Description
Problem. The mining industry's Vision Zero commitment, pursuing continual improvement with the goal of zero harm depends on accurate data. Effective safety management requires accurate incident and risk classification, near-misses reported accurately, root causes and corrective actions attributed correctly. When workers in operations lack safety expertise or face time pressure, events are frequently misclassified. If 20% of events are incorrectly classified, 100% of the prevention budget gets allocated based on only 80% accurate information, systematically misdirecting resources away from actual risk patterns and undermining critical control effectiveness. Conversational AI appears to solve this knowledge gap, workers can describe events naturally, receiving help with the classification form the AI Agent.
Approach. This paper argues that conversational safety assistants must be engineered systems, not standalone chat interfaces. Drawing from our deployment experience across Latin American mining and industrial operations, the building blocks required for trustworthy implementation, must include knowledge bases containing approved company classification taxonomies, organizational procedures for reference, structured databases encoding relationships, so the system understands the operational context. These systems must understand the processes and workflow enabling user guidance through classification steps with mandatory verification checkpoints, connecting to enterprise safety management systems. We need to ensure that AI retrieves the right information for any specific question.
Boundaries of use. This model is aligned with Vision Zero principles, guiding workers through classification taxonomies, explaining control requirements, and pre-populating structured reports including recommendations on risk ratings. This requires human revision and approval. According to Deloitte reliance on unverified outputs remains a critical vulnerability, underscoring the urgent need for Human-in-the-Loop (HITL) protocols.
Contribution. The presentation will offer a practical review of important topics for managers in building safety AI assistants, aligned to improve data quality that drives prevention investments toward zero harm.