Noname manuscript No. (will be inserted by the editor) I-C-E Framework: Concepts for Group Dynamics Research in Human-Robot Interaction Revisiting Theory from Social Psychology on Ingroup Identification (I), Cohesion (C) and Entitativity (E) Anna M. H. Abrams · Astrid M. Rosenthal-von der P¨utten. Received: date / Accepted: date Abstract The research community of human-robot in- be co-workers, servants, maybe even companions, and teraction relies on theories and phenomena from the thus, will be integrated in a social network consisting social sciences in order to study and validate robotic of more than only one human interaction partner, and developments in interaction. These studies mainly con- possibly more than only one robotic partner. cerned one (human) on one (robot) interactions in the Research in the field of social psychology investigat- past. The present paper shifts the attention to groups ing human groups shows that group-level processes are and group dynamics and reviews relevant concepts from fundamentally different from individual-level processes. the social sciences: in-group identification (I), cohesion One of the earliest research in this field was done by (C) and entitativity (E). Ubiquitous robots will be part Kurt Lewin who is known as the founder of field the- of larger social settings in the near future. A concep- ory in social sciences [2]. The theory builds upon the tual framework, the I-C-E framework, is proposed as assumption that individual behaviour results from per- a theoretical foundation for group (dynamics) research sonal and environmental factors, and the interaction in HRI. Additionally, we present methods and possible of both. A group influences an individual's behaviour measures for these relevant concepts and outline topics due to the individual's interaction with other members for future research. in a social setting. This is what Lewin called "inter- Keywords human-robot interaction · HRI · group actionism". Group phenomena can only be understood dynamics · ICE framework · ingroup identification · when a group as a whole is studied on the group-level cohesion · entitativity and cannot be fully understood by observing individ- uals, thereby ignoring group influences or social set- tings [3]. An example of group processes that cannot 1 Introduction be explained by individual preconditions is the phe- nomenon of groupthink which is described as an er- The field of Human-Robot Interaction (HRI) has devel- roneous process of decision making due to social dy- oped over the last 20 to 30 years [1]. Studies in this field namics in groups. Janis Irvin analyzed failed decision often involve a human participant and the robot, as an making processes by expert groups, among others the interaction partner. Considering potential, ubiquitous Challenger disaster where security concerns somehow deployments of robots and their integration in daily got lost in the decision process which finally led to future life, the scenario of one human and one robot the tragic decision the spacecraft was ready for flight. interacting with each other in isolation will be unreal- His explanation of this erroneous decision making is "a istic in most of the deployment scenarios. Robots will mode of thinking that people engage in when they are A. M. H. Abrams deeply involved in a cohesive in-group, when the mem- Chair Individual and Technology, RWTH Aachen University bers' strivings for unanimity override their motivation E-mail: [email protected] to realistically appraise alternative courses of action" A. M. Rosenthal-von der P¨utten (1982, as cited by [4]). In other words, groupthink ap- Chair Individual and Technology, RWTH Aachen University pears in groups that are highly cohesive which leads E-mail: [email protected] group members to commit mistakes because agreement 2 Anna M. H. Abrams, Astrid M. Rosenthal-von der P¨utten. is set as the highest goal. The example of groupthink tions (e.g., [12{14]). Motion in human groups has to be shows why research on a group-level is so important. interpreted in real time to anticipate future actions of In consideration of findings from group dynamics re- human group members and synthesize the robots` own search and the likely event of robots playing a part in motion accordingly [15]. social settings in the near future, research on robots and groups is necessary. Still, groups that interact with 1.1.2 How Humans Perceive Robot Groups robots or robot groups are seldom studied, although the need for a paradigm shift has been acknowledged [5,6]. Other research groups concentrate on humans percep- tion of entitativity of robot groups or social effects in minimal group paradigms. In online studies featuring 1.1 HRI and Group Research pictures or videos of single and groups of robots, Fraune and colleagues [16] examined when a quantity of robots Many different fields and subjects of study have emerged is perceived as a group and found that number, type, that are involved in group research with artificial agents. similar colour and synchronized behaviour lead