Tech Giants Dive into AI Healthcare Race

 

Whenever the smoke of war rises, internet giants are never far behind in raising their battle flags.

From the "microblog" wars over a decade ago to the "hundred regiments" battle in group buying, food delivery, news clients, and live video streaming... and now to the "hundred model" wars in AI large models, giants have never been able to stand aloof.

Currently, the competition in large models has shifted from "technology competition" to the second half of "application competition". A trend is that, in addition to general large models, more and more vertical large models targeting specific industries are being released, with the medical scenario receiving particular attention from all parties; and with the continuous emergence of intelligent agents (Agents), the application penetration of AI large models in the medical industry will further accelerate.

On September 5th, Alipay, based on the previously released native multimodal medical large model foundation, officially launched the "AI Health Butler" at the "2024 Inclusion·Bund Conference", with the first batch of more than 20 medical intelligent agents, covering more than 30 health services such as finding doctors, reading reports, accompanying consultations, asking about medical insurance, and managing health.

Alipay's release this time intends to connect all the services users need in the medical and health field through a single product or entry point.

Advertisement

The medical industry, this "pearl on the crown of large models", will it open up new battlefields in the direction of intelligent agents and make this giant competition more and more intense?

And a batch of medical AI startups that emerged around 2015, such as Yingteng Technology, Shukun Technology, Tuixiang Medical, Shenrui Medical, etc., should they focus on their own business or join in response?

Perhaps, the one who tied the bell still needs to untie it.

In fact, since 2023, in addition to the medical large model released by Ant Group, many internet giants have bet on medical large models, such as Baidu's first domestic "industrial-level" medical large model Lingyi Large Model; Tencent's "Tencent Medical Large Model"; JD Health's "Jingyi Qianxun" and so on.

These medical vertical large models all have a general large model as the foundation. For example, the foundation of Ant's medical large model is the self-developed Bailing large model, the technical foundation of Lingyi large model is Baidu's Wenxin large model, and Tencent's medical large model is based on the self-developed Hunyuan large model.

The large model foundation provides a solid technical support for the medical large model, and the user base of hundreds of millions and ample marketing budgets of internet giants also make the promotion and application of medical large models relatively easy, and more confident to win the price war for users and traffic.

Judging only by the domestic large model AI product application data, after a large-scale deployment in Q2 this year, the giants have already taken an absolute lead.

According to AI industry list data, in August 2024, among the top ten domestic AI applications (APP) with monthly active users (MAU), only three are from startups, namely Kimi Intelligent Assistant, Xingye, and Zhi Pu Qingyan, and the rest are from giants such as ByteDance, Baidu, Alibaba, and Kunlun Wanwei, iFLYTEK, etc.

Among them, Douyin Doubao and Baidu Wenxin Yayan have a clear leading advantage, and they are the only AI applications in the country with more than ten million monthly active users, reaching 40.31 million and 11.9 million respectively.

Although the number of users who have used AI large models is in the hundreds of millions, every "urban slave" has more or less generated text, pictures, and videos with AI, but what users can truly feel is: why is AI still so far from changing our lives?

The industry also generally believes that as a capability, large models want to innovate the entire industry, the road is long and arduous. Under the huge large model bubble, there is an urgent need to explore killer scenarios and applications, to let people see a clear commercial path, in order to let the onlooking investors pull the trigger of investment.

This point is especially true for the medical industry.

The arterial network report shows that in the first half of 2024, the number of domestic medical health financing transactions and the total amount of financing decreased by 32.3% and 12.2% respectively. From a not-very-optimistic signal, the total amount of domestic medical health financing in 2024 may break through ten billion US dollars.

Of course, the giants also understand that simply rolling traffic and competing for user scale is ultimately limited in significance, the key is to find the right scenario and run through the payment logic.

Before Alipay launched the "AI Health Butler" this time, Baidu Health also released four medical large model applications including AI Health Assistant, Online Medical Copilot, AI Smart Clinic, CDSS+LLM, and one open platform;

Tencent Health also emphasized many times in the PR draft that rolling large models is meaningless, and rolling applications can truly reflect the value of medical large models. Last September, it released a matrix of AI products in multiple scenarios such as intelligent Q&A, family doctor assistant, and digital medical imaging platform. According to disclosed data, its medical AI and related products have been landed in more than 1,300 institutions.

Under the trend of medical large models, the launch of large model applications or "intelligent agents" is an exploration in process and architectural innovation. On the one hand, it allows the public to directly perceive the convenience that AI brings to medical and health services. On the other hand, it also provides a more specific and clearer commercial path.

