LLMs and Emergency Management: A Critical Analysis

The use of large language models (LLMs) for crisis preparedness and response, especially in multilingual communication, is growing. However, a recent study warns about the risks associated with using these tools in emergency contexts, where the perception of urgency is fundamental.

Distortion of Urgency in Automatic Translation

The research analyzed the performance of LLMs and machine translation systems in translating messages related to crisis situations. The results show that both dedicated translation models and LLMs tend to worsen and show instability in rendering the original message. A critical aspect is that even linguistically adequate translations can distort the perception of urgency, compromising the effectiveness of communication and triage operations.

Risks in Using General-Purpose Language Technologies

The study highlights how urgency classifications based on LLMs vary significantly depending on the language of the prompt and input. These findings underscore the risks in using general-purpose language technologies for communication in crisis situations and the need for specific evaluation frameworks for these contexts.