LATIDIA · Investigación

Auditoría de la confiabilidad de la escalada en agentes de LLM de atención médica: un punto de referencia controlado por la perturbación de las fallas del estado del flujo de trabajo

arXiv:2610.21270v1 Tipo de anuncio: nuevo Resumen: La IA biomédica confiable requiere una selección de acción confiable cuando falta evidencia, está obsoleta, es conflictiva o tardía, no solo respuestas precisas en entradas estáticas. Instantia

La noticia

arXiv:2610.21270v1 Announce Type: new Abstract: Trustworthy biomedical AI requires reliable action selection when evidence is missing, stale, conflicting, or late, not only accurate answers on static inputs. We instantiate an episode-level evaluation protocol as 400 synthetic healthcare workflow episodes: 40 hand-authored semantic seeds across four task classes, two surface parameterizations, and five controlled variants, scored on ANSWER/ASK CLARIFICATION/ABSTAIN/ESCALATE under a pre-declared 5:1 missed-versus-unnecessary escalation cost. Reference actions are single-author study-policy labels, not clinician-adjudicated ground truth. Across five open-weight configurations (1.7B--9.7B), clean decision accuracy reaches 0.97, yet uncertainty-prompted models miss 32--97% of required escalations. The stronger 9.7B configuration narrows but does not remove the gap, falling from 0.97 clean accuracy to