Compared with the medical large model applications launched by giants such as Baidu and Tencent, Alipay's previous attempts in the field of electronic medical insurance have accumulated more experience in the field of medicine and health, and to a certain extent, have also cultivated the user habits of online medical services.

However, the medical field is still a "tough bone" for big manufacturers. In addition to selling drugs, major manufacturers have encountered layers of resistance in their strategic layout in the medical field in the past, with little return.

In the medical AI field, where the early investment is high, it is difficult to land, and it is even more difficult to monetize, how long can the internet giants, who have emphasized cost reduction and efficiency in recent years, persist this time?

Due to the large medical needs and the urgent need to improve diagnostic efficiency, the medical industry is undoubtedly one of the fastest vertical fields to land in the entire AI industry. Therefore, the penetration of large models into the medical industry is also the earliest and fastest.

The "2023 Medical Health AI Large Model Industry Research Report" shows that as of October 2023, the number of large models publicly disclosed in China has reached 238, of which nearly 50 are medical large models.

But at the same time, the medical industry just happens to have a great seriousness. When ChatGPT was first introduced with a larger error tolerance, no one dared to truly apply such an error tolerance to the medical industry. Because even a single mistake may lead to a medical accident, a human life.

It is for this reason that the medical large models released by internet giants are still a certain distance from the actual applications.

The ones that have achieved landing applications in the medical AI industry are precisely those "small factories" (startups).

These small factories generally have several characteristics:

First, they focus on more vertical segments. For example, companies such as Yingteng Technology and Zhiyuan Hui Tu focus on AI-assisted diagnosis of retinal images in the ophthalmology track; Shenrui Medical and Tuixiang Medical first focused on the popular segment of lung nodule imaging-assisted diagnosis; Shukun Technology is more deeply involved in the cardiovascular field; Qianglian Zhichuang is positioned in the intelligent diagnosis and treatment of cerebrovascular diseases.

Second, many of the founding teams have medical backgrounds or compound backgrounds. Shenrui CEO Qiao Xin once served as the Vice President of Siemens Medical Division in Greater China; Yingteng founder Zhang Dawei graduated from the Second Military Medical University, and because he loved programming since high school, he successively served as executives in Microsoft, Sina and other internet companies after graduation.

Finally, and most crucially, these startups are not based on Internet logic but medical logic in terms of product development and market promotion.

Under the logic of mobile internet, under the guidance of "traffic is king", all netizens are users, and it is easy to "hold a hammer to find a nail"; under the logic of medicine, users are individuals with specific diagnostic and treatment needs, and solving needs is far more important than creating needs.

Of course, "large models" and "AIGC" not only bring efficiency improvements but also have sufficient attractiveness and imagination space for capital. This has also led to such differences: internet giants first have large models and then look for scenarios to release AI application products, while these startups, which have been established for nearly 10 years, first have scenarios and AI products, and then develop large models, staging a philosophical conundrum of "which came first, the chicken or the egg".

That is to say, unlike giants extending from general large models to vertical large models with general large models as the foundation, medical AI startups develop medical vertical large models from the stage of special (special disease) small models.

For example, the medical-specific large model ShukunGPT independently developed by Shukun Technology, and the SAMI medical image general segmentation large model developed by Shenrui Medical, are all based on their technical accumulation and data accumulation in specific scenarios of medical image-assisted diagnosis. The application of large model technology has further improved the efficiency of their respective business operations.

However, it is worth noting that medical large models are still in the early stages of development. If startups want to go all in on large models, they need to consider how much large models can help the original AI applications or develop new AI applications.

At this stage, startups are putting relatively limited funds together with giants to roll out large models, or focusing more on the research and development and application promotion of special small models, and the ROI is a matter worth careful consideration.

After all, after a wave of enthusiasm for large models and generative AI, capital has more or less demystified AI.

The medical AI field is still waiting for a mature commercial path.

Whether it is rolling traffic or rolling applications, rolling large models or small models, the ultimate goal should be to solve practical problems.

For both giants and startups, the medical industry is destined to be a difficult, lonely, and long-term path that requires persistence.

Faced with the "impossible triangle" of the medical field of "affordable, accessible, and quality medical care", medical AI applications need to be done, and more people who believe in long-term need to do it.

When people forget the technology itself and see the actual medical and health needs of patients or consumers, capital will resonate.