The Friendlier the Robot, the Harder Its Mistakes Land, Brain Study Finds

A humanoid robot that makes eye contact, nods and says "uh-huh" earns people's attention fast. But according to a new study, that same charm makes its mistakes hurt more, not less.

KEY NUMBER
Once the expressive robot began making mistakes, its measurable sway over participants' decisions fell by more than half.

Why it matters

Companies building social robots have generally assumed that a warmer, more lifelike design earns forgiveness when things go wrong, the same way it does with people. This study is a direct test of that assumption, and it points the other way: an engaging robot's errors were read less like a glitch and more like a person breaking a social rule, and people pulled back harder as a result. For a field racing to put expressive humanoids into healthcare, retail and homes, that is a design trade-off worth knowing about before deployment, not after.

What happened

Each of the 50 participants spent roughly two and a half hours face to face with one of two versions of Pepper, a commercial humanoid robot: one animated, using gestures and eye contact, the other identical in what it said but physically motionless. Every session began with two short, error-free conversations and two collaborative tasks, building rapport and trust. In a third round, the robot started interrupting, giving illogical reasoning and making irrelevant remarks.

Throughout, the team recorded participants' prefrontal brain activity with a wearable optical sensor, tested saliva for the hormone oxytocin, surveyed people on trust after each stage, and tracked how often they actually followed the robot's advice. Self-reported trust dropped for both robot versions once errors began. But only the expressive robot's drop in trust carried through to real behavior, cutting its influence over participants' choices by more than half.

How the brain responded

When the expressive robot erred, two brain regions tied to reasoning about others' intentions became more active and more closely coordinated with each other, a pattern the researchers associate with treating a mistake as a social violation rather than a technical fault. The motionless robot's identical errors did not produce that pattern; participants appeared to read them as ordinary malfunctions instead.

The most unexpected result involved oxytocin, often called the bonding hormone because it typically rises between people who trust each other. Here, oxytocin rose as the expressive robot broke the rules of conversation, and the more it rose, the less participants trusted the robot and the less they followed its advice. The researchers interpret this as a vigilance signal rather than a sign of connection.

What happens next

The team, whose work spans Drexel, the U.S. Air Force Academy, George Mason University and the University of Southern California, says a natural next step is testing whether robots can repair trust after an error, for instance by apologizing or explaining what went wrong, the way people often do. They also want to see whether the same pattern holds for women, mixed groups and other robot designs before treating it as a general rule.

Takeaway

Making a robot more expressive can win people over quickly, but the same design that builds a bond also raises the price of any mistake it makes. For engineers weighing how "social" to make the next generation of humanoid robots, reliability, the study suggests, has to come first.