Study: How Robot Witnesses Can Stop Abuse Through Empathy
As people encounter more and more robots in public places, abuse of these devices may become common. Abuse can take many forms, from milder behavior to outright cruelty. Unfortunately, people are willing to abuse or harass robots, and bystanders will not step in to stop it, even if the robot itself asks for help.
Without nearby people to rely on for rescue, robots have no choice but to rely on themselves and their fellow machines when it comes to safety. Researchers from Yale University's Interactive Machines Group conducted an experiment to determine whether emotionally expressive robot witnesses can prevent human abuse.
Consider this example: there is a small group of robots and a small group of people. If one person starts abusing one robot, will the other people do anything if the remaining robots react to the situation with sadness? Based on past research on robot abuse, empathy and bullying, the answer was "maybe." That is why the scientists decided to run a new experiment.
The experiment involved a group of three COZMO robots, a participant, and a researcher pretending to be a second participant. The humans and robots had to work together to complete construction tasks using wooden blocks. During this time, one of the robots would occasionally make mistakes, and the researcher would respond to each error with signs of abuse: calling the robot stupid, pushing it, shaking it, and throwing it across the table.
After each insult, the targeted robot reacts by displaying a sad face and shutting down for 10 seconds. During this time, the two other COZMO robots either do nothing or turn to the abused robot and express sadness in response to the abuse through animation. The researchers hypothesized that when the other robots responded to the abuse of their "friend" with sadness, the participant would feel more empathy and be more likely to intervene to stop the violence.
Such interventions were classified as "strong" or "weak" and could be verbal or physical. Strong interventions included physically interrupting the abuse and actively interrupting verbally using social pressure. Weak interventions included things like touching the robot after it was abused to make sure it was okay.
Participants did intervene when COZMO was subjected to abuse, and they were more likely to intervene when the other robots were sad. However, the sad robots did not significantly increase participants' empathy and perception, which made the results inconsistent. The researchers explained three reasons why participants did what they did:
- Subconscious empathy: broadly speaking, empathy is one person's reaction to another person's observed experience. People often feel empathy without being aware of it. This could have happened to the participants. They may have felt subconscious empathy while watching the robot being mistreated. The effect may have been stronger because of the sad reactions of the robot witnesses.
- Group dynamics: people tend to define themselves in terms of social groups. This can shape how they process knowledge and the emotional meaning of events. In the experiment, the participant, the researcher and the robots were part of one group because of the shared task. However, the researcher abusing one robot triggered negative emotions. This was more noticeable when the other robots expressed sadness, which only emphasized the negative perception. As a result, the participant stopped considering the "aggressive" researcher a member of the group and was able to intervene.
- Conformity to inaction: this is a type of social influence in group interactions. Participants may have failed to intervene in the abuse because the robots exerted social influence on them. They saw the lack of reaction from the robot witnesses, and this became the norm for their group interaction.
Participants explained why they chose not to intervene to stop the abuse: six people did not consider intervention necessary because they believed the robots had no feelings, five because they felt shy, scared or uncomfortable, and two because they feared that intervening would affect the task.
The researchers believe that the appearance of robots may change how people perceive abuse directed at them. In this case, COZMO's design may have been a factor that made it easier for participants to intervene.
“When robots are deployed in public places, we should not assume they will avoid mistreatment by users; it is very likely they will receive it. It is therefore important to design robots so they are safe when people act against them, both physically and computationally”
In one experiment, the robot destroyed a project the whole team had been working on. Then, after witnessing the abuse directed at the robot, the participant also insulted it, though more gently. This showed that there is a certain threshold at which people are willing to defend a poorly performing robot.
The value of this research lies in several things: it helps us understand how humans and robots interact when working together, and how to benefit from that.
“Our attitude toward machines is a mirror of ourselves. Why do people feel they have a free pass for the cruel behavior we constantly see on social media? The goal of the research is not to build better robots, but to build robots that can help us do better than ourselves”
Robots can indeed use their social context to reduce the impact of cruel actions directed at them. For example, they can motivate witnesses to intervene or help.
