Assessing LLM-mediated Formal and Informal Safety Communication Methods to Enhance Safety Performance for Construction Workers

Abstract

Safety communication substantially affects construction workers’ cognitive processing, comprehension, and behavioral outcomes. Persistent workforce constraints such as literacy, language diversity, and limited range of safety communication practices that reduce the effectiveness of traditional safety communication. Prior research studies have distinguished between informal and formal safety communication, however, there are limited experimental studies that examine these communication styles within an artificial intelligence mediated setting. Therefore, this study conducted an evaluation on whether formal and informal large language model mediated safety communication agents result in statistically significant differences in hazard recognition performance, safety attitude, and perceived usefulness. This study used an exploratory mixed-condition quasi-experimental study design guided by communication accommodation theory and cognitive load theory, with participants being randomly assigned to a linguistic distinct safety communication agent transmitting equivalent safety information from public OSHA regulations and rules. Twenty-eight construction professionals ranging from laborers to safety directors participated in this study. The participants were assigned to either an informal or formal communication condition and evaluated on the following survey constructs; perceived clarity, perceived usefulness, perceived cognitive load, trust, engagement, behavioral intention, perceived authority, safety attitude, hazard recognition, and risk perception. The results demonstrated no statistically significant differences between the two communication conditions for perceived hazard recognition, safety attitude, and perceived usefulness. However, at the 90 percent confidence level, differences were observed for perceived authority and cognitive load. An exploratory analysis was conducted based on exposure duration, and it indicated significant differences in perceived clarity and hazard recognition. These findings suggest AI-mediated safety communication platforms can support construction safety communication while incorporating accessible features such as for literacy and language barriers.