Human-Agent Trust Relationships in a Real-Time Collaborative Game

This research investigates the dynamics of human-agent trust within real-time collaborative environments, specifically focusing on how agent behavior, reliability, and predictability influence user reliance and performance. By utilizing a custom-built video game framework modeled after classic arcade mechanics, the study captures both objective interaction data and subjective survey responses to analyze how trust evolves during high-stakes, uncertain tasks. The primary objective is to develop trust-aware agents capable of adapting their behavior to individual users in real-time.

Key findings indicate that agent predictability is a critical driver of trust, as users consistently report higher trust levels and lower cognitive loads when interacting with predictable agents, even when performance levels remain constant. Furthermore, the nature of agent errors significantly impacts user behavior; errors resulting from agent inaction are generally perceived less negatively than errors of planning or commission. Preliminary data regarding environmental uncertainty suggests that users may report higher trust in agents when visual information is restricted, despite no significant change in overall situational awareness.

The research methodology employs a series of controlled lab-based and online experiments, including a within-groups study with 30 participants and subsequent investigations into error types and visual uncertainty. By integrating real-time logging with established psychometric instruments, the framework successfully demonstrates that user trust ratings can be inferred with high accuracy. Future research directions include exploring trust-repair mechanisms, such as emotional communication or voice capabilities, to help users re-engage with agents following negative experiences. This work contributes to the broader field of human-computer interaction by providing an empirical basis for designing collaborative systems that maintain user engagement and appropriate levels of reliance.