AI in healthcare: making information more accessible without compromising trust

I've recently had the chance to join ELO Digital Office's ELO Expert Talks podcast (in Spanish) to discuss how artificial intelligence is being applied in healthcare and, more importantly, what it takes to use it responsibly in such a critical environment.

The conversation covered a wide range of topics — from AI hallucinations and the importance of medical evidence to the security of hospital information, regulation and the role of European technology. I found the discussion particularly useful because it brought together several ideas that we think about every day at Elna.

I've since taken the opportunity to structure some of those thoughts into this article. The central question is simple: how can we make information more accessible with AI without compromising the trust, evidence and safeguards that healthcare requires?

More information does not automatically mean better information

Generative artificial intelligence is changing how we access information. In healthcare, however, making information easier to find is only part of the challenge. The more important question is whether clinicians can understand where an AI-generated answer comes from, assess the evidence behind it, and decide how much they can rely on it.

This is particularly important in medicine, where information is not simply information. It can influence clinical decisions and, ultimately, patient care.

Healthcare professionals already work with an enormous and constantly growing body of knowledge: scientific publications, clinical studies, systematic reviews, meta-analyses and clinical practice guidelines.

Generative AI makes it possible to interact with this information in a much more natural way. Instead of navigating multiple databases and documents, a clinician can ask a question in plain language and receive a concise answer in seconds.

That accessibility is valuable. But it also introduces a new risk.

AI systems can generate answers that sound convincing while containing errors or information that is not adequately supported by evidence. These are commonly referred to as AI hallucinations. In a healthcare context, even a very low probability of error deserves careful consideration.

The answer, therefore, is to use AI in a way that makes its limitations visible and verification straightforward.

In medicine, the source matters as much as the answer

Medical knowledge has different levels of evidence. An individual case report, a clinical study, a systematic review and a clinical practice guideline do not carry the same weight.

This distinction matters when an AI system answers a clinical question.

A useful clinical AI system should therefore do more than produce a response. It should help the clinician understand what evidence supports that response and provide a direct path to the original sources.

At Elna, this principle is central to how we approach AI for healthcare: traceability should be part of the user experience, not an afterthought.

When a clinician can inspect the underlying evidence, AI becomes a tool for navigating medical knowledge rather than a black box that simply asks to be trusted.

Trust requires critical thinking — and system design

Responsible use of AI is not only a question of technology. It is also a question of how people use it.

Clinicians, like all users, need to understand that an AI-generated answer is not automatically a verified fact. Critical thinking remains essential: checking the evidence, considering the context and applying professional judgment.

Technology should make this process easier, not harder.

That means clearly communicating the limitations of AI, making sources accessible, and designing systems that encourage verification. The goal is to give healthcare professionals better tools for exercising clinical judgment, not to replace it.

The same principle applies to hospital information

The challenge becomes even more interesting when we move from external medical knowledge to the information generated inside a hospital.

Hospitals are complex organizations with large volumes of information: policies, procedures, quality documentation, clinical protocols, operational processes and many other documents. Different professionals need access to different information, according to their roles and responsibilities.

Generative AI can make this information much easier to find and use. A clinician or employee could ask a question in natural language instead of searching through multiple repositories and documents.

But accessibility must go hand in hand with security, governance and access control.

An AI assistant for a hospital has to do more than retrieve information. It should respect the organization's existing permissions and answer from trusted, structured sources.

This is where the combination of secure information management and AI becomes particularly important. If the underlying information is well governed, access is controlled and the sources can be traced, generative AI can become a practical interface for navigating complex organizational knowledge.

Building AI on trust

Healthcare is one of the areas where the consequences of an incorrect answer can be significant.

The objective should therefore be different: make useful information easier to access while preserving the controls that make healthcare information trustworthy.

For us at Elna, this means focusing on three connected principles:

• Evidence: AI-generated medical information should be grounded in reliable sources.

• Traceability: clinicians should be able to inspect and verify those sources.

• Security and governance: sensitive healthcare information must remain protected and accessible according to clearly defined permissions.

These principles are particularly relevant as European healthcare organizations adopt AI within an increasingly structured regulatory environment. Regulation, data security and technological innovation should not be seen as opposing objectives. They are different parts of the same challenge: developing AI that can be used responsibly in environments where the consequences of error matter.

The future of AI in healthcare will not be defined simply by how intelligent an AI system appears to be. It will also depend on whether healthcare professionals can understand, question and verify what it tells them.

For Elna, that is what trustworthy AI in healthcare should mean: not asking clinicians to trust the machine, but giving them the information and tools they need to make informed decisions about when to trust it.