# Building and Repairing Trust in Chatbots: The Interplay Between Social Role and Performance During Interactions

## TL;DR

This study explores how the social role of chatbots affects users' trust and how trust is repaired after failures.

## Contribution

The study introduces a novel approach to understanding trust in AI through a simulated investment task with chatbots.

## Key findings

- Chatbots with a friend-like role fostered higher behavioral trust than servant-like ones.
- Friend-like chatbots were more effective in mitigating trust erosion during failures.
- Friendship with AI can act as a buffer for trust recovery after violations.

## Abstract

Trust (or distrust) in artificial intelligence (AI) is a critical research topic, given AI’s pervasive integration across societal domains. Despite its significance, scholarly attention to process-based learned trust in AI remains limited. To address this gap, this study designed a virtual non-fungible token (NFT) investment task, featuring seven rounds of risk decision-making scenarios, to simulate an investment/trust game to explore participants’ multifaceted trust under the influence of different chatbots’ social role. The findings suggested the chatbot’s social role had a significant impact on participants’ trust behaviors and perceptions over time. Trust in the two chatbot types diverged until the system-induced failures occurred. The friend-like chatbot elicited a higher level of behavioral trust than the servant-like counterpart. During those trust-damaging moments, the friend-like chatbot proved more effective in mitigating trust erosion and facilitating trust repair, as evidenced by relatively stable investment behaviors. The findings reinforce the notion that friendship with AI can function as a relational buffer, softening the impact of trust violations and facilitating smoother trust recovery.