Previous research has demonstrated reliable effects of social pressure on conformity and social decision-making in human-human interaction. The current study investigates whether non-human agents are also capable of inducing similar social pressure effects; in particular, we examined whether the degree of physical human-likeness of an agent (i.e., appearance) modulates conformity and whether potential effects of agent type on conformity are modulated further by task ambiguity. To answer these questions, participants performed a line judgment task together with agents of different degrees of humanness (human, robot, computer) in either a high or low ambiguity version of the task. We expected an increase in conformity rates for agents with increasing levels of physical humanness, as well as for increasing levels of task ambiguity. Results showed low-level conformity with all agents, with a significant difference in conformity between the high and low ambiguity version of the task (i.e., stronger compliance for the high versus the low task); the degree of humanness, however, did not have an influence on conformity rates (neither alone or in combination with task type). The results suggest that when performing a task together with others, participants always conform to some degree with the social interaction partner independent of its level of humanness; the level of conformity, however, depends on task ambiguity with stronger compliance across agents for more ambiguous tasks.

Results suggest that users may react differently to the influence of nonhuman agent groups with the potential for variability in conformity depending on the domain of the task.

As nonhuman agents become integrated into the workforce, the question becomes whether humans are willing to consider their advice, and to what extent advice-seeking depends on the perceived agent-task fit. To examine this, participants performed social and analytical tasks and received advice from human, robot, and computer agents in two conditions: in the Agent First condition, participants were first asked to choose advisors and were then informed which task to perform; in the Task First condition, they were first informed about the task and then asked to choose advisors. In the Agent First condition, we expected participants to prefer human to nonhuman advisors, and to subsequently trust their advice more if they were assigned the social as opposed to the analytical task. In the Task First condition, we expected advisor choices to be guided by stereotypical assumptions regarding the agents' expertise for the tasks, accompanied by higher trust in their suggestions. The findings indicate that in the Agent First condition, the human was chosen significantly more often than the machines, while in the Task First condition advisor choices were calibrated based on perceived agent-task fit. Trust was higher in the social task, but only showed variations with the human partner.

As interactions with non-human agents increase, it is important to understand and predict the consequences of human interactions with them. Social facilitation has a longstanding history within the realm of social psychology and is characterized by the presence of other humans having a beneficial effect on performance on easy tasks and inhibiting performance on difficult tasks. While social facilitation has been shown across task types and experimental conditions with human agents, very little research has examined whether this effect can also be induced by non-human agents and, if so, to what degree the level of humanness and embodiment of those agents influences that effect. In the current experiment, we apply a common social facilitation task (i.e., numerical distance judgments) to investigate to what extent the presence of agents of varying degrees of humanness benefits task performance. Results show a significant difference in performance between easy and difficult task conditions, but show no significant improvement in task performance in the social presence conditions compared to performing the task alone. This suggests that the presence of others did not have a positive effect on performance, at least not when social presence was manipulated via still images. Implications of this finding for future studies, as well as for human-robot interaction are discussed.

This study examines to what extent mixed groups of computers and humans are able to produce conformity effects in human interaction partners. Previous studies reveal that nonhuman groups can induce conformity under certain circumstances, but it is unknown to what extent mixed groups of human and nonhuman agents are able to produce similar effects. It is also unknown how varying the number of human agents per group can affect conformity. Participants were assigned to one of five groups varying in their proportion of human to nonhuman agent composition and were asked to complete a social and analytical task with the assigned group. These task types were chosen to represent tasks which humans (i.e., social task) or computers (i.e., analytical task) may be perceived as having greater expertise in, as well as roughly approximating real-world tasks humans may complete. A mixed analysis of variance (ANOVA) revealed higher rates of conformity (i.e., percentage of time participants answered in line with their group on critical trials) with the group opinion for the analytical versus the social task. In addition, there was an impact of the ratio of human to nonhuman agents per group on conformity on the social task, with higher conformity with the group opinion as the number of humans in the group increased. No such effect was observed for the analytical task. The findings suggest that mixed groups produce different levels of conformity depending on group composition and task type. Designers of systems should be aware that group composition and task type may influence compliance and should design systems accordingly.