- Project number: F 2536
- Institution: Federal Institute for Occupational Safety and Health (BAuA)
- Status: Completed Project
Description:
This research project investigated the potential of Artificial Intelligence (Al), specifically Large Language Models (LLMs), to support risk assessment in occupational safety and health (OSH). Risk assessment is the central tool for ensuring workplace safety but poses significant challenges, particularly for small and medium-sized enterprises (SMEs), due to resource and expertise constraints. The project aimed to determine if AI can assist experts in this complex task without compromising safety standards.
The research focused on designing and testing an Al-based assistance system grounded in German OSH regulations. The methodology followed a Human-Centered Design approach: First, the specific needs of safety professionals were identified through expert interviews. Based on this, practical scenarios for human-AI interaction were developed in workshops. A core component was the implementation of a functional software prototype utilizing "Retrieval Augmented Generation" (RAG) technology. This ensures the AI generates answers derived from a trusted database of regulations (e. g. Technical Rules). Finally, the prototype was evaluated in a study with 28 experts.
The results demonstrate that AI can effectively support the creation of risk assessments. Experts accepted the system as a competent "sparring partner" that helps identify hazards and proposes safety measures. The prototype proved that technically sound responses are achievable when the AI is subjected to strict constraints.
In conclusion, the use of AI in OSH is feasible and beneficial, provided the system operates transparently and humans retain final decision-making authority. The technology transforms the workflow from laborious content creation to the efficient verification of proposals, thereby helping to make expert safety knowledge more widely accessible.