Authors: Alfred Castillo, Arturo Castellanos, Jeffrey Parsons, Monica Chiarini Tremblay, Roman Lukyanenko, Veda C. Storey
Tags: 2020, AI, conceptual modeling, explainable AI, machine learning, XIA
Machine learning has become almost synonymous with Artificial Intelligence (AI). However, it has many challenges with one of the most important being explainable AI; that is, providing human-understandable accounts of why a machine learning model produces specific outputs. To address this challenge, we propose superimposition as a concept which uses conceptual models to improve explainability by mapping the features that are important to a machine learning model’s decision outcomes to a conceptual model of an application domain. Superimposition is a design method for supplementing machine learning models with structural elements that are used by humans to reason about reality and generate explanations. To illustrate the potential of superimposition, we present the method and apply it to a churn prediction problem.
Pre-print available:
