By Lisiane Pruinelli, PhD, MSc, RN, FAMIA, FAAN
A new academic year brings new energy, new ideas and new people. This year, it also arrives at a time when artificial intelligence continues to advance at an extraordinary pace.
Across healthcare, we are talking about increasingly sophisticated predictive models, generative AI, multimodal systems and agentic AI. The conversation is rapidly shifting from what AI can do to how we implement these systems safely, responsibly and meaningfully in real clinical environments.
But as we begin this academic year, I keep coming back to a much less glamorous question:
Do we actually have the informatics infrastructure to build the AI systems we keep talking about?
The Infrastructure Nobody Sees
It is easy to be impressed by what happens at the end of the AI pipeline: a prediction, a generated summary, a clinical recommendation or an intelligent agent.
What we rarely see is everything underneath.
AI depends on data that are accessible, meaningful, representative and sufficiently complete. It depends on standards that allow information to be interpreted consistently, systems that can communicate with one another, secure computational environments, reproducible data pipelines and mechanisms for validating models beyond the environments in which they were developed.
In nursing, many of these foundations remain uneven. Some of our most valuable information is still buried in clinical notes, flowsheets, local documentation practices and systems that do not easily communicate with one another. Even when the data exist, understanding what they actually mean requires knowledge of how care is delivered, how nurses document and how information changes as it moves through health information systems.
The sophistication of our AI cannot exceed the reliability of the infrastructure beneath it.
From AI Models to Trustworthy Systems
This summer reinforced this challenge for me.
At AINurse-26, our conversation deliberately moved beyond whether AI models can perform toward implementation, governance and what it takes to build trustworthy AI systems. We also put those ideas into practice. Together with workshop participants, the NAIL Collaborative launched the NAIL Digest: AI-Driven Nursing Informatics, a community-curated, AI-assisted resource designed to help the field keep pace with the rapidly growing nursing informatics literature. Participants helped shape the sources, topics and rules guiding how the literature is summarized, providing an example of building AI with a community rather than simply delivering AI to one.
Even a resource like this reminds us that trustworthy AI is never just about the model. It depends on the data sources, rules, oversight and people surrounding it.
Those broader conversations are also captured in our newly published report, Implementation, Governance, and Trustworthy AI in Nursing.
At the Nursing Knowledge: Big Data Science (NKBDS) Initiative in Minneapolis, we approached another part of the same challenge: how nursing knowledge and nursing data are represented, standardized, shared and ultimately implemented to improve health and healthcare.
Different initiatives, but a remarkably consistent message:
We cannot implement trustworthy AI without building the informatics infrastructure to support it.
Data governance is an important part of that infrastructure. Privacy, ethics, security and responsible data use must remain foundational. But governance also needs to evolve alongside the technologies it supports. Decisions about data access, linkage, reuse, provenance, sharing and validation directly shape what AI systems can be developed, how rigorously they can be evaluated and whether they can ultimately move into practice.
The challenge is not choosing between protection and progress. We need governance capable of enabling both.
Infrastructure Is Also People
There is another part of infrastructure that receives even less attention: people.
Building trustworthy AI requires more than data scientists or increasingly powerful models. It requires people who understand nursing knowledge, clinical data, terminology standards, care delivery, workflows, human factors, implementation, quality, policy and the realities of healthcare systems.
That is why I am particularly excited to begin this academic year by welcoming two new tenure-track faculty members to the UF College of Nursing.
Seo-Yoon Lee, PhD, MPH, RN, brings expertise in health informatics, standardized nursing terminologies, EHR data, workload modeling and clinical decision support. Amelia Schlak, PhD, RN, brings expertise in nursing resources, care delivery, healthcare quality and the system-level factors that shape patient outcomes.
Their areas of expertise are different, and that is exactly the point.
Building the future of nursing informatics does not mean assembling people who all build AI models. It means creating an ecosystem of complementary expertise capable of connecting data and technology with nursing knowledge, people, organizations and care.
What Are We Really Building?
Perhaps this is the question we should carry into the new academic year.
AI readiness should not be measured by how many AI tools an organization has adopted or how many models researchers have developed. A health system can have access to the most sophisticated AI available and still lack the infrastructure to use it safely and meaningfully.
Trustworthy AI requires trustworthy infrastructure.
Better data. Stronger standards. Interoperable systems. Modern computational environments. Governance that protects while enabling responsible use. Interdisciplinary partnerships. And people who understand how all of these pieces connect.
For nursing informatics, this is both a challenge and an opportunity. Our role is not simply to help nursing use the next generation of AI. It is to help build the knowledge, data, systems and environments that make responsible AI possible in the first place.
As we begin another academic year, perhaps the question is not simply what will AI be able to do next?
Everyone wants AI. The harder question is whether we are building what AI actually needs.