Study: Expressive Humanoid Robots' Mistakes More Damaging to Trust

The study was led by Drexel University, in collaboration with the U.S. Air Force Academy, George Mason University, and the University of Southern California's Institute for Creative Technologies. It was published in Science Robotics and funded by the Air Force Office of Scientific Research under the U.S. Department of Defense.

The study involved 50 healthy adult males, each interacting with the Pepper humanoid robot for about two and a half hours. Half of the participants faced an expressive version (with eye contact, gestures, nodding, and responses), while the other half faced a stationary version (saying the same words but without expressions).

Pepper was controlled by a human operator following a script, unbeknownst to participants. Participants first completed two error-free interactions; in the third interaction, the robot began interrupting, making irrelevant comments, and providing illogical explanations.

Researchers simultaneously used wearable functional near-infrared spectroscopy (fNIRS) to monitor prefrontal cortex activity, collected saliva samples to measure oxytocin, administered questionnaires, and tracked whether participants changed their decisions based on the robot's suggestions.

The study found that participants engaged more deeply with the expressive robot, but when it made mistakes, they were less willing to trust its suggestions. The robot's measurable influence on decisions dropped by more than half after it began making errors.

When the expressive robot made mistakes, brain regions associated with social reasoning showed increased activity, which researchers interpreted as people treating the robot's behavior as a social violation rather than a mechanical failure. The same pattern was not observed with the stationary robot.

Oxytocin levels rose when the robot made mistakes, even though reported trust declined. Researchers suggest this indicates that oxytocin may sometimes act as a social vigilance signal, not just a sign of closeness.

Study author Frank Krueger noted: 'A charismatic robot pays a higher price when it makes mistakes than an ordinary robot. Expressiveness is not free; it brings engagement, but also vulnerability.'

Researchers suggest that designers may need to carefully weigh expressiveness against reliability. Robots that behave more like social partners may attract more engagement, but the same behavior may make errors more consequential when the robot fails to meet the expectations it creates.

Corresponding author Ewart J. de Visser emphasized: 'Reliability must come first. An unreliable expressive robot is not a safer robot, but a more disappointing one, because expressiveness raises the bar that the robot cannot meet.'

The study concludes that robot design should not consider cuteness in isolation, but should incorporate neurobiology and psychology to maximize performance.