Medical Korea 2026: Smart Hospitals Face Real-World Barriers in the Age of Medical AI

Medical Korea 2026 convened policymakers, clinicians, hospital leaders, and technology developers at a time when artificial intelligence is rapidly moving from experimental pilots into clinical workflows. Across global health systems, AI is increasingly positioned as a solution to mounting pressures—from physician burnout and workforce shortages to rising healthcare costs and the growing complexity of data-intensive medicine. Yet as enthusiasm accelerates, so too does a more difficult question: what does it actually take to operationalize AI inside hospitals at scale?

One discussion at the conference explored this transition from ambition to implementation, focusing on how “smart hospitals” are being built in practice. Professor Kim Dai-jin of Seoul St. Mary’s Hospital offered a detailed and at times unfiltered account of both progress and persistent barriers, drawing on his role leading digital transformation across one of South Korea’s largest hospital systems.

Rather than centering on algorithmic breakthroughs, his presentation reframed the conversation around systems—data architecture, infrastructure capacity, regulatory environments, and the evolving role of clinicians. In doing so, it underscored a broader reality emerging across global healthcare: the success of medical AI will depend less on technological capability alone and more on whether institutions are structurally prepared to support it.

Kim Dai-jin emphasized that scaling smart hospitals depends not only on algorithms, but on unified data systems, regulatory clarity, and clinical trust. Featured. Image: GeneOnline

Building a Data Empire: Scale as the Foundation of AI

To ground the discussion, Kim began with the scale and structure of the Catholic Medical Center (CMC), a network that illustrates both the opportunities and prerequisites of data-driven healthcare. Comprising eight hospitals under a unified governance model, the system operates approximately 6,400 beds—representing close to 9% of the nation’s total bed capacity—and manages roughly five million outpatient visits each year. This volume, he explained, continuously generates a dense and longitudinal stream of clinical data that becomes foundational for any meaningful AI deployment.

What distinguishes CMC is not only its size but its standardization. While many hospital networks—both in Asia and globally—operate on fragmented IT systems that limit interoperability, all eight CMC hospitals share an identical database structure. This alignment allows data to be aggregated, compared, and analyzed across institutions without the need for extensive harmonization, a challenge that often slows or prevents AI adoption elsewhere.

“Without integrated data at this level, most AI models simply cannot function in real clinical environments,” Kim said.

Over the past seven years, this approach has resulted in a dataset encompassing more than 15 million patients. The effort reflects a longer-term strategic shift in healthcare, where institutions are increasingly positioning themselves not only as care providers but also as stewards of large-scale, structured medical data capable of supporting advanced analytics and AI development.

The Clinical Data Warehouse: Turning Information into Insight

At the center of this transformation is CMC’s Clinical Data Warehouse (CDW), designed to unify diverse data types into a single, queryable environment. The system integrates diagnosis codes, prescriptions, laboratory results, vital signs, imaging data from PACS systems, and next-generation sequencing (NGS) outputs—creating a comprehensive longitudinal view of patient health.

This level of integration reflects broader global efforts to move beyond siloed datasets toward interoperable health data ecosystems, which are widely seen as essential for training robust AI models. However, Kim emphasized that data accumulation alone is insufficient; usability and accessibility are equally critical.

“We don’t just store data—we make it usable,” he said. “That’s where innovation begins.”

To that end, CMC has extended its infrastructure beyond internal use, allowing external AI companies to access anonymized datasets through a secure cloud-based analysis platform. This model of controlled openness mirrors a growing trend in healthcare systems seeking to balance data privacy with the need for collaborative innovation, particularly as AI development increasingly relies on diverse and large-scale datasets.

AI at the Bedside: From Diagnostics to Predictive Care

With this infrastructure in place, Kim outlined how AI is being applied across clinical settings, illustrating a shift from theoretical potential to operational use. In pediatric care, for instance, an AI model analyzes complete blood count (CBC) data to assist in identifying early indicators of leukemia. By referencing similar historical cases within the system, it provides physicians with additional context for diagnosis and treatment planning.

In pathology, digital models are being developed to classify multiple subtypes of lung cancer from biopsy slides, calculating the proportional distribution of each subtype with a level of granularity that is difficult to achieve through manual review alone. Such applications reflect a broader trend in AI-assisted diagnostics, where computational tools augment human interpretation rather than replace it.

