Compete, Cooperate, or Both? Exploring How Interaction Settings Shape Human-Robot Interaction

Robots and humans interact in different settings, typically encompassing tasks with cooperation and competition. However, research in human-human interaction indicates that a combination of both might be most beneficial for enjoyment and task performance. Moreover, as the cooperative competition builds a potential in-group with the robot and a potential out-group of another team, mind perception might be influenced. In a laboratory experiment participants (N =66) interacted with a social robot in either cooperation, competition, or cooperative competition. The results, despite a successful manipulation check, revealed no significant differences in enjoyment and engagement. However, the robot was ascribed with significantly more agency in the cooperative competition compared to the competition. Moreover, exploratory analyses revealed higher performance in the competitive compared to the cooperative interaction but not compared to the cooperative competition. Our study, while needing validation with larger sample sizes, suggests a potential positive effect of combining cooperation and competition, notably in increasing agency.


INTRODUCTION
In an ever-evolving technological landscape, robots are fnding their place in a growing array of social settings aiming for emotional, cognitive, and physical stimulation of the human counterpart [17].In the so-called edutainment area, (commercial) robots play physically or digitally with or against humans to achieve positive outcomes like engagement [14].In these human-robot interactions (HRI), much of the research centers around fostering cooperative relationships where robots and humans work together [21].Cooperation is generally defned as involving a group of agents working together to attain a common goal [4].In certain tasks, not only cooperation but also competition could serve as an efective interaction setting in HRI [2,11,16].In contrast to cooperation, competition involves interactions in which one agent attempts to outperform the other [4].While both cooperative and competitive settings have been incorporated into diferent HRI scenarios, it is still widely unclear which of both settings and respective goal orientation is more benefcial for the stimulation of the human counterpart.
One of the frst studies comparing cooperation and competition in HRI revealed mixed results in a video game scenario [15].On the one hand, the fndings indicated that participants perceived the cooperating robot as more sociable and intellectual compared to the competing robot.On the other hand, they reported a higher level of enjoyment and engagement during the competitive task in contrast to the cooperative task.In a later study [10] again a lack of clear superiority of one interaction setting was revealed.Even though, participants who competed against the robot made fewer mistakes than those who cooperated with the robot, no diferences in task time, likeability, pleasure, comfort, and rapport were revealed.Moreover, current research indicates that the outcome of the task (winning vs. losing) might be important and that the pure main efect of setting might not occur concerning mind perception of robots [11].Taken together, potential positive efects of competition are highlighted in HRI research, despite the prevalence of a lack of diferences between the two interaction scenarios.
Beyond contrasting cooperation and competition, inspiration for combining elements of both interaction settings can be drawn from research in human-human interactions (HHI).In the realm of HHI, numerous researchers have engaged in debates surrounding the optimal interaction context, contemplating whether they should be organized cooperatively or competitively to foster subjective and objective outcomes [4,8,19].Indeed, some researchers have put forth the argument that a combination of both cooperative and competitive elements, here referred to as cooperative competition, can yield the most favorable outcomes [19].Specifcally, the combination of competition and cooperation resulted in higher levels of intrinsic motivation and improved performance.Cooperative competition proves applicable to not only a wide range of HHI but also HRI scenarios, as it can be easily implemented by facilitating competition between two human-robot teams in a game.Drawing on the results in HHI [19] we hypothesized that enjoyment as well as engagement are greater in cooperative competition compared to competition and cooperation.
Moreover, cooperative competition could also be coined as intergroup competition [19] creating an in-group with the robot and an out-group with another human-robot team.Hence, group identity might play an important role in the perception of the robot during cooperative competition.A meta-analysis in HHI [1] obtained evidence supporting the hypothesis that inter-group bias in cooperation results from in-group favoritism rather than out-group derogation.In particular, the meta-analysis [1] demonstrated that the presence of an out-group is not a prerequisite for the emergence of in-group favoritism in cooperative scenarios.Hence, one could contend that cooperation, as opposed to competition, could already establish a substantial in-group deserving of favoritism.One of the crucial consequences of categorizing a robot as a member of one's social group is the change in perception of the robot's mental attributes [5,6].In particular, humans tend to attribute agency to robots, acknowledging their ability to act independently and make decisions [7].Mind perception and especially agency, in turn, serve as crucial prerequisites for the robot being perceived as a social companion [20].As cooperative competition compromises both the formation of an in-group and an out-group, it might be assumed that it leads to increased mind perception.Based on the combination of the results in HRI and HHI, we hypothesized that mind perception is greater in the cooperative competition condition compared to the cooperation condition compared to the competition condition.
To investigate the efects of interaction setting on enjoyment, engagement, and mind perception in a realistic social HRI, participants interacted with the commercially available robot Cozmo in the reaction game "Quick Tap" [3].Participants were instructed to either cooperate with the robot, compete with the robot, or compete together against another human-robot team.Performance, as well as the social and verbal behavior of participants, were assessed during the interaction for exploratory analyses.Besides subjective measures of engagement, enjoyment, and mind perception, the voluntarily spent interaction time after the game was fnished was used as a behavioral indicator for engagement.

