Rather a nurse than a physician - Contrastive explanations under investigation

Ilias Chalkidis, Oliver Eberle, Laura Cabello, Stephanie Brandl

Keywords: Natural Language Processing, Explainability, Contrastive Explanations, Human-centered AI

TL;DR: We compare contrastive explanations with non-contrastive explanations to investigate the claim that the former is more plausible for humans

Abstract: Contrastive explanations pose an interesting alternative to non-contrastive explanations from a human-centered perspective. They have been claimed to be closer to how humans explain a decision. We carry out a comparison between contrastive and non-contrastive explanations provided by models and humans for a text classification task to investigate this claim for the first time.

Submission Number: 5