Challenges and Opportunities in Applying Large‑Model AI to Healthcare

The biggest change in medical AI is the application of large models. Currently, medical AI adopts new technologies quickly, but delivering results is difficult because of a shortage of high‑quality doctors and a lack of advanced technology.

The implementation difficulty stems from the fact that changing doctors' behavior is far harder than altering the practices of pharmaceutical or device companies. Technology aims to improve efficiency and eventually replace tasks, which many physicians resist. For instance, without intelligent medical imaging, patients cannot receive efficient CT services due to overwhelming workloads, and existing models only address specific disease types.

Strategic alignment is not about whether to act, but about what to do, how to do it, and how to invest at each stage, as well as securing technical partners and internal talent—major challenges for enterprises.

Enterprises must break data, business, and process silos and demonstrate the value of digitization, data‑driven operations, and intelligence to business units. Assessing value is especially hard given large investments and generally poor data quality.

In terms of scenario maturity, R&D is the least mature due to long drug‑development cycles and point‑based innovation. Production is largely automated but still relies on traditional data and information systems. Marketing is currently the hottest area, using CDP platforms or sales‑enablement tools to make marketing more compliant and systematic, resulting in clearer, more quantifiable outcomes. Customer‑service scenarios also benefit from scale and infrastructure, enabling data‑driven automation.

Medical AI applications fall into two categories: text processing, which is already thriving abroad, and domestic efforts focused on doctor‑assistant or agent applications.

Doctor‑assistant models consider the physician’s perspective, distinguishing tasks that improve efficiency, replace, or surpass human capability, thereby reducing implementation friction.

Domestic medical AI faces high safety requirements that suppress grassroots innovation and can create monopolies, while large companies lack strong innovation incentives. The absence of high‑quality SaaS offerings further limits AI penetration. Large‑model providers now deliver services akin to SaaS, but both talent and technology are expensive, and data challenges include collection, quality, governance, and sharing.

Consequently, transitioning from project to product carries significant risk. In markets with mature internet ecosystems and strong competition, SaaS adoption is higher, and large models follow suit—explaining why AIGC firms focus on short‑video formats where business models are more viable.

In the United States, engineering culture and financing environments are more mature, making AI startups easier to fund compared with China. Many industry‑specific AI applications in entertainment, law, education, and healthcare originate abroad, with developers leveraging OpenAI frameworks for innovation.

Upholding the belief “Data aggregation in verticals, intelligence leading the future,” DataFunCon2024 Shanghai will continue to follow trends, organizing forums on data intelligence and new biopharma, inviting experts to discuss the present and future of data‑driven implementation.

The event schedule (April 19‑20) is announced, and readers are invited to click “Read the original article” to register.