Paper Type
ERF
Abstract
Artificial Intelligence (AI) systems are increasingly embedded in high-stakes human decision-making, yet achieving appropriate reliance remains elusive. Although Explainable AI (XAI) aims to address this, static explanations often fail to meet diverse user needs, while interactive natural-language explanations risk functioning as unintended persuasion agents by heightening perceptions of system humanness. Grounded in the Persuasion Knowledge Model (PKM), this study develops and tests a theoretical framework showing how explanation type (static versus interactive) influences perceived humanness, which shapes beliefs about the XAI agent’s assistive versus persuasive intent and ultimately affects reliance behavior. The hypotheses are tested in a between-subjects experiment (N=100) using a deception-detection task with hotel reviews, offering insights for the ethical design of conversational XAI that fosters complementary human-AI performance rather than over- or under-reliance.
Paper Number
1947
Recommended Citation
Lomo, Marvin Adjei Kojo; Singh, Rahul; and Ray, Arindam, "Interactive XAI in AI-Augmented Decision-Making: A Persuasion Knowledge Perspective for Understanding the Effects of Interactive XAI on Appropriate Reliance" (2026). AMCIS 2026 Proceedings. 29.
https://aisel.aisnet.org/amcis2026/sig_dsa/sig_dsa/29
Interactive XAI in AI-Augmented Decision-Making: A Persuasion Knowledge Perspective for Understanding the Effects of Interactive XAI on Appropriate Reliance
Artificial Intelligence (AI) systems are increasingly embedded in high-stakes human decision-making, yet achieving appropriate reliance remains elusive. Although Explainable AI (XAI) aims to address this, static explanations often fail to meet diverse user needs, while interactive natural-language explanations risk functioning as unintended persuasion agents by heightening perceptions of system humanness. Grounded in the Persuasion Knowledge Model (PKM), this study develops and tests a theoretical framework showing how explanation type (static versus interactive) influences perceived humanness, which shapes beliefs about the XAI agent’s assistive versus persuasive intent and ultimately affects reliance behavior. The hypotheses are tested in a between-subjects experiment (N=100) using a deception-detection task with hotel reviews, offering insights for the ethical design of conversational XAI that fosters complementary human-AI performance rather than over- or under-reliance.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIG DSA