Dynamic Effects of Multiround Positive and Negative Feedback from Robots and Humans on Trust and Collaboration

Abstract: Human–robot collaboration is increasingly common in intelligent organizations, making it important to understand how trust changes across repeated interactions. Evaluative feedback can trigger trust updating, yet prior research has largely relied on single-time-point measures or one-off feedback. This research examined the dynamic effects of multiround positive and negative feedback from robot and human teammates on behavioral trust and willingness to collaborate.
A pilot study and two laboratory experiments were conducted. In the pilot study, 192 adults evaluated candidate feedback statements, from which four positive and four negative statements of comparable intensity were selected. In Experiment 1, 56 participants collaborated with either a Pepper robot or a human teammate and received four rounds of negative feedback. Experiment 2 used the same design with another 56 participants but presented positive feedback. In both experiments, participants completed five rounds of an investment decision task. Behavioral trust was indexed by the extent to which participants revised their judgments toward the teammate’s recommendation, whereas willingness to collaborate was measured after each round.
Under multiround negative feedback, behavioral trust and willingness to collaborate declined over time, but their trajectories differed by feedback source. Feedback from the robot teammate produced a relatively gradual and delayed pattern: behavioral trust changed little initially but declined in later rounds, while willingness to collaborate decreased as feedback accumulated. Feedback from the human teammate produced a sharper early decline followed by relative stabilization. Segmented models fit the data better than linear or quadratic models, supporting a lagged-cumulative pattern for robot feedback and an immediate-responsive pattern for human feedback.
These findings suggest that robots and humans may produce similar overall declines under negative feedback but different temporal processes of trust updating. Human feedback appears to have a stronger immediate interpersonal impact, whereas robot feedback may influence trust through accumulation. The results highlight the importance of examining when and how trust changes rather than relying only on end-point comparisons. The apparent asymmetry between positive and negative feedback should be interpreted cautiously because the higher-order interaction involving feedback valence was not consistently significant.