Peer Pressure from Machines: When Robots Sway What We Say

Classic conformity experiments are being rerun with robots in the role of the crowd, and the results suggest that adults mostly shrug off a chorus of machines while children, and adults who have come to trust the robots, often do not.

Asch, revisited with servos

In the 1950s the psychologist Solomon Asch showed that people will sometimes give an obviously wrong answer to a simple perceptual question if everyone else in the room has just given it. Participants compared the lengths of lines on a card, and confederates planted in the group confidently chose the wrong one. The finding became a cornerstone of social psychology because it exposed how heavily ordinary judgement leans on what other people seem to think.

Robots raise an obvious question. If a group of machines all give the same wrong answer, does that create the same pull? The question is not academic. Social robots are being designed for classrooms, care homes and shops, all places where a machine might try to steer what people believe or do. Knowing whether robots can exert social pressure is a precondition for deciding whether they should.

Adults resist, children yield

An early attempt came from Jürgen Brandstetter, Christoph Bartneck and colleagues, who recreated the Asch setup at IROS 2014 with a group of small humanoid robots and added a verbal judgement task alongside the visual one. Human peers produced conformity in both tasks, as expected, and conformity was higher when the judgement was ambiguous. Robot peers, however, did not significantly sway the adult participants.

Anna-Lisa Vollmer, Robin Read, Dries Trippas and Tony Belpaeme took the paradigm further in a 2018 paper in Science Robotics. They first replicated the classic result that adults conform to human peers, then showed that adults largely resisted a panel of small humanoid robots giving wrong answers. The striking part came next. When the same task was given to children aged seven to nine, the children did conform to the robots. According to the university press coverage of the study, the children were right about 87 per cent of the time when answering alone, and this fell to about 75 per cent with robots present, with roughly three quarters of their wrong answers matching the robots' mistakes exactly.

The authors were careful to present this as a double-edged finding. Conformity can be useful, for example when a robot tutor nudges a child towards a correct method. But the same susceptibility means that a faulty or badly designed robot could lead young users astray, and the paper explicitly raised the possibility of misuse. The press release accompanying it went further and floated the idea of regulatory safeguards.

It depends on whether the robots seem to know

So are adults simply immune? A 2018 HRI study by Nicole Salomons, Michael van der Linden, Sarah Strohkorb Sebo and Brian Scassellati suggests not. They argued that earlier replications might have failed because the Asch task has an obviously correct answer, so a robot giving the wrong one quickly loses credibility. Drawing on the idea that people weigh truth, trust and social solidarity when deciding whether to go along with a group, they designed a game with three robots in which there was no objective right answer.

In that setting, participants who saw the robots' answers before giving their own changed their responses to match much more often than a control group, at roughly 29 per cent against 6 per cent in a small sample. Early in the game the rate of conformity was comparable to what Asch reported with human groups. But once the robots made a choice that turned out to be wrong, many participants stopped following them. Conformity, in other words, tracked trust, and trust tracked apparent competence.

From conformity to persuasion

Put these studies together and a more nuanced picture emerges than the headline that robots can or cannot pressure us. Adults seem to discount a robot crowd when the truth is plain, perhaps because they see the machines as instruments that can be wrong rather than as peers whose approval matters. When the answer is uncertain and the robots have earned a reputation for being right, adults defer too. Children, who are still learning how much to credit different sources, may extend social standing to robots more readily.

This matters because persuasion is a design goal for many social robots, from health coaches to retail assistants. The mechanisms these studies expose are not exotic. They are the ordinary levers of human influence: consensus, credibility and the desire not to stand out. A robot does not need to be a full social peer to pull them; it needs to seem informed, or to be in front of someone who is not yet sure how far to trust it.

The ethical question, then, is less whether robots can persuade and more who gets persuaded and about what. A machine that subtly amplifies a group view in a classroom, or quietly agrees with a sales pitch, is exercising influence that its users may not notice. The research so far suggests that the most susceptible audiences are exactly the ones we most want to protect.

References

1. Brandstetter, J., Rácz, P., Beckner, C., Sandoval, E. B., Hay, J. and Bartneck, C. (2014). A peer pressure experiment: Recreation of the Asch conformity experiment with robots. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 1335-1340. Link
2. Vollmer, A.-L., Read, R., Trippas, D. and Belpaeme, T. (2018). Children conform, adults resist: A robot group induced peer pressure on normative social conformity. Science Robotics, 3(21), eaat7111. Link
3. ScienceDaily (2018). News report on robot-induced peer pressure in children, summarising Vollmer et al. (2018). Link
4. Salomons, N., van der Linden, M., Strohkorb Sebo, S. and Scassellati, B. (2018). Humans conform to robots: Disambiguating trust, truth, and conformity. Proceedings of the 2018 ACM/IEEE International Conference on Human-Robot Interaction (HRI), 187-195. Link

Feature drafted with AI from the published research listed above, for social-robotics.com.