to higher Most research has addressed technical challenges to en- groupness (entitativity) perceptions of the observed robots. able robots to identify, keep track of and attend to mul- Synchronicity in movement and similarity in appear- tiple humans in interactions. It has been researched in ance in a group of robots was found to be perceived online studies and in interaction studies how humans more negatively, i.e. threatening [17]. Eyssel and col- perceive and evaluate robot groups and whether hu- leagues further investigated the impact of social catego- mans tend to prefer robots that were marked as in- rization of social robots on its perception. When a robot group members. In laboratory settings and observa- is presented as an in-group member (German partici- tional studies in the field, researchers explored how in- pants evaluate robot developed in Germany in contrast teraction in dyads deviate from interaction in groups to fabricated in Turkey) participants show an in-group involving robots and humans with the goal to derive rel- bias evaluating the robot developed in Germany more evant concepts that need further investigation such as favourably [18]. Moreover, the ingroup robot was iden- emotional climate, social signal modelling, group norms tified to be psychologically closer, warmer and having and so forth. In addition, robots have been used to pos- more mind [18]. The social category of robots was also itively shape interactions between humans. investigated in a study done by H¨aringand colleagues [19]. Their results support earlier findings by Eyssel and 1.1.1 Technical Solutions to Handle Multiple Users Kuchenbrandt: the in-group robot was more positively evaluated and anthropomorphized to a higher extent. New fields of research in HRI and other disciplines such Taijfels minimal group paradigm [20] also shows ef- as computer vision emerged to find technical solutions fect in HRI. Participants who were assigned into a blue to perception of and behaviour in multi-agent interac- group together with a robot showed an extent of anthro- tions. For instance, regarding computer vision, a robot pomorphic inferences about the robot and more positive has to identify multiple objects of interest [7], e.g., po- evaluations compared to participants in the condition tential interaction partners, decide upon relevant ob- were the robot was not in the same group [21]. Simi- jects [8], and keep track of these relevant objects/people larly, intergroup bias can significantly affect how close [9]. Scholars from the field of social signals processing humans are approaching an in-group robot and how come into play when a robot takes on the challenge to much they trust the robots suggested answers with re- recognize and interpret social behaviours shown by the spect to task difficulty [22]. However, this research on identified interaction partners. Human communication perception of robot groups (online studies) and minimal is multi-modal (verbal and nonverbal). 60-65% of com- group paradigms (one robot grouped with one human) munication is nonverbal [10]. Using diverse mechanisms has so far not been conducted in group settings that go and nonverbal cues, we are able to structure our conver- beyond the dyad. sations, regulate turn taking, establish conversational roles, and convey intentions and emotional states and 1.1.3 Group Dynamics in Human-Robot Groups so forth. Especially communication management (e.g., turn-taking, back-channeling) and relational communi- First studies have begun to examine group dynamics in cation [11] are heavily based on nonverbal messages and interactions between humans and robots. The research have been addressed in HRI dyads for years. Hence, re- predominantly looks into how group interactions differ search groups now shifted to work on realising attention from interactions in dyads. For instance, engagement or management, turn-taking gaze behaviour and other so- disengagement with the interaction is expressed differ- cial gaze behaviour for robots in multi-party interac- ently according to the type of interactions and changes I-C-E Framework for HRI 3 across the group size in HRI [23]. Hence, robots should general, the field of HRI is in need of well-studied and have different prediction models and, depending on the validated sociological and psychological phenomena and number of people around it, use the most appropri- concepts to work with. As Irfan and colleagues describe, ate (p.104). Alves-Oliveira and colleagues developed a many researchers in this field are engineers who begin framework to distinguish individual-level and group- with building and designing a robot. When it comes level emotional expression in interactions in HRI and to testing the robot in a study, some researchers are introduced the concept of emotional climate to HRI confronted with their lack of training and experience in [24]. Observational research of robots interacting with fundamental theories of the social sciences and empir- human individuals and groups in the wild show ample ical methodology [32].