'Victims and Observers: How Gender, Victimization Experience, and Biases Shape Perceptions of Robot Abuse' by Hideki Garcia Goo, Katie Winkle, Tom Williams and Megan Strait in (2023) IEEE International Symposium on Robot-Human Interactive Communication comments
With the deployment of robots in public realms, researchers are seeing more and more cases of abusive disinhibition towards robots. Because robots embody gendered identities, poor navigation of antisocial dynamics may reinforce or exacerbate gender-based violence. Robots deployed in social settings must recognize and respond to abuse in a way that minimizes ethical risk. This will require designers to first understand the risk posed by abuse of robots, and how humans perceive robot-directed abuse. To that end, we conducted an exploratory study of reactions to a physically abusive interaction between a human perpetrator and a victimized agent. Given extensions of gendered biases to robotic agents, as well as associations between an agent’s human likeness and the experiential capacity attributed to it, we quasi-manipulated the victim’s humanness (via use of a human actor vs. NAO robot) and gendering (via inclusion of stereotypically masculine vs. feminine cues in their presentation) across four video-recorded reproductions of the interaction. Analysis of data from 417 participants, each of whom watched one of the four videos, indicates that the intensity of emotional distress felt by an observer is associated with their gender identification, previous experience with victimization, hostile sexism, and support for social stratification, as well as the victim’s gendering.
The authors state
Much of existing research on robot abuse has focused on the potential for robot abuse to impact those perpetrating that abuse (typically negatively [11], [12], although cf. [13]). However, the impacts of abuse, and a victim’s response to it, extend not only to abusers, but to bystanders and observers as well. Although people may believe themselves disaffected in aggressing an artificial agent, the impacts of abuse, and an agent’s response to it, extend beyond individual inter- actions. Research on human-robot interaction dynamics, for example, has found that people react to the abuse of robotic technologies similar (to a lesser degree) to how they react to seeing the abuse of other people [14], [9], [10], and even the abuse of Cozmo – Anki’s minimally agentic, toy-like robot – has been observed to induce substantial distress in bystanders witnessing the interaction [15], [5].
Moreover, the effects of abusing a robot – as well as witnessing a robot’s abuse – likely extend beyond a single interaction (e.g., [12]). For example, the inability of a robot to respond to social aggression may risk normalization [16] – or even escalation [17] – of that behavior. This suggests that abuse, if left unaddressed, has the potential to weaken moral norms surrounding those abusive behaviors, both in perpetrators, observers, and ultimately those with whom perpetrators and observers interact.
Consequently, agents unable to navigate antisocial dy- namics risk replicating, reinforcing, and exacerbating extant social inequities [18]. For example, consistent with the observations outlined above, many people verbally abused Microsoft’s chatbot Tay upon its 2016 deployment. Because Tay was designed to learn from its interactions with users – but lacked any mechanisms to recognize and respond to antisocial content – the bot quickly morphed from its intended cheery, teenage girl-like persona into an overt white supremacist, directing racist, sexist, and xenophobic hostility toward unconsenting users before Microsoft intervened [19].
People ascribe robots gender [20], even in the absence of intentional gender cueing [21], a phenomena that emerges at least as early as 8 years of age [22]. This enables robots to similarly evoke and reinforce gendered stereotypes in a complex way that interacts with interactants’ gender identities [23], [21], [24]; but also allows for the intentional subversion of gender norms and stereotypes [25], [26]. Given differences in perceptions (and realities) of gender-based violence [27], [28], [29], [30], it is critical for robot designers to have a nuanced understanding of these complex gender- mediated perceptions and their implications in the context of robot abuse.
