This study explores human psychological and emotional responses to witnessing perceived "harm" or "violence" towards anthropomorphic and humanoid robots. As robots become more integrated into social environments, understanding these reactions is crucial for ethical design and deployment.
The analysis draws from research on the spectrum of human responses—emotional, physiological, and behavioral—and the factors that modulate these reactions, such as robot anthropomorphism, perceived agency, and the specific context in which the "harm" occurs.
A. Defining "Robot Harm": From Testing Protocols to Perceived Abuse
The concept of "robot harm" encompasses a wide spectrum of actions and interpretations. At one end, it includes intentional damage inflicted during rigorous resilience testing protocols, such as those famously depicted in videos by companies like Boston Dynamics, where robots are pushed, kicked, or otherwise destabilized to evaluate their balance and recovery capabilities. Such actions, while potentially appearing harsh, are typically framed within an engineering context of ensuring robustness and safety.
Further along the spectrum lies accidental damage, which may or may not elicit strong observer reactions depending on the circumstances. However, a significant area of HRI research focuses on actions perceived by observers as deliberate mistreatment, cruelty, or abuse. This perception is critical: it hinges less on the robot's actual capacity for sentience or suffering and more on the observer's interpretation of the perpetrator's intent and the robot's perceived capacity to be "wronged".
An operational definition of robot abuse cited in the literature is "persistent offensive action, either verbal or nonverbal, or physical violence that violates the robot's role or its human-like (or animal-like) nature". Further research has categorized patterns of robot abuse, such as "Outburst" (often due to frustration with malfunction), "Clash" (stemming from unmet relational expectations), "Oppression" (abuse for entertainment or power display), and "Rebellion" (aggression towards a robot perceived as a competitor or threat). These categories highlight the diverse motivations and contexts that can frame an action as "harmful."
B. Spectrum of Human Responses: Emotional, Physiological, and Behavioral
When humans witness actions perceived as harmful towards robots, they can exhibit a range of responses spanning emotional, physiological, and behavioral domains.
Emotional Responses:
A consistent finding is that humans are capable of experiencing empathy, sympathy, concern, and pity for robots they perceive as being harmed or mistreated, even when consciously aware that the robots are insentient machines. This can manifest as distress and other negative feelings; for example, witnessing a robot being "tortured" or destroyed can elicit discomfort and negative emotional reactions in observers. The public outcry following the destruction of hitchBOT, a simple hitchhiking robot, serves as a notable real-world example of such emotional investment. While negative empathetic responses are common, reactions can vary. Some individuals might find certain scenarios amusing, particularly if the robot is not highly anthropomorphized or if the "harm" is framed as slapstick. The "Oppression" pattern of abuse, for instance, describes situations where mistreatment is enacted for the user's entertainment.
Physiological Responses:
The emotional responses to robot harm are often accompanied by measurable physiological changes. Functional Magnetic Resonance Imaging (fMRI) studies have revealed activation in brain areas typically associated with empathy, such as limbic structures, when participants observe robots being harmed. These patterns can be similar to those seen when observing human-human harm, although often less intense. Other studies have employed electroencephalography (EEG) to demonstrate empathic processing for robots and have noted increased physiological arousal, such as changes in skin conductance response (SCR), during the observation of robot "torture". However, it's worth noting a counter-example where participants rated a robot appearing to be electrocuted as experiencing pain, but corresponding activation in their brain's pain matrix was not observed, suggesting complexities in the neural correlates of empathy for robots.
Behavioral Responses:
Observed harm can also influence human behavior. People may exhibit hesitation to harm a robot themselves, particularly if the robot has been personified (e.g., given a name, interacted with positively) or if it expresses distress cues. In one experiment, participants who had personified baby dinosaur robots were extremely reluctant to "torture and kill" them, even under instruction. Furthermore, bystanders may intervene to protect a robot they perceive as being abused. Inducing empathy towards a robot can also trigger broader prosocial behaviors, such as increased helpfulness.
C. Modulating Factors Influencing Human Reactions
The intensity and nature of human reactions to robot harm are not uniform but are modulated by several factors, including the robot's characteristics and the context of the harmful act.
1. Robot Anthropomorphism: Humanoid vs. Non-Humanoid/Animal-like Designs
A primary modulating factor is the robot's degree of anthropomorphism. Generally, increased human-likeness or animal-likeness correlates with stronger empathetic responses and greater condemnation of perceived harm. People tend to empathize less with robots that appear overtly mechanical. This directly addresses the inquiry about whether reactions are stronger for more humanoid robots; the evidence largely supports this assertion.
The form of anthropomorphism also matters. Animal-like robots, such as Boston Dynamics' Spot, can evoke strong empathetic reactions akin to those felt towards biological animals. The art exhibit "Dynamics of a Dog on a Chain," featuring robotic dogs, reportedly elicited significant empathy and discomfort from many viewers who perceived the chained, lunging robot as oppressed. Even non-humanoid "robotic objects" with minimal features can be perceived through a social lens and elicit social responses, largely due to the pervasive human tendency to anthropomorphize inanimate objects that exhibit life-like motion or behavior. Indeed, one study found that rude treatment of non-embodied, voice-based smart speakers could elicit empathetic reactions from observers.
A potential caveat exists at the extreme end of human-likeness: androids that fall deeply into the uncanny valley might, paradoxically, elicit less empathy than a less perfectly human-like but still clearly "alive" or engaging robot. This suggests that extreme realism, if flawed, can create an emotional distance that dampens empathetic connection.
