Play MNIST For Me! User Studies on the Effects of Post-Hoc, Example-Based Explanations & Error Rates on Debugging a Deep Learning, Black-Box Classifier

TL;DR

This study investigates how example-based explanations and error rates influence user perceptions of a deep learning classifier, revealing that explanations improve perceived correctness and higher error rates reduce trust.

Contribution

It provides empirical evidence on the effects of post-hoc, example-based explanations and error rates on user trust and perception of black-box classifiers.

Findings

Case-based explanations increase perceived correctness of misclassifications.

Error rates above 4% decrease trust and perceived reasonableness.

User perceptions are significantly affected by explanation type and error rate.

Abstract

This paper reports two experiments (N=349) on the impact of post hoc explanations by example and error rates on peoples perceptions of a black box classifier. Both experiments show that when people are given case based explanations, from an implemented ANN CBR twin system, they perceive miss classifications to be more correct. They also show that as error rates increase above 4%, people trust the classifier less and view it as being less correct, less reasonable and less trustworthy. The implications of these results for XAI are discussed.