METHODS
The study was approved by the local ethics committee.The preregistration and collected data are accessible via the Open Science Framework (https://osf.io/2q6n9/).

Participants
A sample size of 66 participants was targeted to obtain close to .80 power to detect a medium to large efect size of .40 at the standard .05alpha error probability.Participants were recruited from the participant pool of a local university and received course credit.All 66 recruited participants were included in the data analysis, and they were evenly distributed among three groups, resulting in 22 participants per group.They were predominantly students with a mean age of 25.44 (SD=5.41).Slightly more men (56%) than women (44%) and no non-binary (0%) participants took part in the study.Overall, the participants were rather inexperienced in the interaction with robots as a value around two on a scale from one to fve indicated (SD=1.97;SD=1.09).

Apparatus and Task
The human-robot interaction consisted of the game Quick Tap and is illustrated in Fig. 1.Quick Tap is a go-no-go task and involves two cubes, which operate as buzzers.Both interaction partners are assigned a buzzer and the aim is to tap as fast as possible each time the cubes light up in the same color pattern (go reaction).If the patterns do not match or are completely red, no tap is required (nogo reaction) and a tap is considered an error.The task, which has also been employed in previous HRI experiments [3], was selected for its ideal suitability in facilitating the manipulation of cooperation, competition, and cooperative competition.In the cooperation condition, the participant was instructed to work together with the robot as a team, tapping the cube together to complete a trial.In the competition condition, the participant was informed to contest against the robot, with the winner's tap completing a trial.In the cooperative competition condition, the participant was instructed that both interaction partners needed to tap the cube to complete a trial, aiming to be faster than a fctive team in another laboratory.The task itself was the same in all conditions, i.e., tapping as fast and correctly as possible.
The robot used in this experiment was Cozmo 2.0 from Anki.Cozmo is equipped with wheels for independent navigation and components to interact with the included cubes.Cozmo's head is moving, and its LED eye display efectively communicates various emotions.

Design
The experiment consisted of a one-factorial between-subjects design.Each participant was assigned to one of three conditions.Participants either cooperated with Cozmo, competed against Cozmo, or engaged in cooperative competition with another fctive team.The conditions were manipulated via the task description and framing of the robot.

Dependent Measures
As control variables, trait cooperation, and trait competition were assessed via the Cooperative/Competitive Strategy Scale on a fvepoint Likert scale [18].To assess the success of the manipulation, two custom items asking for perceived cooperation (i.e., How much cooperation did you perceive in the interaction with Cozmo?) and perceived competition ( i.e., How much competition did you perceive in the interaction with Cozmo?) on a fve-point Likert scale from none to very much were included.
For enjoyment and engagement, the respective two subscales of [9] were adapted to Cozmo.Both consisted of four items (e.g., I think interacting with Cozmo is enjoyable.)each of which was rated on a ten-point Likert scale from totally disagree to totally agree.
Mind perception was assessed via the mind perception scale of [13].This scale consists of two subscales (i.e., agency and experience), and all items (e.g., The robot can feel tired.)were answered on an eight-point Likert scale from not at all to very well.Moreover, voluntary time spent with Cozmo at the end of the experiment was used as a behavioral indicator of engagement.
For exploratory analyses, the experimenter noted any peculiarities in the study protocol during the experiment.The occurrences were categorized into seven diferent categories: technical malfunctions (e.g., Cozmo freezing in motion), helping behavior (e.g., positioning the cube), indicating positive emotions (e.g., having fun), indicating negative emotions (e.g., being angry), positive comments on the robot (e.g., the robot being cute), directly addressing the robot (e.g., insulting it), and making assumptions about the robot (e.g., robot being motivated).Moreover, performance was assessed from the output of the code.

