Nowadays, almost all industries are discussing AI, believing that AI will be the engine of the next era, and many companies are launching their own AI models.
In the model wars of various industries, the medical industry is undoubtedly the most lively one. The massive amount of data, policy support, and huge demand jointly drive the development of medical AI. The Internet giants who seized the opportunity to rise earlier will not miss the opportunity of AI. For example, Baidu released the "big model of Lingyi", Tencent launched the "big model of Tencent Health", Alibaba's Alipay released the big model of multimodal health care, JD Health launched the "Beijing Medical Qianxun" medical model, and many new medical AI stars joined the competition, such as the big medical model released by Shangtang Technology, MedGPT released by the Medical Federation, and so on.
Based on various large-scale models, medical AI applications have also begun to be implemented and used. Among them, Baidu Health has launched five major model application products: a family of health intelligent agents for ordinary users; Copilot for online healthcare for doctors and patients; AI intelligent outpatient service for hospitals; The privatization product CDSS (Clinical Decision Support System) for large models; An open platform for spiritual medicine for enterprises. Alibaba's Alipay also released a multimodal medical model in the field of medical AI, and launched a number of digital intelligent solutions including medical trusted all-in-one computers, trusted cloud, and AI application product "AI health butler".
Major companies are laying out medical AI based on their own advantages, but inevitably their products are clustered and homogenized. In the face of this situation, He Mingke, Senior Vice President of Baidu Group and President of Baidu Health Business Group, stated that Baidu Health's advantages lie in its long-term accumulation in the field of large models and its own search traffic advantage.
Despite the emergence of various medical vertical models and medical AI applications like mushrooms after rain, the commercialization path of medical models has not yet been fully realized. He Mingke frankly stated that it seems that there is no application that can solve all problems, not only in the medical and health industry, but also in the entire industry. Nowadays, all big model applications are focused on search, such as ChatGPT which makes search very strong. However, apart from that, no one has found a 'killer app' that can bring value in some narrow scenarios
So, what more needs to be done to achieve large-scale implementation of the medical big model? He Mingke believes that whether it is patients, hospitals, or pharmaceutical companies, what they need is not models, but practical applications. That is to say, identifying the pain points of users is the key.
Like the development track of the Internet, AI is still in its infancy. At the beginning, the rapid development of the Internet benefited from the construction of new infrastructure such as 5G network and data center, while the foundation of AI development was computing power. AI technology has three major elements: data, algorithms, and computing power. If data is the foundation of AI, AI applications require a large amount of data support; Algorithms are the core of AI, helping it understand and process complex data; Computing power is the foundation that supports algorithms, and the more complex the algorithm, the more computing resources it requires. He Mingke believes that in order for the AI industry to achieve rapid development, the first step is to do a good job in infrastructure construction.
The following is an excerpt from the interview with He Mingke by Interface News, briefly edited:
AI empowers healthcare
Interface News: Baidu Health is now launching AI products in three directions: users, hospitals, and pharmaceutical companies. What are Baidu Health's plans in these three directions, especially in the pharmaceutical industry?
He Mingke: Many people have asked us this question, and everyone thinks that when doing a business, we should focus on it. Why does Baidu Health do it on all three ends? The healthcare industry is a unique one among various industries, with most industries achieving a trinity - decision-makers, consumers, and payers are one person. The complexity of the healthcare industry lies in the complete separation of the three: decision-makers are hospitals and doctors, payers are medical insurance or commercial insurance, and consumers are patients themselves.
If it is difficult to commercialize only one end of the service or product, we need to focus on all three aspects, but the ultimate commercialization will be achieved at the pharmaceutical company level. The companies that provide services to pharmaceutical companies abroad are called CXO companies (a general term for pharmaceutical outsourcing), including CRO (Contract Customized R&D Institutions), CDMO (Contract Customized R&D Production Institutions), and CMO (Contract Customized Production Institutions). Companies like Aconwei provide CRO services such as data support and clinical research, which are excellent learning benchmarks for us.
