2026 Zhongguancun Forum Annual Conference | From Efficiency Competition to Disease Cognition Revolution, AI Empowers Innovation and Development in Medicine and Medical Devices

(Source: Beijing Business Today)

Smart healthcare is becoming one of the application scenarios with the greatest social value for AI technology. On March 26, at the 2026 Zhongguancun Forum annual conference “Empowering Life and Health with Intelligence · AI Leading the Future” forum, experts attending the meeting engaged in in-depth exchanges around strategic layout, changes in research paradigms, industrial implementation paths, and explorations of technological frontiers. They reached a consensus that enabling life and health with AI is both a challenge and an opportunity. It requires the government, academia, and industry to work in the same direction to jointly connect the full chain—from data foundations, to evaluation and rulemaking, and then to application deployment. From the predicament of “a decade and $1 billion” in drug R&D, to the intelligent upgrade of medical devices, from the accurate simulation of digital human body models to the cutting-edge exploration of virtual clinical trials, AI not only brings a leap in efficiency, but also is driving a profound transformation: from experience-driven to data-driven, and from an efficiency race to a revolution in disease understanding.

Data Foundations and Evaluation Rulemaking

Liu Yuanli, an adviser to the State Council and a long-term appointed professor at the School of Health Management and Policy, Peking Union Medical College, combined the results of his national strategic special-topic research on medical-health artificial intelligence led by the Adviser Office of the State Council. He systematically sorted out the three major bottlenecks facing China’s AI-enabled healthcare sector today.

Liu Yuanli summarized the difficulties that urgently need to be overcome with three keywords: data, evaluation, and deployment. The first bottleneck is “the data dilemma.” China has a huge population base and a healthcare system mainly comprising public hospitals, which theoretically provides both advantages in data resources and institutional mechanisms. However, there is insufficient supply of high-quality, standardized, multimodal medical and health data, and an efficient, trustworthy, secure data sharing and circulation mechanism has not yet been formed.

As for the reasons, Liu Yuanli summarized them as “three don’ts”: cannot, dare not, and do not want to. “Cannot” is because medical data are multimodal, highly complex, and highly specialized; many hospitals lack mature data governance and development capabilities. “Dare not” stems from the fact that health and medical data are highly sensitive; the pressure of privacy protection and safety responsibility is heavy, and sharing concerns are substantial. “Do not want to” is because there is a lack of reasonable incentive and value-reward mechanisms, so the motivation of data contributors is clearly insufficient.

The second bottleneck is “difficulty in evaluation.” Medical AI directly concerns people’s life and health, so there is no room for ambiguity. With large models iterating quickly—showing enormous value—issues such as an unexplainable black box, algorithmic bias, and the risk of misdiagnosis and missed diagnoses have emerged alongside them. “The more advanced the technology is, the more the regulation needs to keep up.” Liu Yuanli emphasized that it is necessary to accelerate the establishment of an authoritative evaluation mechanism and platform covering the entire chain of R&D, approval, and application supervision. Using unified, scientific, authoritative standards to draw the safety boundary for technological innovation, and to set a quality baseline.

The third bottleneck is “pain in deployment.” No matter how good the technology is, it only has real value when it is actually put to use. Currently, medical AI faces the “last mile” problem, requiring multiple obstacles—policy, pricing, processes, habits, and more—to be cleared so that mature and reliable intelligent products can truly enter hospitals, enter homes, and benefit the public. Liu Yuanli said that from data foundations, to evaluation and rulemaking, and then to application deployment, every step requires the government, academia, industry, and the medical and health sector to move in the same direction. Under national strategic guidance, they need to pool efforts to break bottlenecks, tackle difficulties, and connect broken links.

Research Paradigms and Talent Development

Xing Xinhui, vice president in charge of work at the Shenzhen International Graduate School, Tsinghua University, and president of the Institute of Biomedical and Health Engineering, as an adviser to the People’s Government of Beijing, shared innovative experience in the deep integration of AI and life sciences, based on the institute’s six years of exploration and practice.

