Bridging Explanations and Logics: Opportunities for Multimodal Language Models

As subsymbolic Artificial Intelligence (AI) systems have become increasingly integrated into decision support tools, there is a consequent need for transparent and interpretable models. In this sense, eXplainable AI (XAI) techniques offer insights into model behavior; however, they often lack the formal rigor –typical of symbolic AI– required for causal interpretation and verification. This paper presents a framework designed to bridge the gap between subsymbolic explanations and symbolic reasoning through the application of Multimodal Language Models (MLMs). By combining the output of XAI methods with symbolic knowledge bases encoded in logic programming languages, we enable abductive reasoning to yield causal interpretations of the explanations produced over a prediction. In this instance, MLMs serve as intersymbolic translators, converting visual or textual explanations into structured logical assertions that can be processed by inference engines or subsequent verification purposes. With this approach, we aim not only to enhance the interpretability of existing AI systems, but also to promote the use of sound reasoning as a means to increase the trustworthiness of the entire AI-based system. ... mehrIn this paper, we outline our proposed methodology, illustrating it with a running example, and discuss both future directions (including prompt optimization and ontology generation) and challenges (like hallucinations and scalability). Our work aims at contributing to the emerging field of intersymbolic AI, which calls for the integration of symbolic and subsymbolic paradigms in the pursuit of trustworthy AI.