Expressive Robots Lost Half Their Influence After One Mistake

Then the robot broke the rules

Every session built trust first. Two error-free rounds of small talk and collaborative choices: pick a survival item on a deserted island, rank five paintings. In the third round, Pepper interrupted, repeated itself, defended illogical options, and drifted off topic. That is social rule-breaking, not a servo stutter.

Lead author Yigit Topoglu and senior author Hasan Ayaz measured four streams at once: prefrontal activity with wearable fNIRS, salivary oxytocin at three points, self-reported trust after each interaction, and whether participants revised choices to match the robot. Frank Krueger of George Mason and Ewart de Visser of the Air Force Academy share corresponding authorship. The EurekAlert write-up and a Conversation brief by the team walk the design.

Self-reported trust fell for both robots once the errors began. The expressive robot paid a steeper behavioral price. Its measurable influence over participants' later decisions fell by more than half. Prefrontal regions tied to social reasoning lit up when the animated Pepper failed. The motionless twin's identical mistakes looked more like a glitch.

Oxytocin rose as trust fell. The hormone usually associated with bonding tracked vigilance here. Higher oxytocin during expressive errors predicted lower trust and less willingness to take the robot's advice. The bonding chemical was reading the room as a social threat.

The designers' temptation is obvious: make the machine likable so people forgive it. This study says the opposite trade is live. Expressiveness moves a mistake out of the mechanical category and into the interpersonal one. Reliability has to clear the bar charm raises. An unreliable expressive robot is a more disappointing partner, which is a worse product for any setting where trust decides whether the machine gets used at all.