Humanoid robot Pepper study shows trust collapses after errors

A study published in *Science Robotics* examined how people trust an expressive humanoid robot, Pepper, during face‑to‑face conversations. Fifty adult participants interacted with either an animated version of Pepper that used eye contact, gestures and nods, or a motionless version that delivered the same dialogue without non‑verbal cues. After two error‑free sessions, a third session introduced conversational mistakes such as irrelevant replies, interruptions and illogical suggestions.

Researchers measured brain activity with wearable functional near‑infrared spectroscopy, collected saliva samples to assess oxytocin levels, recorded self‑reported trust ratings and observed participants' decision‑making behavior. Errors caused oxytocin to rise, indicating heightened suspicion, and led to a marked decline in trust and reduced reliance on the robot’s advice. Brain activity increased in the dorsolateral prefrontal cortex and medial prefrontal cortex, reflecting greater effort to resolve the awkward interaction.

The findings suggest that while expressive robots can initially boost engagement, their trustworthiness is fragile when they violate social norms. Designers of social robots for healthcare, education, customer service and home environments may need to prioritize error mitigation and transparent communication to maintain user confidence.