Expressive Humanoid Robots' Mistakes Trigger Suspicion, Not Trust—Brain Study Reveals
Key Takeaways
- ▸Expressive robots' errors activate brain regions that infer social intent, causing humans to perceive mistakes as social violations rather than technical failures
- ▸Counterintuitively, oxytocin levels rise during interactions with expressive robots that make errors, correlating with decreased trust—opposite the common assumption about the 'love hormone'
- ▸Motionless robots' mistakes trigger no coordinated brain activity and are judged more charitably as mechanical failures, preserving user trust
Summary
A new study published in Science Robotics challenges a core assumption in robot design: that expressive, lifelike humanoid robots earn more human trust. Researchers conducted experiments with 50 participants interacting with SoftBank's Pepper robot, measuring brain activity, oxytocin levels, trust, and decision-making behavior. Surprisingly, when an animated, expressive Pepper made errors—such as interrupting or offering illogical suggestions—participants' oxytocin levels increased, but so did their distrust and resistance to the robot's advice.
Using portable brain imaging (functional near-infrared spectroscopy worn on the forehead), researchers found that expressive robots' mistakes activate two brain regions in tandem: the dorsolateral prefrontal cortex (which flags norm violations) and the medial prefrontal cortex (which infers social intent). This coordinated brain activity predicted the rise in suspicion-linked oxytocin. In contrast, participants who interacted with a motionless Pepper treated its errors as mere technical glitches, with no such coordinated brain response.
The key insight: when robots display social cues, humans unconsciously reclassify their errors from technical malfunctions into social violations—the same category they apply to people. This finding upends conventional robot design wisdom and suggests that expressiveness alone cannot protect a robot's reputation. As robots increasingly enter homes, hospitals, and workplaces, understanding how humans judge robot reliability at a neurological level becomes critical for deployment success.
- The finding challenges the core design assumption that lifelike social expressiveness earns robot trust, suggesting more nuanced approaches to human-robot interaction design are needed
Editorial Opinion
This research delivers an important reality check to the robotics industry: anthropomorphizing robots doesn't automatically build trust—it can backfire spectacularly. By making humans apply social-judgment machinery to robot errors, expressiveness creates a higher bar for performance. The implication is profound: companies deploying humanoid robots in sensitive contexts (healthcare, elder care, customer service) must carefully weigh whether social expressiveness serves their use case or undermines it. The most trusted robot might not be the most human-like.
