Trust Evolution Over Time in Explainable AI for Fake News Detection

Shiva Kumar Pentyala

Abstract:

The need for interpretable and accountable intelligent systems is strong as artificial intelligence (AI) becomes more
prevalent in human life. We study the effects of interpretability on user’s trust in an AI assistant tool designed for fake
news detection. In our study, we expose participants to
different types of AI and Explainable AI (XAI) assistants,
measure their perceived accuracy of algorithm, and cluster
user trust changes over time into five types of trust evolution. We present quantitative results and analysis from
human-subject studies and discuss our findings regarding
how model explanations affect on user trust evolution over
time.