The Impact of Demonstrated Untrustworthiness and Severity of Mistake on Trust Emergence in an Agent-Based Model of The Trust Game

Abstract

Development of trust is critical to cooperation games, especially when trust breaches majorly influence the outcome. The Trust Game is used to study trust between an investor and trustee by comparing their generosity levels.

Current models of The Trust Game measure investor trust in response to selfish trustee behavior. This model in NetLogo does the same while attempting to understand the impact of timing and severity of an untrustworthy event.

In this simulation, agents play 15 iterations of The Trust Game under 2 conditions: 30% trustee return on round 2; 10% trustee return on round 7. For remaining rounds, the trustee determines a 40%-100% return with q-reinforcement-learning. Using ACT-R’s Base Level Activation (BLA) Formula, investor behavior is categorized as distrust: BLA of demonstrated untrustworthiness > BLA of neutral events; moderate trust: retention threshold < BLA of demonstrated untrustworthiness < BLA of neutral events; trust: BLA of demonstrated untrustworthiness < retention threshold. BLA of demonstrated untrustworthiness is scaled to the severity of the mistake, calculated as 10 X (1-return rate).  

Early, minor mistakes required longer recovery: 6 iterations. Meanwhile, it took 4 iterations to recover from a late, severe mistake. The BLA of the untrustworthy event never fell below the retention threshold; keeping the investor at moderately trusting for the remaining iterations. Also note that the BLA of an untrustworthy event declined more rapidly than the BLA of neutral events.

These findings suggest that early, minor trust breaches require longer recovery than late, severe breaches, despite similar long-term trust levels.