DEAR EDITOR,

We thank the authors for their insightful letter regarding our review on artificial intelligence (AI) in medicine. We strongly agree with their premise that the macro-level ethical and regulatory frameworks we discussed must be actualized directly at the bedside and within clinical training programs, and we are pleased to witness these discussions arise more and more in the field.

In our review, we highlighted the black-box nature of AI models, pointing out that the internal reasoning behind an AI system’s outputs often cannot be readily interpreted even by experts.¹ The letter excellently articulates the clinical manifestation of this opacity: the “epistemic placebo” generated by fluent but potentially inaccurate language models.² As the authors astutely point out, using the shared example of pulmonary function test interpretations: while rule-based models encode established guidelines, generative models lack such explicit rule sets. In evaluating AI applications for complex respiratory conditions such as interstitial lung diseases or obstructive syndromes, robust biostatistical and methodological scrutiny is of utmost importance to differentiate true diagnostic utility from mere generative linguistic confidence.

Furthermore, our article emphasized that the integration of AI into medicine is profoundly shaped by neoliberal logic, which prioritizes cost-effectiveness, performance-based metrics, and output targets over collective well-being.¹ This systemic pressure trickles directly and indirectly down to medical education. When clinical and educational environments demand high efficiency, the unearned reassurance provided by generative AI becomes a highly attractive, yet perilous, shortcut for overworked residents and students. If trainees consult these tools for complex physiological and pathological mechanisms or differential diagnoses they do not yet fully comprehend, they risk profound deskilling and automation bias, causing them to become copycats of whichever AI model they use. This process risks exacerbating the individualization of responsibility that neoliberal frameworks often promote, blaming the individual trainee for a structural failure in the educational architecture rather than the system that produces it.

Therefore, the framework proposed in the letter–epistemic authority, calibrated trust, and an architecture of responsibility–is a vital addition to modern medical education.² We fully support the practical recommendations provided.

Ensuring that AI contributes to justice, inclusivity, and patient safety requires transparent clinical workflows and educational environments that model explicit accountability. When the faculty uses AI transparently and teaches the methodological limits of these tools to their trainees, a culture of critical appraisal is passed down. We share the conviction that clinicians and trainees must be equipped to rigorously appraise their sources, ensuring that we utilize these powerful tools to advance health outcomes rather than being led blindly by them.