Biblioteca Central EsSalud

Viewed through a relational lens, artificial intelligence (AI) has entered the clinical space as a copilot — a computer system that informs decision making by means of human-like interactions. AI copilots augment clinicians’ skills in the service of patient care. Indeed, systems based on large language models now draft clinical notes, generate accurate differential diagnoses, and propose management plans with increasing fluency. But this partnership poses a risk: the more capable the AI copilot appears, the more likely clinicians are to offload tasks to it, eroding the very skills they need for detecting when it’s wrong. The deskilled clinician — and especially the never-skilled or mis-skilled learner — is left exposed to AI’s fluency traps and silent failures. This “automation paradox” is a human hazard, compounding AI limitations such as confabulations, susceptibility to misinformation, and sycophancy. How can clinicians realize AI’s promise while avoiding its perils, and how should educators prepare trainees to take control when the copilot fails?