XAI in Automated Fact-Checking? The Benefits Are Modest and There's No One-Explanation-Fits-All

TL;DR

It is suggested that XAI has limited effects on users’ agreement with the veracity prediction of the automated fact-checker and on their intent to share news, however, XAI nudges users towards forming uniform judgments of news veracity, thereby signaling their reliance on the explanations.

Abstract

The massive volume of online information along with the issue of misinformation has spurred active research in the automation of fact-checking. Like fact-checking by human experts, it is not enough for an automated fact-checker to just be accurate, but also be able to inform and convince the user of the validity of its predictions. This becomes viable with explainable artificial intelligence (XAI). In this work, we conduct a study of XAI fact-checkers involving 180 participants to determine how users' actions towards news and their attitudes towards explanations are affected by the XAI. Our results suggest that XAI has limited effects on users' agreement with the veracity prediction of the automated fact-checker and on their intent to share news. However, XAI nudges users towards forming uniform judgments of news veracity, thereby signaling their reliance on the explanations. We also found polarizing preferences towards XAI and raise several design considerations on them.

Figures (10)

• Figure 1: The XAI conditions for the automated fact-checker. The fact-checker reports, each with a "How to understand this?" guide, are based on the TRUE news shown in Figure \ref{['fig:NewsPrediction']}. The LIME, SHAP and Attention conditions display similar data with a graphical visualization while the Evidence and Comments conditions use a textual visualization.
• Figure 2: News that received a TRUE or FALSE prediction by the automated fact-checker. These visualizations were used as Control for our experiment.
• Figure 3: Chart (a) shows the effects of XAI Condition on Agreement. Agreement scores are between 1 (strong disagreement) and 4 (strong agreement) with the averages shown. Chart (b) shows the variance in the agreement scores for each condition. Error bars show .95 confidence intervals. *: $p$<.05, **: $p$<.01.
• Figure 4: Interaction of XAI Condition and News Veracity on Agreement. Scores are between 1 (strong disagreement) and 4 (strong agreement) with the averages shown. Error bars show .95 confidence intervals.
• Figure 5: Frequencies (n) of the subcodes for the qualities of each XAI condition.