Institute of Formal and Applied Linguistics

In this talk, we will start from the general motivation behind Explainable Artificial Intelligence (xAI) and point out the advantages of using natural language explanations (NLEs). However, NLEs, as well as other xAI methods, face difficulties in providing both plausible and faithful explanations, i.e. explanations that are trusted by users and reflect the real decision-making process of the AI model. We will illustrate this problem for two NLE methods for item recommendation and image classification tasks. In addition, we will explore simple NLE techniques for improving faithfulness while maintaining high plausibility.

*** The talk will be delivered in person (MFF UK, Malostranské nám. 25, 4th floor, room S1) and will be streamed via Zoom. For details how to join the Zoom meeting, please write to sevcikova et ufal.mff.cuni.cz ***