possible due to overwhelming interest from hospital clients. Similarly, a Tencent-backed internet healthcare firm indicated they are actively researching DeepSeek, with plans for detailed implementation within weeks.
While some tech experts might argue that DeepSeek isn’t a “disruptive innovation” in the purest sense, its refined performance in reasoning and decision-making scenarios after “fine-tuning” is undeniably valuable. It offers a significant boost to AI medical products, especially in handling complex medical data and supporting precise clinical decisions. This is precisely what AI healthcare companies desperately need right now.
Zhang Hanxiang, CEO of Shule Information, the company behind the Med-Go AI medical model, emphasized the critical reasoning and mathematical capabilities of DeepSeek-R1 (671B). He told 36Kr Pro that integrating DeepSeek has improved diagnostic accuracy by over 10%, with even more pronounced improvements in complex case diagnoses. Launched in collaboration with Shanghai Eastern Hospital last November, Med-Go, now enhanced by DeepSeek, is proving its mettle in high-pressure environments like the ICU.
Imagine a patient with multiple organ failure. Doctors face a daunting array of treatment options: mechanical ventilation, blood purification, medication cocktails, and more. This is where DeepSeek shines. By rapidly processing a patient’s vital signs, lab results, and imaging data, alongside their medical history and the latest research, DeepSeek can generate a comprehensive decision-support framework for doctors, offering multiple potential diagnostic pathways. This isn’t about replacing the doctor, but empowering them with an incredibly efficient and knowledgeable AI assistant that works tirelessly, 24/7.
The impact isn’t limited to hospital settings. Anecdotal evidence is surfacing on social media platforms like Xiaohongshu, a popular Chinese social media and e-commerce platform. During the recent Spring Festival holiday, parents reportedly used DeepSeek to interpret their children’s blood test reports and seek medication advice. In one instance, DeepSeek’s diagnosis reportedly aligned with that of a chief physician at a major public hospital. While a doctor’s in-person consultation often comes with limited explanation due to time constraints, AI medical models can provide continuous, detailed responses to concerned parents’ questions, demystifying medical jargon and easing anxieties.
Of course, the “AI hallucination” problem – where AI models can generate nonsensical or inaccurate information – is still a factor with DeepSeek. While the base model possesses impressive reasoning power, its medical knowledge base isn’t yet specialized enough. The level of detail in data processing needs further refinement. Direct application in serious clinical scenarios could lead to inaccurate diagnoses and potentially misleading treatment or medication recommendations.
However, this limitation also creates an opportunity for “deep customization.” Companies can leverage the open-source nature of DeepSeek and enhance it with expert-annotated datasets, incorporate the diagnostic reasoning processes of experienced medical teams, and fine-tune the model to improve accuracy. This is where the years of accumulated expertise of existing AI healthcare companies become invaluable.
Another significant advantage of DeepSeek’s approach is its ability to address data privacy concerns, a long-standing issue hindering the adoption of AI in healthcare. DeepSeek’s open-source model, coupled with the option for localized deployment, offers a viable solution. AI agents can be deployed within hospital networks, ensuring data privacy at the source and reducing reliance on third-party cloud platforms, which also translates to cost savings.
Beyond diagnostics and treatment, DeepSeek is showing promise in the business-to-business (B2B) medical distribution sector. Shangyao Cloud Health, a company focused on pharmacy retail, told 36Kr Pro they are already applying DeepSeek in areas like intelligent Q&A, pharmacist training, and personalized patient management. For example, in prescription verification, DeepSeek could help pharmacists quickly identify potential drug interactions or dosage errors, thereby reducing medication risks.
The Quest for the Medical “Killer App”
AI in healthcare in China has been on a rollercoaster ride for a decade. Remember the buzz around AlphaGo’s victory over Ke Jie in Go? Similar “AI vs. Doctor” competitions have been staged in China, generating initial excitement and investment. Over these ten years, AI tools have evolved from “toys” to doctor’s assistants, and now, with advancements like DeepSeek, we’re inching closer to the concept of a truly intelligent “AI doctor.”
