RIXA - Real-time Interactive eXplainable AI

Short description of the project

RIXA (Real-time Interactive eXplainable AI) is a natural language dashboard that can explain AI models in a way that is understandable for end users. It combines methods from the field of explainable AI with large language models (LLMs). As a result, explanations that are typically highly technical can be made accessible to laypeople.

Project objectives

AI is increasingly used in production, whether in assembly assistance, shift planning, or threat detection. This broadens the user base of AI systems, including more people without technical expertise interacting with these systems. In work contexts, the European AI Regulation also applies, which mandates that certain AI systems are made transparent and comprehensible. Transparency and explainability can additionally help increase acceptance of AI systems. Existing explanation methods often fail because they are too technical and cannot be understood by non-experts; they are at the moment mostly used in model development.

Project results

RIXA offers users a chat through which they can communicate with the system in natural language. Here, they can ask questions about the underlying AI system and how it functions. The chat is powered by an LLM, which, however, does not respond freely but based on function calls and background knowledge. The functions and background knowledge can be adjusted as needed for different application areas, enabling RIXA to support a wide range of applications.