Artificial Intelligence (AI) is often presented as a magical solution in healthcare: easing doctors’ workload, improving diagnostic accuracy, personalizing treatments, and enhancing service quality. However, in reality, AI technologies are still rarely integrated into hospital routines. Why? The answer lies in a recent study published in PLOS Digital Health, which involved in-depth interviews with 13 healthcare professionals, researchers, and policy experts (Leenen et al., 2025). This study sheds light on the real-world complexities of AI implementation in clinical practice.
1. No Real Demand, Just Tech Push: “Did we need it, or did we just build it because we could?”
One of the most striking findings is that many AI systems are developed based on what technology allows, not what the clinical environment truly needs. For example, a tool is developed for a patient group that already has sufficient diagnostic support. Since it doesn’t address a real unmet need, it fails to gain traction.
As one surgeon put it: “If the AI model produces too many false positives, we might end up with unnecessary surgeries. That risk makes me avoid using it.”
2. Plenty of Publications, But No Practical Use
Most AI projects are designed to generate academic publications. A model trained on thousands of images may be scientifically valuable, but to implement it in a hospital, you need to address data security, IT infrastructure, and legal compliance—areas often overlooked. As a result, many projects stay in the “great idea” phase.
3. Everyone Has Different Priorities: “There’s the nurse, the IT team, the legal team…”
Successful AI implementation requires collaboration between a wide range of stakeholders: clinicians, nurses, engineers, lawyers, administrators… But these people are rarely involved early enough. For instance, after a model is developed, someone may realize the hospital’s IT system can’t even support it. Or legal experts may raise red flags about compliance. Different stakeholders having different goals can derail the entire project.
4. Lacking Technical Skills: “Even the IT team doesn’t fully understand AI”
Hospital IT departments are typically focused on networking and data security, not on machine learning or algorithm integration. This skill gap leads to delays and added costs. Nurses and clinical staff are also often unaware of AI’s potential. Yet AI can significantly support nursing tasks and interdisciplinary care—if only the staff knew how to use it.
5. No Hospital-Level Strategy = Fragmented Projects
A hospital might have dozens of AI initiatives, but if they are disjointed, the overall impact is minimal. Hospitals must answer strategic questions like: “Which patient groups are a priority?”, “Are we developing our own models or using external ones?”, “Who assumes legal responsibility?” Without clear answers, even well-intentioned projects stall.
6. Legal Grey Areas: “Who’s liable when AI makes a mistake?”
One of the biggest roadblocks is the legal uncertainty surrounding AI. The EU’s AI Act and Medical Device Regulation (MDR) have introduced new rules, but these are evolving fast and are difficult to interpret. Doctors and administrators worry: if AI makes a wrong diagnosis, who’s accountable? Tracking these regulations is already difficult for policy experts—imagine the burden on frontline physicians.
Sample Scenario: The Night-Shift Doctor
It’s 3:00 AM. A sleep-deprived on-call doctor receives a radiology image from an assistant. An AI system highlights a potential brain hemorrhage. The doctor double-checks the image and feels reassured: “It gives me peace of mind,” they say. This scenario shows the value AI can bring—but only if the system is well-integrated and trusted. Reaching that point, however, is the hard part.
What Can Be Done?
- Assess Readiness: Hospitals should evaluate their AI maturity—data quality, existing systems, staff knowledge—and identify gaps.
- Define Your Role: Will your hospital develop AI in-house or only implement third-party solutions? This choice affects budget and hiring.
- Standardize Implementation: Develop SOPs that assess ethical, technical, and clinical feasibility. Create clear bridges between research projects and actual adoption.
- Foster Collaborative Learning: Encourage long-term, interdisciplinary collaboration among doctors, engineers, nurses, and legal teams. Use participatory approaches like design thinking or action research to build mutual understanding.
Conclusion
AI holds enormous potential to revolutionize healthcare. But this potential can only be realized through clear strategy, multidisciplinary collaboration, robust infrastructure, and legal clarity. Without these, AI remains a promising yet underused tool—gathering dust on the shelf rather than transforming patient care.
Reference:
Leenen, J. P., Hiemstra, P., Ten Hoeve, M. M., Jansen, A. C., van Dijk, J. D., Vendel, B., … & Hettinga, M. (2025). Exploring the complex nature of implementation of Artificial intelligence in clinical practice: an interview study with healthcare professionals, researchers and Policy and Governance Experts. PLOS Digital Health, 4(5), e0000847. https://doi.org/10.1371/journal.pdig.0000847
