Abstract

Manual communication is often used during emergency medical situations, causing a delay in receiving the correct help. During an emergency, the individual needs to recognise the situation, tell someone where they are, determine which hospital would be best, organise an ambulance, and decide what first aid to give. Each of these can cause delays. Recent studies have looked into implementing machine learning, large language models (LLMs), and multi-agent systems into emergency situations in order to improve various parts of the emergency situation. However, most models only focus on certain tasks. Sixteen studies were reviewed between 2018 and 2026 covering emergency triage prediction, using large language models in multi-agent decision support, ambulance dispatch, ambulance booking, first-aid chatbots, and medical resource recommendation. Nine key studies were compared using a year-based survey matrix. Five main issues were found to be issues within the current systems. These include: limited scope, lack of capability in autonomous agents, lack of real-time integration between location services, hospitals, ambulance and emergency contacts, missing privacy, security and trust issues, and lack of validation in real-life scenarios. In order to overcome these issues this paper proposes an emergency taxonomy linking different types of emergencies to the right priority in first aid and capabilities in hospitals. It also proposes a way of prioritising resources using a scoring system based around how urgent the emergency is. A conceptual model of a five-module AI Emergency Response Agent was also proposed which has not yet been implemented. An appropriate evaluation protocol will be discussed in future work.

Keywords

How to Cite This Article

Rights