Equity – Bias
Some users do not feel part of the intended audience (algorithmic bias).
Bias in human-robot interaction refers to the ways in which robots may systematically exclude, misrepresent, or disadvantage certain users or groups. Unlike overt design failures, bias often emerges subtly: through assumptions embedded in datasets, physical design choices, or interaction patterns that work well for some people but poorly for others. In everyday use, this can translate into a robot that feels less usable, less understandable, or even irrelevant to parts of its intended audience, raising concerns about fairness and inclusion in real-world deployments.
These conditions can vary significantly: from differences in geography affecting maintenance and performance, to differences in user demographics shaping interaction success. In practice, this means a robot that works well in one setting or for one population may systematically underperform or misalign with others, not because of technical failure alone, but because of uneven design assumptions.
This concern is reflected consistently in empirical studies and design critiques. In one in particular, participants reported feeling excluded from the intended user group of a companion robot, highlighting how design choices can implicitly signal who a robot is "for." Older adults have expressed concern about socially assistive robots lacking spontaneity, with interactions feeling overly scripted and uniform, which can further reinforce a sense of exclusion for users whose needs do not fit the default interaction model. Broader design research warns that anthropomorphic design decisions can encode and reproduce social categories such as gender, race, class, or disability, often in ways that reflect existing power imbalances. Similarly, it has been noted that humanoid robots may unintentionally normalise stereotypical representations of human roles, reinforcing biased portrayals through their appearance and behavior.
Together, these findings show that bias in HRI is not only about algorithmic fairness, but also about how robots are perceived, who they implicitly include or exclude, and how they reproduce social categories in everyday interaction.
Excerpts from the paper:
About the value "Equity"
Equity entails treating people differently based on the circumstances, to ensure an equal outcome. In contrast, equality – treating everyone the same regardless of their situation – did not emerge as a relevant value during the focus groups discussions. Indeed, the experts suggested focusing on equity as a key value. They noted that equity is closely linked to demographics (who) and the environment (where), which can significantly impact a robot's performance. For instance, the geographical location can influence how often a robot overheats and how easily it can be repaired.
About "Bias"
The focus groups participants and some publications voiced the concerns of certain communities and populations which felt not being part of the intended user base of certain robots. This can be caused by the physical aspects of the robot or by the interaction design. Especially when powered by data-driven technologies, the robots can reflect the bias present in the dataset on which their algorithms are trained, as highlighted by the focus groups participants. The experts also noted that a robot might struggle to accurately identify emotional states in children with varying cognitive abilities or socioeconomic backgrounds. Or, in another example, it could treat students in undesirable ways, such as reinforcing gender stereotypes.