Evaluating the Influences of Explanation Style on Human-AI Reliance

TL;DR

This study investigates the influences of feature-based, example-based, and combined feature- and example-based XAI methods on human-AI reliance through a two-part experimental study with 274 participants comparing these explanation style conditions.

Abstract

Explainable AI (XAI) aims to support appropriate human-AI reliance by increasing the interpretability of complex model decisions. Despite the proliferation of proposed methods, there is mixed evidence surrounding the effects of different styles of XAI explanations on human-AI reliance. Interpreting these conflicting findings requires an understanding of the individual and combined qualities of different explanation styles that influence appropriate and inappropriate human-AI reliance, and the role of interpretability in this interaction. In this study, we investigate the influences of feature-based, example-based, and combined feature- and example-based XAI methods on human-AI reliance through a two-part experimental study with 274 participants comparing these explanation style conditions. Our findings suggest differences between feature-based and example-based explanation styles beyond interpretability that affect human-AI reliance patterns across differences in individual performance and task complexity. Our work highlights the importance of adapting explanations to their specific users and context over maximising broad interpretability.

Figures (7)

• Figure 1: The three stages of the Judge-Advisor System (JAS) Framework.
• Figure 2: Experimental design of our study. Participants are be randomly allocated to one of four explanation-style groups after reading and agreeing to the consent form, remaining assigned to this condition for both Part 1 and Part 2. Before beginning each Part, participants are provided information about task followed by a short quiz to confirm their understanding.
• Figure 3: Equivalence of information as determined through combined explanations between an image (left) and its two nearest-neighbour example-based explanations (middle and right). Ovals of the same colour indicate features shared across the three images.
• Figure 4: Participants' average initial and final accuracy per experimental condition, relative to the accuracy of the AI.
• Figure 5: Participants' average initial and final accuracy per experimental condition split by task ability, relative to the accuracy of the AI. Error bars show standard error of the mean.