Criminal Law Library Blog
Introduction
A recent Tech Xplore article reports on research suggesting that making humanoid robots more socially expressive (through eye contact, gestures, nodding, and other humanlike behaviors) can increase engagement but may also carry an unexpected cost: when an expressive robot makes a mistake, people may react to the error more as a social violation than as a simple technical failure.
The article, “A Humanoid Robot’s Social Expressiveness May Backfire When It Makes Mistakes,” was written by Ingrid Fadelli and published by Tech Xplore/Phys.org on August 28, 2026, with editing by Robert Egan. It reports on research conducted principally by investigators at Drexel University and published in Science Robotics. The underlying study, by Yigit Topoglu and colleagues, is titled “Multilevel Dynamics of the Brain, Hormones, Mind, and Behavior in Social Human-Robot Interaction.” The following is an overview of the two articles:
Studying Trust in Human-Robot Interaction
As robots increasingly move beyond industrial settings and into homes, health care facilities, classrooms, workplaces, and service environments, researchers are paying greater attention to a fundamental question: What causes people to trust, or distrust, a robot?
Researchers led by Hasan Ayaz and Ewart J. de Visser at Drexel University examined whether a humanoid robot’s social expressiveness affects the way people respond when the robot behaves reliably and when it makes mistakes. Rather than measuring trust in only one way, the researchers examined several dimensions simultaneously, including participants’ reported trust, observable behavior, brain activity, and levels of oxytocin, a hormone involved in social processes.
The experiment involved 50 adult male participants, each of whom interacted for approximately two and a half hours with Pepper, a semi-humanoid robot. Participants engaged in conversations and collaborative decision-making exercises with the robot. In some interactions Pepper behaved appropriately and gave logically consistent responses; in others, it deliberately violated conversational expectations by interrupting, making irrelevant comments, or offering poorly reasoned suggestions.
Half of the participants interacted with a socially expressive version of Pepper that made eye contact, gestured, nodded, and provided conversational responses such as “uh-huh” or “hmm.” The other half interacted with a stationary version of the robot that delivered comparable verbal responses without the expressive physical behavior.
Reliability Matters More Than Social Polish
One of the study’s clearest findings was that robot reliability remained fundamental to trust. When Pepper made errors, participants became less willing to accept its recommendations, regardless of whether the robot was expressive or stationary.
Social expressiveness, however, appeared to change the way participants interpreted those mistakes. According to Ayaz, the findings suggest that an expressive robot’s error may be processed more like an interpersonal violation, whereas a similar error from a motionless robot may be perceived more readily as a technical malfunction.
Exploratory measurements of brain activity supported this distinction. Errors by the expressive robot were associated with increased activity in areas of the prefrontal cortex involved in monitoring uncertainty, recognizing violations of expectations, and attempting to infer another agent’s intentions. Comparable patterns were not observed when the stationary robot made errors.
The researchers therefore describe social expressiveness as something of a double edged sword. Humanlike behaviors may encourage people to engage more deeply with a robot, but that increased engagement can also raise expectations. When the robot subsequently behaves badly or unreliably, the resulting loss of trust may acquire a distinctly social dimension.
An Unexpected Finding About Oxytocin
The study also produced a particularly interesting physiological finding. Oxytocin levels increased when the expressive robot made errors even as participants’ reported trust declined.
Oxytocin is ” a natural body chemical—known as a hormone and brain messenger—often called the love hormone ‘because it helps build social bonds, trust, and loving feelings’ ” Although oxytocin is popularly characterized as a “bonding” or “trust” hormone, researchers increasingly recognize that its effects are highly dependent on social context. In this experiment, increased oxytocin appeared to be associated not simply with greater attachment but potentially with heightened attention or vigilance toward socially significant behavior.
This finding reinforces a broader methodological point emphasized by the researchers: human trust cannot necessarily be understood through a single measurement. What people say they trust, how they behave, what occurs in the brain, and their physiological responses may not always move in the same direction.