Xing Xinhui introduced that within the institute’s curriculum modules, all interdisciplinary and innovative courses include AI-related content, and strong AI faculty participation brings cross-disciplinary integration. In research, over the past six years, in the master’s and doctoral thesis research conducted at the institute, 90% incorporated AI into research practice. This deep integration is not only reflected in talent cultivation, but also has produced substantive breakthroughs in research innovation.

Taking examples of polypeptide drugs and active peptide mining: traditional approaches mainly rely on experience and trial and error, with a success rate of less than 1%. By involving AI, the activity prediction and information screening of sequences and targets are deeply integrated, and efficiency can be increased by ten times, a hundred times, or even a thousand times. “Before we even run experiments, we can judge which molecule will have an easier path to go down after being constructed through the AI model, significantly reducing the trial-and-error rate and mining target molecules more precisely.” Xing Xinhui said.

In addition, the institute has also built a globally leading digital human body model, covering different groups such as infants and young children, adult males, adult females, and elderly people, with metabolic models covering more than 100 different organs and cells. It can predict changes, toxic side effects, and impacts on the gut before a drug or food enters the human body. In the equipment domain, the high-throughput cell screening equipment developed by the team not only serves domestically, but also is exported to developed countries such as Japan, the United States, South Korea, and France. In digital pathology, by combining AI with high-throughput pathology slide systems, tasks such as accurate tumor diagnosis, gene mutation prediction, and prognosis assessment can be completed precisely, providing strong support for clinical precision diagnosis and treatment.

Industrial Practice and Technological Frontiers

Cui Song, co-founder, chairman, and chief executive officer of Beijing Nuocheng Jianhua Pharmaceutical Technology Co., Ltd., from the perspective of a practitioner on the front lines of innovative drug companies, shared practical pathways for using AI to enable drug R&D.

Cui Song pointed out that from target selection to molecular design and then to clinical trials, new drug R&D takes as long as a decade and requires investments exceeding $1 billion. AI has already played an important role in improving operational efficiency within the company, such as patient enrollment optimization, real-time data transformation dashboards, summaries of production data, and more. However, the core problem that AI has not yet solved is: how to find drugs that truly work in clinical settings from 0 to 1. “Molecules designed by AI have great binding strength to proteins and very strong affinity, but can they be directly skipped over from animal experiments into clinical use? The regulators don’t allow it now,” Cui Song frankly stated. From AI prediction to becoming an approved上市 drug, there is still a huge gap in between. In the future, if it becomes possible to replace some laboratory validation steps with AI and if it is recognized by the NMPA, the timeline for new drug R&D could potentially be shortened from a decade to two or three years.

Zhao Yu, professor at the Western Institute of Computing Technology, and vice director of the Turing–Darwin Laboratory, put forward a more disruptive viewpoint from the perspective of technological frontiers. He noted that the industry’s current use of AI mostly remains at the level of statistics and information technology, and AI in the true sense has not yet been sufficiently understood.

Zhao Yu emphasized that the most difficult part of drug R&D is “don’t get it wrong at the root.” Currently, innovative drugs show significant effects in animal experiments, but 95% fail in clinical trials. The fundamental reason is that our understanding of disease mechanisms is not truly clear. “What we lack is not molecules; what we lack is disease-treating molecules. The first-principles approach of this industry is understanding diseases.” After nearly three decades of accumulation, Zhao Yu’s team has built a complete computational medicine system. Starting from diseases, they clarify the mechanisms of target action and the beneficiary populations, and then proceed with molecular design. They have completed the world’s first virtual clinical trial, achieving 100% accuracy by prospectively predicting patient treatment outcomes. In multiple fields such as rare disease chordoma and early breast cancer diagnosis, this disease-grounded approach based on the underlying logic has achieved substantive breakthroughs.

Zhao Yu said he hopes to transform drug R&D from “serendipitous inspiration like a scientist’s genius” into “engineering-driven inevitability.” If life is self-consistent, and both diseases and health are encoded in DNA, then from a mathematical perspective, life can be encoded and interpreted.

Beijing Business Today reporter Wang Yinhao Song Yuying

A massive flow of information and precise interpretation—available in the Sina Finance app