2. Perceived Agency, Intentionality, and Emotional Expression of the Robot
Beyond physical form, a robot's perceived inner states and behavioral cues significantly influence human empathy. Attributing mind, sentience, or agency to a robot generally increases the likelihood of empathetic responses to its mistreatment and strengthens the condemnation of such acts. Studies have shown that explicitly attributing a mind (the capacity to think and feel) to a robot can lead participants to view its mistreatment as less acceptable.
The robot's own emotional expression is a powerful modulator. If a robot audibly or visually expresses distress, fear, or pain—even if these expressions are known to be simulated—it tends to heighten empathetic concern in observers and increase their hesitation to participate in or condone harmful actions towards it. An "empathetic robot" (one that demonstrates empathetic behaviors towards humans) was also found to garner more self-reported interventions from bystanders when it was subsequently abused. Similarly, attributing a personality to a robot, for example by giving it a name or a backstory, can increase people's reluctance to see it harmed.
3. Context of Harm: Observer Interpretation of Intent (Testing, Accident, Malice, Frustration)
The context in which the "harm" occurs and the observer's interpretation of the perpetrator's intent are crucial. Actions performed during engineering tests (e.g., Boston Dynamics videos) often elicit mixed reactions. While the animal-like or human-like movements of robots like Spot and Atlas can trigger empathetic discomfort in some viewers who perceive the actions as "abuse", others may understand the necessity of such tests for resilience and safety, thus moderating their perception of malicious intent. The justification of "testing" can frame the actions differently than unprovoked aggression.
Conversely, deliberate, seemingly malicious harm is likely to evoke stronger negative emotional responses and moral condemnation, especially if the robot target is highly anthropomorphized and appears defenseless. The intent behind the harm is a key factor in moral judgment, with intentional harm generally viewed more severely than accidental harm, a principle that extends from human-human morality to HRI.
Harm arising from user frustration due to robot malfunction (the "Outburst" pattern of abuse) presents another context. Observers might react with a degree of understanding for the user's frustration, even if they find the aggressive act itself problematic. This differs from reactions to unprovoked malice or systematic testing. The perceived "justification" or trigger for the harm heavily influences the observer's cognitive appraisal and subsequent emotional and moral judgment. This makes studying reactions to robot harm particularly complex, as controlling for and accurately assessing the perceived context is vital for interpreting results. The use of simulation for testing robot safety and failure modes offers a way to explore these scenarios without physical harm, though observing simulated harm may evoke different responses than witnessing physical interactions.
D. The Interplay of Human-Likeness and Empathy in Robot "Mistreatment" Scenarios
The phenomenon of humans reacting emotionally to the perceived mistreatment of robots, despite full awareness that these entities are non-sentient machines, highlights the "Computers Are Social Actors" (CASA) paradigm. This suggests a dissociation between rational knowledge (it's a machine) and more automatic social and emotional responses (it seems alive/suffering). This "it's just a machine" paradox is central to understanding human reactions to robot harm.
This leads to discussions about whether robots can, or should, have any form of moral standing or be considered "victims" in a meaningful sense. While they lack sentience, the act of abusing a robot, especially a human-like one, can be seen as reflecting poorly on the human perpetrator's character, potentially desensitizing them to violence, or having negative broader societal consequences, such as impacting the moral development of children who witness such acts. The concern shifts from the robot's "feelings" to the impact on human individuals and society.
Highly anthropomorphic or emotionally expressive robots might be perceived as more "vulnerable," thereby increasing protective instincts in some observers or heightening their distress when these robots are subjected to harm. This suggests that as robots become more socially sophisticated and human-like, they may not only elicit more protective empathy but also become entangled in more complex, and sometimes negative, human social dynamics. For instance, frustration arising from unmet social expectations from a highly anthropomorphic robot, or perceiving such a robot as a social competitor, could lead to specific forms of mistreatment not typically directed at simpler machines. This implies that simply making a robot "more human-like" to elicit empathy is not a straightforward solution to prevent all forms of mistreatment; it could inadvertently open doors to different types of negative interactions if not managed with a nuanced understanding of human psychology.
The following table summarizes how robot characteristics and the context of harm can influence human reactions:
| Robot Characteristic(s) | Context of Harm | Typical Human Emotional Reaction | Typical Human Physiological Arousal |
|---|---|---|---|
| High Human-Likeness (Humanoid); Expresses Distress | Deliberate Abuse by Stranger | High Empathy/Distress, Moral Outrage | Increased SCR/HR, Limbic Activation (fMRI) |
| Animal-Like (e.g., Spot); Realistic Movement | Stability Testing (e.g., kicking, pushing) | Mixed: Empathy/Discomfort for some, Neutral/Understanding for others | Potential for SCR/HR changes in empathetic individuals |
| Low Human-Likeness (Mechanical); No Distress Cues | Deliberate Abuse by Stranger | Low Empathy, Indifference, or Amusement in some | Minimal or absent |
| High Human-Likeness; Perceived Agency/Mind Attributed | Any Form of Perceived Mistreatment | Increased Empathy, Condemnation of Mistreatment | Likely similar to humanoid/distress above |
| Any Robot Type | Frustration-Induced Harm by User (Robot Malfunction) | Mixed: Potential understanding of frustration, but disapproval of aggression | Variable |
| Non-Humanoid (e.g., Smart Speaker); Voice-based Interaction | Rude Verbal Treatment by User | Empathetic reaction from observers | Not directly measured in snippet for observers |