In 2024, the market’s evaluation of medical large language models is shifting from simply assessing model capabilities to a more business-oriented approach. The focus now is on identifying “killer applications” – use cases that are not only technologically impressive but also practically valuable and commercially viable. Creating a successful “killer app” in medical AI requires a holistic approach, addressing technology, product design, doctor and patient education, business models, regulatory compliance, and ethical considerations – a complex ecosystem that needs to work in harmony.
DeepSeek’s value lies not only in its technological prowess but also in its broad public visibility. Its widespread attention serves as a crucial “AI usage education” for both doctors and patients. While some doctors, especially younger generations, are eager to embrace new technologies, a significant portion of the medical community remains conservative. DeepSeek’s demonstrated capabilities are starting to sway more doctors towards adopting AI. For patients, as AI tools become increasingly integrated into daily life and work, using AI for health consultations, image analysis, and even initial diagnoses feels like a natural progression.
Currently, even popular Chinese chatbots like Baidu’s Ernie Bot (豆包) have integrated health consultation assistants, such as the “AI Health Consultation Assistant” launched by Xiaohe Health, guiding users in health-related dialogues. However, unlike casual conversations in other contexts, medical consultations and cancer screenings demand high accuracy and have a low tolerance for AI “hallucinations” or fabricated information. Imagine an AI health assistant making a misdiagnosis and a patient following its medication advice, leading to adverse health consequences. This raises serious questions about medical liability and responsibility, underscoring the need for robust accuracy and stringent regulatory frameworks.
Furthermore, the perennial challenge of commercialization persists. For AI products aimed at clinical diagnosis and treatment, the “pay-per-use” model seems to be the most sustainable path to generate continuous revenue while meeting patient needs. In the past, many AI healthcare companies offered similar services, leading to intense competition and a race to the bottom, often resulting in a “one-time software sale” business model that struggled to generate profits for the industry as a whole.
However, past failures are merely prologues. Rapid technological advancements, particularly in models like DeepSeek, are opening up new potentials for AI in healthcare, both inside hospitals and across the pharmaceutical supply chain. The industry is poised for further evolution, and the next few years will be crucial in determining whether AI can truly revolutionize healthcare in China and beyond.
Patient Skepticism and Doctor “Collapse”: The Human Element Remains
Interestingly, DeepSeek’s foray into healthcare isn’t without its human drama, even a bit of comedy. Recent online buzz highlights a fascinating dynamic: patients using DeepSeek to question their doctors’ diagnoses. One Guangdong-based doctor, known online as “Lone Ranger Appreciation,” shared a humorous yet thought-provoking experience online. After prescribing a treatment plan, the patient, armed with DeepSeek’s analysis, challenged the doctor’s recommendations. Initially annoyed, the doctor double-checked the medical guidelines and, to their surprise, discovered the guidelines had been updated – DeepSeek was right! The doctor jokingly lamented online about feeling like their “world had collapsed.” This anecdote went viral, garnering over 150,000 likes and thousands of comments, sparking widespread discussion about the role of AI in healthcare.
This scenario isn’t isolated. Many doctors online have shared similar experiences. One doctor commented, “Yesterday, during my outpatient clinic, a patient showed me DeepSeek’s results, which were identical to my diagnosis. The patient actually praised me for being professional!” Another doctor quipped, “I feel like my retirement is getting closer.”
However, the same doctor who initially felt his “world collapsing” later clarified that DeepSeek’s diagnosis was indeed correct. He admitted that the hospital was using a slightly different medication due to local availability, but this had been explained to the patient. This incident highlights a critical point: AI can be a powerful tool for patients, empowering them to be more informed and engaged in their healthcare.
