Evaluating the Interpretability of Tooth Expressions

In Knowledge Representation, Tooth expressions have been shown to behave like linear classification models. Thus, they can provide a powerful yet natural tool to represent local explanations of black box classifiers in the context of Explainable AI. In this extended abstract, we present the result of a user study in which we evaluated the interpretability of Tooth expressions compared to Disjunctive Normal Forms (DNF). In the user study, we asked respondents to carry out two classification tasks using concepts represented either as Tooth expressions or as different types of DNF formulas. We evaluated interpretability through accuracy, response time, confidence, and perceived understandability by human users. In line with our hypothesis, the study revealed that Tooth expressions are generally faster to use, and that they are perceived more understandable by users who are less familiar with logic.