Interface News: In the field of specialized services for pharmaceutical companies, iKunwei is already a relatively mature company. What are the differentiated advantages of Baidu Health compared to it?
He Mingke: Firstly, we are a Chinese company with local advantages. Moreover, Chinese pharmaceutical companies are still in a thriving state. To be honest, in most industries, Chinese companies are basically on par with the United States, but there are relatively few top innovative pharmaceutical companies in China in the pharmaceutical industry. But we believe that China's pharmaceutical industry will also see world-class enterprises emerge, just like the manufacturing or technology industries. We believe it is worth accompanying such enterprises to grow from obscurity to industry leaders. Our advantage in providing CXO services is that we are closer to Chinese companies and have a better understanding of the Chinese market.
Secondly, in the pharmaceutical industry, previously everyone was talking about digitization, but now we are talking about digitization. There is a lot of integration between digitization and AI, which is Baidu's specialty.
Third, Baidu is an Internet company. We have the minds of users and doctors and the corresponding data, so we naturally have data to provide digital and intelligent services.
Interface News: Does Baidu Health mainly engage in preclinical patient recruitment in the pharmaceutical industry, and do they have any other businesses?
He Mingke: In addition, we also have related businesses similar to CRO, such as patient education and doctor education, providing digital systems and products to pharmaceutical companies, etc. These are all our businesses.
Interface News: Which pharmaceutical companies have we already cooperated with?
He Mingke: Basically, most pharmaceutical companies have some related cooperation with us because they have many clinical practice projects in China, which account for a very high proportion globally, second only to the United States. These projects require basic patient recruitment. Moreover, as clinical recruitment of patients is no longer as coarse and granular as before, it will become increasingly refined. At this time, both Baidu's massive user base and AI technology can play a greater value.
Interface News: Baidu Health now has a presence in patients, pharmaceutical companies, and hospitals. Is there any room for expansion in the future?
He Mingke: Although it seems that we have coverage in all three aspects, there is still a lot that needs to be done. The medical industry is already a trillion or ten trillion dollar industry, and as it gradually expands towards the direction of health, especially from the prevention and treatment of diseases to the prevention and treatment of diseases, there is still a lot of space in the health field. Even though it basically covers patients, doctors, and hospitals, there is still a long way to go in reality. Taking hospitals as an example, with so many departments, no one dares to claim that they cover all of them.
In terms of pharmaceutical companies, as mentioned earlier about CXO, there are also many services that can be provided with high ceilings. These top international companies have revenues ranging from tens of billions or higher. So just by doing these three things well, we have a lot to do.
Interface News: Currently, what type of drugs does Baidu Health mainly focus on recruiting patients before clinical trials?
He Mingke: This is related to the difficulty of recruiting patients. Currently, the tumor field is still more popular, with more focus on the field of immunotherapy.
Interface News: Apart from Baidu, other major companies are also laying out medical AI. What are the advantages of Baidu Health?
He Mingke: On the one hand, we have been doing this for a long time and have been using AI to empower hospitals, so our understanding of hospitals and doctors is quite deep. For example, our "AI Smart Clinic" is widely recognized in hospitals and has been promoted throughout Wuhan Union Medical College Hospital. This application can indeed help doctors improve efficiency, enhance the service experience of hospital patients, and is highly consistent with the national policy of intelligent management and intelligent services. This is our advantage.
Secondly, when people want to find a good doctor or hospital, they still prefer to use Baidu search. We can better combine this traffic with digitalization.
Interface News: What do you think is the most important thing when working with hospitals?
He Mingke: I think it's still a pain point for doctors. Firstly, doctors are a professional group that understands their needs and provides services. I think there is a certain threshold, so I always emphasize the accumulation of time.
Secondly, hospitals and doctors place great emphasis on the stability of cooperation. If you do it today and don't do it tomorrow, others won't trust your service. The same goes for pharmaceutical companies. In addition to good product service, stability is also important. For them, if they keep switching suppliers, there are also a lot of internal switching costs.
What are the remaining paths for the large-scale application of medical AI?
Interface News: Baidu Health launched the Lingyi Model last year. What are the difficulties in the development and application of the medical model?
He Mingke: First, Baidu has its own general model, ERNIE Bot, which has great advantages over other competitors. Secondly, AI technology has three pillars - computing power, algorithms, and data. In the healthcare industry, data is a rare resource and data privacy management is also very strict. It is very difficult to obtain data in a reasonable, legal, and compliant manner and use it to train models. For new players, one is the basic big model, and the other is the distribution of data and high-quality data, both of which are difficult points.
Interface News: What are the differences and advantages and disadvantages between medical models and other general models?
He Mingke: The superiority or inferiority of general models and industry models has always been a topic of debate. In fact, the final one is to see how detailed the scene division is and whether it can solve fixed problems in fixed scenarios; Another is to look at data. Medical big models have specific requirements for data, but general big models need to be applicable to all scenarios and may not necessarily use medical data for model optimization. These two points are the core competitive advantages of the medical big model.
Interface News: To evaluate the goodness or badness of a medical model, should we mainly look at these two aspects?
He Mingke: These two dimensions are still very rough, and the thickness of the scene and the amount of data are not quantifiable standards. Finally, it still needs to be examined in practical scenarios, such as user evaluations or evaluations from doctors and hospitals.
Interface news: The development of the large model is only the first step, and there are still many difficulties in its subsequent implementation. From your observation, what other paths are needed for the large-scale implementation of medical models?
He Mingke: I think whether it's patients, hospitals, or pharmaceutical companies, what they need is not models, but practical applications. When we return to the industry perspective, users are more concerned about what the application is, such as our applications in hospitals, such as the intelligent waiting room function of AI intelligent clinics. Through large models, we can generate medical records that meet the requirements, saving doctors' time. Doctors can spend more time communicating with patients, discovering the real causes, and developing better treatment plans. I think this is very valuable.
Our recent medical insurance policy inquiry is aimed at entering the industry. Because medical insurance policy is a very complex matter, it has several dimensions. Firstly, the medical insurance policies in each province of China may vary; Secondly, different categories of drugs correspond to different medical insurance policies; Thirdly, whether one's personal identity is an employee or a farmer, and whether they are insured after retirement, the corresponding policies are also different; Fourthly, the corresponding medical insurance policies vary depending on an individual's status this year, such as whether they have been reimbursed this year, whether they have exceeded the basic line, and whether they have reached the ceiling.
For ordinary people, it is very troublesome to figure out the medical insurance policy. Not just ordinary people, whether they are doctors or insurance companies, will face this problem. At this point, AI models can come into play to solve this problem.
Interface News: Has the medical big model now developed a successful business model?
He Mingke: Still serving pharmaceutical companies. At present, it seems that there is no application that can solve all problems, not only in the medical and health industry, but also in the entire industry. Nowadays, all large model applications are aimed at search, such as ChatGPT, which makes search very strong. However, apart from that, no one has found a "killer app" that can play value in some narrow scenarios.
Interface news: It seems that we have returned to the initial stage of the Internet and started from search.
He Mingke: I think there must be a stage in social development where everyone knows this thing is very useful, but they can't find users. I started to invest in 2008. At that time, the most popular thing was to do mobile Internet. Everyone knew that mobile Internet was valuable, but they did not find a valuable application compared with PC Internet. It was not until 2010 that online car hailing and takeaway appeared that they could be regarded as true mobile Internet applications. Therefore, the development of both the Internet and AI needs a process.
Interface News: Are you also looking for an application scenario that can be implemented now?
He Mingke: Yes, but one problem facing AI development now is that the technology stack and infrastructure are not yet mature. The AI big model actually relies heavily on computing power, and the high-precision AI that everyone expects requires computing power support. From a national perspective, the current computing power cannot support large-scale use of AI. At present, the country has built a computing center in the western region, and the mobile phone hardware needs to be upgraded in computing power. Maybe after the computing center is built and the hardware upgrade is completed, just like the original development of the Internet, the development of Internet companies can only be achieved after the mobile base station and hardware are improved. Similarly, the development of the AI industry is no exception.