Implications for the Design of Humanoid Robots
The research has potentially significant implications as socially interactive robots become more common. Designers may be tempted to make robots increasingly humanlike on the assumption that gestures, facial orientation, eye contact, and conversational behaviors will automatically make people more comfortable with them.
The Drexel research suggests that reliability should come before social sophistication. Giving a robot more humanlike characteristics may also cause users to apply more humanlike expectations to its behavior. When those expectations are violated, the robot may not merely appear defective, it may appear socially inappropriate, uncooperative, or untrustworthy.
That distinction could become particularly important in settings such as health care, education, elder care, customer service, and other environments in which sustained cooperation between humans and robots may depend heavily upon trust.
The findings also raise an intriguing question for future research: If an expressive robot’s mistake is interpreted as a social violation, might trust also need to be repaired socially? Researchers are therefore interested in examining whether robots can regain trust by acknowledging mistakes, apologizing, explaining their behavior, or otherwise signaling good intentions.
Important Limitations
The researchers caution against treating these findings as universally applicable. The principal experiment involved 50 adult male participants, and the researchers themselves note the need for future studies involving women, different age groups, diverse cultural and ethnic populations, and other types of robots. The study also relied on particular physiological and brain imaging techniques that have methodological limitations.
Accordingly, the findings should be understood as an important contribution to an emerging field rather than as a definitive account of how all people will respond to socially expressive robots.
Broader Perspective
Perhaps the most important lesson from the research extends beyond robotics. As artificial systems increasingly communicate through voices, gestures, personalities, and other humanlike characteristics, people may begin evaluating their mistakes differently. The more a machine presents itself as a social actor, the more humans may expect it to behave according to social as well as technical standards.
The study therefore suggests that successful human-robot interaction may depend not simply on making machines appear more human, but on understanding the expectations that humanlike behavior creates—and what happens to trust when those expectations are not met.
Why Librarians and Researchers Should Care
Although this research focuses on humanoid robots, its implications extend to librarians, researchers, educators, and other information professionals who increasingly work with artificial intelligence systems designed to communicate in humanlike ways.
As AI systems become more conversational, responsive, and socially expressive, users may begin to evaluate them not simply as information-retrieval technologies but as seemingly knowledgeable partners. A confident voice, conversational fluency, apparent empathy, or other humanlike qualities can make interaction easier and more engaging. But these characteristics do not necessarily make the information produced by the system more accurate or reliable.
The Drexel research offers a useful reminder that social expressiveness and reliability are different qualities. In the study, reliability remained fundamental to trust, while greater social expressiveness could make mistakes more consequential when they occurred.
For librarians and researchers, this distinction reinforces a longstanding principle of information literacy: authority and reliability should be evaluated through evidence, sources, methodology, and verification, not through the persuasiveness or humanlike qualities of the messenger.
This principle may become increasingly important as the boundary between information tools and social technologies becomes less obvious. Researchers using conversational AI should therefore remain alert to the possibility that humanlike interaction can influence perceptions of credibility independently of actual accuracy.
In this sense, the study raises a broader question extending well beyond robotics: As machines become increasingly skilled at communicating like humans, will people become better at critically evaluating what those machines tell them—or more inclined to trust them because of how convincingly they communicate?
For librarians, researchers, and other information professionals, helping users understand that distinction may become an increasingly important component of AI literacy and critical thinking.
Primary Sources
Tech Xplore / Phys.org
Ingrid Fadelli, A Humanoid Robot’s Social Expressiveness May Backfire When It Makes Mistakes, Tech Xplore (Aug. 28, 2026), edited by Robert Egan.
Live link: https://techxplore.com/news/2026-08-humanoid-robot-social-backfire.html
Underlying Research Study
Yigit Topoglu, Frank Krueger, Shawn Joshi, Nina Rothstein, Adrian A. Franke, Xingnan Li, Jonathan Gratch, Ewart J. de Visser & Hasan Ayaz, Multilevel Dynamics of the Brain, Hormones, Mind, and Behavior in Social Human-Robot Interaction, Science Robotics, vol. 11, no. 116, eaec1762 (July 29, 2026).