Logic Explained Deep Neural Networks: A General Approach to Explainable AI

Although deep learning models are playing increasingly important roles across a wide range of decision-making scenarios, a critical drawback is their inability to provide human-understandable motivations for their opaque or complex decision-making processes. This so-called "black box" issue has hindered the deployment of deep neural networks in safety-critical and other domains such as industry, medicine or courts, where human experts and concerned parties naturally desire more insight into just how the machine is formulating its decisions. In the paper Logic Explained Networks, a research team from Università di Firenze, Università di Siena, University of Cambridge and Universitè Côte d'Azur proposes a general approach to explainable artificial intelligence (XAI) in neural architectures via interpretable deep learning models called Logic Explained Networks (LENs). The novel approach yields better performance than established white-box models while providing more compact and meaningful explanations. Previous research has shown that one possible way to provide human-understandable explanations is through the use of an expressive formal language such as first-order logic (FOL).