Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Recurso digital |
| Lenguaje: | |
| Publicado: |
Zenodo
2026
|
| Materias: | |
| Acceso en línea: | https://doi.org/10.5281/zenodo.19269540 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
Tabla de Contenidos:
- <p>Medical students use AI at the exact moments their cognitive load is highest and their <br>supervision is lowest. Patient presentations are the surface where that use is most concentrated <br>and least governed. Written work can be screened; spoken work cannot. A student who builds an <br>entire presentation with AI delivers it in their own voice, sounds competent, and is assessed on a <br>performance the system produced. No current detection tool can see that. This case study <br>examines a single, high-stakes clinical vignette — a 45-year-old with crushing chest pain — <br>presented word-for-word to four AI systems. All four collapsed the differential at Turn 1 and <br>moved toward aspirin, a treatment that can kill a patient with aortic dissection. The Meaning <br>Audit Protocol (MAP) was applied to each interaction record and produced consistent, <br>classifiable findings across four different architectures. This document presents the complete <br>MAP audit for CCA-ISF-02. It shows how the failure emerges at the interaction level, how it is <br>internalised by students as clinical knowledge, and why responsibility sits with the institutions <br>that had the distance and authority to govern these tools and did not. </p>