Researchers have developed an AI system for triaging eye-related medical issues over the phone without requiring human annotation. The system, called Guideline-as-Oracle (GAO), uses a table of operational rules compiled from American Academy of Ophthalmology guidance to train a model that can provide accurate diagnoses and recommendations. In tests, GAO-Triage outperformed other systems on both metrics, with an improvement in agreement with a reference standard from 61.7% to 74.1%. The system's success is attributed to the use of guideline-derived assignment rather than dialogue surface form. --- Why it matters: This matters because it shows that AI can be trained without relying on expensive and time-consuming human annotation, which is often a bottleneck in developing medical chatbots. This could lead to more widespread adoption of such systems in healthcare settings. Source: https://arxiv.org/abs/2608.04772