Early Exploration of Generative AI in Nursing Education
Project Overview
In June 2024, following exposure to conversational AI applications at a professional conference, I explored whether generative AI could function as an interactive patient rather than simply as an information or writing tool.
I developed a simulated 55-year-old patient presenting with acute chest pain, cardiovascular risk factors, nausea, and shortness of breath. The AI was instructed to assume the role of “Mrs. Johnson” and respond only as the patient while a nursing student conducted an assessment and provided care.
The interaction allowed the learner to determine what questions to ask, what assessments to perform, what interventions to initiate, and what education to provide. The AI patient responded dynamically to the learner's actions rather than directing the learner through a predetermined sequence.
What I Was Exploring
The exercise examined whether generative AI could support a low-risk clinical rehearsal environment in which nursing students could practice:
Patient identification and introduction
Symptom and pain assessment
Medication administration and reassessment
Focused cardiovascular and respiratory assessment
Communication and therapeutic interaction
Patient education
Clinical decision-making
Recognition and correction of inaccurate information
One particularly important moment occurred when the student suggested that increasing dietary fat might benefit a patient with heart failure. The simulated patient accepted the incorrect education. This exposed an important limitation of unrestricted generative AI: a realistic patient simulation should not necessarily correct the learner, but the educational system must be capable of detecting unsafe or inaccurate clinical decisions during evaluation.
AI-Assisted Formative Evaluation
After the simulation, I changed the AI's role from patient to nurse educator and asked it to evaluate the learner's performance in communication, nursing process, assessment, patient education, and professionalism.
That experiment revealed both the potential and limitations of AI-generated feedback. Although the AI recognized several strengths, its initial evaluation awarded the learner 98/100 and failed to adequately penalize important clinical concerns.
For example, the student administered nitroglycerin without first verbalizing assessment of blood pressure or other safety considerations and provided inaccurate dietary education. The AI's overly favorable evaluation demonstrated that an educational simulation requires more than a conversational model—it requires a defined clinical framework, explicit safety rules, and a structured scoring rubric.
What I Learned
This early experiment helped shape my understanding of what an effective AI nursing simulation would require. The technology could convincingly portray a patient and sustain an unscripted interaction, but educational validity required a separate layer capable of distinguishing between:
what the patient should say, what the nurse should do, and how the learner's actions should be evaluated.
That distinction became foundational to my later thinking about AI-supported clinical rehearsal.