“Please, Vadalog, tell me why”: Interactive Explanation of Datalog-based Reasoning

Abstract

Integrating Large Language Models (LLMs) with logic-based Enterprise Knowledge Graphs (EKGs) and more generally with Knowledge Representation and Reasoning (KRR) approaches is currently at the forefront of research in many data-intensive areas, as language models may complement EKGs and ontological reasoning with flexibility and human orientation. Conversely, EKGs provide transparency and explainability on the conclusions drawn, a typical weak point of LLMs, which operate opaquely. In this demo, we integrate Llama 2 with our reasoning system Vadalog and use it to turn a chase graph, i.e., the trace of an ontological reasoning process, into a human-readable business report. In other words, we show the amazing capabilities of state-of-the-art LLMs in combination with a principled exploitation of the theoretical underpinnings of logic-based reasoning. We walk the audience through a visual environment, unfolding real-world reasoning settings from the Central Bank of Italy.