Where AI Genuinely Adds Value

Let’s be specific, because the debate deserves specificity.

AI-powered natural language systems can generate adaptive assessment questions, provide immediate formative feedback on written responses, and personalise study pathways based on demonstrated knowledge gaps. A medical student who receives targeted feedback on a pharmacology question at the moment of error learns more efficiently than one who waits for a scheduled tutorial. These are real educational gains.

AI-driven virtual humans — conversational agents capable of simulating patient interactions — offer genuine value for communication skills training. Practicing high-stakes conversations is one of the most underdeveloped areas of clinical education. Breaking difficult news, navigating a distressed family, disclosing a medical error: these conversations require rehearsal, and they are almost impossible to rehearse safely in real clinical settings. Research on simulation-based communication training consistently shows improvements in clinician confidence and patient-reported experience (Kissane et al., 2012; Fallowfield et al., 2002). An AI-driven virtual patient that responds dynamically to what a resident actually says — rather than following a scripted path — raises the fidelity of that training meaningfully.

AI can also do something human educators cannot at scale: track. It can monitor which anatomical structures a student is spending time examining, flag knowledge gaps in real time, and surface patterns across an entire cohort that would take weeks for a faculty member to identify manually. When learning analytics are genuinely intelligent — not just dashboards — they change what is possible for curriculum teams managing hundreds of students.

None of this is nothing. It is, in fact, quite a lot. And any honest evaluation of these technologies has to say so.

Where AI Falls Short — and Why It Matters in Clinical Contexts

The same honesty that acknowledges AI’s strengths requires acknowledging its limits. And in clinical education, those limits have consequences.

AI systems do not build procedural memory. They cannot develop the spatial reasoning that comes from physically navigating a three-dimensional anatomical structure. They do not replicate the physiological response of embodied rehearsal — the recalibration that happens when a learner has, in a meaningful sense, been somewhere and done something. These are not gaps that better models will close. They are categorical differences between an information technology and an experiential one.

Then there is the hallucination problem — and in a clinical education context, this is not a minor inconvenience.

AI language models produce confident, fluent, incorrect information with regularity that the field has not yet resolved. A 2023 study evaluating ChatGPT’s performance on medical licensing exam questions found that while accuracy rates showed promise for general knowledge retrieval, errors were often presented with the same confidence as correct answers (Kung et al., 2023). In anatomy education, a student who receives an incorrect description of a vascular landmark from an AI system is not just misinformed — they are potentially acquiring a dangerous clinical belief with high confidence. That is a patient safety issue, not a product limitation.

There is also something more fundamental. AI is non-human. It does not get tired, anxious, or uncertain. It does not know what it feels like to stand in an operating room for the first time, or to deliver a difficult diagnosis to a parent. Its simulations of human interaction are increasingly sophisticated — but they remain simulations of language, not of experience. In a profession where the human dimension of care is inseparable from clinical competency, that gap is real.

Where VR Falls Short — and What That Means for Investment

VR has its own barriers, and a CXO who isn’t examining them isn’t doing proper due diligence.

Implementation complexity is real. Deploying an immersive learning environment across a large institution requires infrastructure thinking — device management, content governance, faculty training, LMS integration, and a plan for keeping the content library current. Institutions that have purchased VR hardware without an accompanying content and curriculum strategy have, in some cases, found expensive headsets sitting unused in storage rooms.

The no-code content creation shift has materially changed this equation. Platforms that allow faculty to build, update, and manage VR modules without technical support remove the dependency on specialist developers and close the gap between curriculum evolution and platform capability. But the procurement question — is this platform genuinely faculty-led, or does every content update require a support ticket? — remains critical.

Scalability without hardware dependency is the other factor that has shifted the field. Desktop access that extends the full content library to learners without a headset removes the hardware ceiling from the equation. A student who cannot access a headset on a given day still has access to the learning environment. That is not a compromise — it is how modern immersive platforms should be built.

The Question That Should End Every Evaluation Meeting

At this point in the evaluation process, many healthcare education leaders arrive at what feels like an either-or decision. Budget is finite. Both technologies have compelling use cases. Which one wins?

We want to challenge that framing directly — not to avoid the question, but because the framing itself produces the wrong answer.

If a health system deploys AI-based learning tools without immersive simulation infrastructure, it gains adaptive content delivery and improved knowledge assessment. What it does not gain is procedural competency, spatial reasoning, or the embodied clinical confidence that transfers to patient care. The training gap does not disappear. It becomes invisible, until it surfaces as a clinical error.

If a health system deploys VR without an AI layer, it gains exceptional experiential simulation — but delivers the same experience to every learner regardless of where they are in their development. The platform is powerful but undifferentiated. It cannot adapt, cannot personalise at scale, and cannot tell you where the curriculum is failing individual learners.

The business case for both technologies is not additive. It is multiplicative. The value of each is significantly greater when the other is present. That is the argument Issue Three will make in full — with the outcome evidence to support it.