Another doctor, “Phantom Space” online, shared an experience of a patient directly requesting a prescription based on DeepSeek’s diagnosis. He prescribed the medication, which was for a minor condition, and later consulted DeepSeek himself, finding its recommendations reasonable. He emphasized the importance of communication and risk disclosure when dealing with patients who bring AI-generated diagnoses.
These anecdotes, while humorous, underscore a crucial shift: patients are becoming more proactive and informed, leveraging AI as a second opinion. However, experts caution against patients blindly trusting AI. Dr. Hu Wei, a attending physician at the Children’s Hospital of Chongqing Medical University, emphasizes that treatment decisions must be based on accurate diagnoses. While AI excels in certain areas like image analysis, accurately diagnosing complex conditions and their stages remains challenging. He points out that physical examinations rely heavily on a doctor’s experience and intuition – skills that AI currently lacks. For example, diagnosing appendicitis requires tactile examination and clinical judgment, which AI cannot fully replicate.
Dr. Hu acknowledges that AI can provide accurate and up-to-date treatment options, but emphasizes the need for personalized care. Doctors tailor treatment plans based on individual patient conditions, medical history, and other factors. Blindly following AI-generated recommendations without considering these nuances can be risky.
Regulatory Landscapes and the Limits of AI
As AI becomes more integrated into healthcare, regulatory frameworks are playing catch-up. Hunan province recently issued a strict ban on internet hospitals using AI to generate prescriptions, reflecting concerns about patient safety and the potential for misuse. This follows similar “red lines” drawn in other regions. Regulations in China, such as the “Internet Diagnosis and Treatment Management Measures (Trial Implementation),” explicitly prohibit the use of AI to generate prescriptions, emphasizing that prescriptions must be issued by qualified physicians after proper consultation.
Experts like Wei Zining emphasize that AI currently has no prescribing authority. Allowing AI to prescribe medication would be legally and ethically problematic, as AI cannot be held accountable for adverse outcomes. However, AI can be a valuable tool for doctors, assisting with image analysis, report interpretation, and treatment recommendations. But, as Wei Zining stresses, AI should remain a supporting tool, not a replacement for human medical expertise. “For at least the next few decades, AI will only be an assistant to doctors, not a substitute,” she states.
Lawyers like Fu Jian point out that current user agreements for AI medical tools like DeepSeek explicitly state that their recommendations are for reference only and not professional medical advice. If patients use DeepSeek for self-diagnosis and treatment and experience adverse effects, DeepSeek is unlikely to be held legally liable. The responsibility ultimately lies with the individual making healthcare decisions based on AI-generated information.
Zhang Qihuai, another lawyer, reinforces this point, highlighting that medical prescriptions in China must be issued by licensed medical institutions and qualified doctors. Patients who self-treat based on DeepSeek’s recommendations without professional medical confirmation do so at their own risk. He also notes that Beijing has regulations prohibiting doctors from using AI to generate prescriptions, further underscoring the regulatory caution surrounding AI in healthcare.
The Future is Collaborative, Not Competitive
In conclusion, DeepSeek and similar AI models represent a significant advancement in AI healthcare in China. They offer the potential to improve diagnostic accuracy, streamline clinical workflows, and empower both doctors and patients with better information. However, it’s crucial to recognize that AI is currently a tool to augment, not replace, human medical expertise. While AI can analyze data and offer insights, the nuanced judgment, empathy, and personalized care that human doctors provide remain indispensable.
The path forward for AI in healthcare is likely to be collaborative. AI can handle data-intensive tasks, assist with diagnosis, and provide decision support, freeing up doctors to focus on complex cases, patient communication, and the human aspects of care. For patients, AI can be a valuable resource for information and second opinions, but it should not replace professional medical consultations. As AI technology continues to evolve, the key will be to harness its power responsibly, ethically, and in a way that ultimately enhances, rather than diminishes, the human element at the heart of healthcare. For now, the dream of AI revolutionizing healthcare in China is still unfolding, with DeepSeek playing a fascinating and potentially transformative role.
评论