Predictive analytics represents another expanding frontier. By leveraging up to 20 years of longitudinal health check-up data, CMC’s systems can estimate an individual’s risk of developing chronic conditions such as diabetes or hypertension within a defined time horizon. These capabilities point toward a gradual shift from reactive treatment to preventive and anticipatory care models.

At the same time, AI applications are extending beyond traditional hospital environments. A hospice support chatbot, for example, enables terminally ill patients to manage symptoms and medication inquiries at home while maintaining a connection to clinical and support teams. This reflects a broader movement toward decentralized care, where digital tools help extend clinical oversight beyond hospital walls.

Kim also highlighted how regulatory constraints can shape the trajectory of innovation. A brain CT AI system initially designed for human diagnostics was redirected toward veterinary applications, where regulatory barriers are lower, ultimately achieving commercial success.

“Regulation for human applications is so strict that sometimes innovation finds its way through other markets first,” he noted.

The “Wall”: Where Innovation Meets Reality

While these examples demonstrate tangible progress, Kim devoted significant attention to the systemic challenges that continue to limit widespread adoption. Central among these is the issue of liability, particularly in high-stakes clinical scenarios where AI outputs may influence decision-making.

He described a case in which an AI monitoring system indicated normal patient vitals shortly before a fatal cardiac event, raising unresolved questions about accountability.

“If the AI says ‘normal’ and the patient dies, who is responsible?”

Such scenarios highlight the absence of clear legal frameworks governing AI-assisted care, an issue that extends beyond South Korea and remains a topic of debate across global healthcare systems.

Infrastructure constraints present another layer of complexity. Advanced AI models require substantial computational power, often relying on high-performance GPU clusters that significantly increase energy consumption. For hospitals operating within existing physical and financial constraints, scaling such systems can require major upgrades.

“To fully implement this, we would need to rebuild our entire power infrastructure,” Kim said, noting that the associated costs can reach into the millions.

At the same time, international comparisons reveal uneven progress. In the United States, platforms such as Epic Systems have begun integrating large language models directly into clinical workflows, with early reports suggesting significant reductions in physician documentation time.

“We are still trying to integrate basic AI, while others are already improving productivity at scale,” Kim observed.

Regulatory developments add further complexity. South Korea’s forthcoming AI Framework Act classifies medical AI as “high-impact,” subjecting it to stricter oversight. While such measures aim to ensure patient safety, they may also increase administrative burdens for clinicians.

“Doctors may spend more time on documentation and compliance than on patients,” he said.

Together, these challenges illustrate a broader tension shaping the future of medical AI: the balance between innovation, safety, and operational feasibility.

Ethics as Infrastructure: A Human-Centered AI Future

In concluding, Kim turned to the ethical dimensions of AI integration, emphasizing that technological advancement must remain aligned with core principles of patient care. Drawing on guidelines developed in collaboration with the Vatican, he outlined a framework centered on preserving the human role in medicine.

At the heart of this approach is the concept of AI as a “co-pilot,” designed to support rather than replace clinicians. The goal, he suggested, is not automation for its own sake, but augmentation that enhances clinical judgment and strengthens the doctor–patient relationship.

“AI must enhance the doctor–patient relationship, not weaken it.”

He also addressed the question of data ownership and value distribution, arguing that benefits derived from patient data should extend beyond commercial entities to include the individuals and communities that contribute that data.

Transparency remains a critical component of this framework. Physicians must be able to interpret and explain AI-generated insights to patients, ensuring that trust is maintained even as decision-making processes become more complex.

Why This Matters: Medical Korea’s Broader Signal

The discussion reflects a broader shift taking place at Medical Korea 2026, where conversations are increasingly moving beyond the promise of AI toward the realities of implementation. As healthcare systems worldwide grapple with similar challenges—data fragmentation, infrastructure limitations, regulatory uncertainty, and workforce adaptation—the concept of the “smart hospital” is emerging not as a singular technology, but as a multifaceted transformation.

In this context, AI becomes one component within a larger system that must evolve in parallel. The ability to integrate data, manage risk, support clinicians, and maintain patient trust will ultimately determine how—and how quickly—these technologies translate into meaningful clinical impact.

Kim’s presentation, while grounded in the experience of a single hospital network, captures a set of issues that resonate globally. It suggests that the future of medical AI will be shaped not only by innovation, but by the systems that enable—or constrain—it.

Author

Bernice Lottering