Procedure
Participants were assigned counterbalanced to one of the three conditions and signed informed consent at the beginning of the experiment.First, participants were introduced to Cozmo and read an explanation of the game rules.Then, they received a standardized verbal framing and game goal corresponding to the respective condition and performed fve practice trials.Subsequently, two experimental blocks with 25 trials each were performed, highlighting the respective game instructions and the framing before each block.Next, participants answered the questionnaires.Finally, they had the option to interact with Cozmo in another game which was allegedly not part of the experiment.As this should serve as the objective measure for engagement, the experimenter recorded the time participants spent in this voluntary interaction.This was followed by the debriefng about the manipulation of the conditions, and participants were informed about the process of receiving their course credit.The interaction with Cozmo lasted 15 minutes, and the experiment in total lasted 30 minutes.This does not include the voluntary time spent, which difered individually between participants.

RESULTS
All variables were analyzed via one-way between-subjects ANOVAs at a signifcance level of .05.In addition, for all analyses in which a signifcant main efect of the three-level factor interaction setting occurred, Bonferroni-corrected post-hoc tests were conducted.

Exploratory analysis
We conducted exploratory analyses of task performance, with a crucial consideration being the elevated system-related error rates of 9.9% about missing values in participants' reactions.The analysis of the valid reactions revealed a signifcant efect for the achieved score, (2, 63) = 3.73, = .029, 2 = .106.In particular, participants received signifcantly higher scores (=.032) in the competitive condition (=39.90;=5.50) compared to the cooperative one (=33.10;=9.57).No further diferences between the cooperative competition (=34.82;=9.86) and the cooperation (>.999) and competition (=.164) were revealed.On a descriptive level, this was also echoed by the analysis of reaction times with higher reaction times in the cooperative condition (=1.18; =0.17) compared to the competitive one (=1.07;=0.13; =.054), however, the analysis failed to reach the conventional level of signifcance, (2, 63) = 3.07, = .053, 2 = .053.No noteworthy diference was shown between cooperative competition (=1.11; =0.16) and bot cooperation (=.314) and competition (>.999).Lastly, we analyzed the diferent categories in the study protocol including social and verbal behavior the participants showed toward the robot.We did not fnd any notable diferences between the conditions.

DISCUSSION
The purpose of this study was to gain a better understanding of diferent interaction settings and corresponding goal orientations in HRI.Most importantly, we wanted to explore the efects of interaction settings beyond the most commonly studied cooperative one [21] by also incorporating competition and cooperative competition.In line with results in HHI [19], we hypothesized positive efects of cooperative competition compared to pure cooperation and pure competition.In particular, we assumed that cooperative competition leads to the highest enjoyment, engagement, and mind perception compared to both other settings.
It's crucial to note that our manipulation of the diferent conditions efectively infuenced participants' perceived competitiveness and cooperativeness.Nevertheless, most expected efects did not emerge.Of all dependent variables only for mind perception, in terms of attributed agency, a signifcant diference was revealed.Specifcally, the analysis demonstrated a diference between cooperative competition and competition, while no diference was evident between the other comparisons.An essential limitation in this context is that we have not controlled for the outcome of the interaction (win vs. lose).As recently demonstrated, this can afect the attribution of mind, depending on the interaction context [11].Moreover, the research illustrating the interaction efect of interaction setting and game outcome found only diferences in experience but not agency [11].Our fndings did not reveal a signifcant efect of the dimension experience.In general, this fnding aligns with [7] proposition that agency can be achieved by artifcial agents, whereas experience remains predominantly human.This is further supported by a potential foor efect for experience (23% of participants indicated the lowest possible value for experience on all items).For agency, the higher values in the cooperative competition setting than in the competitive one imply that cooperative competition leads to higher perceived robot capacities regarding decision-making and intentionality [13].It is proposed that mind perception enhances our social connectedness, promotes pro-social behaviors, and elevates the signifcance of observed actions, ultimately contributing to improved team performance [11,13,20].
Our exploratory analysis of task performance does not provide support for this assumption, as it indicated that competition resulted in the highest team performance concerning points and trend-wise for reaction time.However, the performance advantage was just shown in comparison to cooperation but not cooperative competition.Crucially, this aligns with previous research that demonstrated a performance beneft of competition over cooperation in HRI [10].However, this outcome should be interpreted with caution.The performance data recorded by Cozmo's Cubes was not adequately precise with approximately 10% missing values.Despite this, it's worth noting that the results support not clearly that cooperative competition could be an interaction setting that potentially combines the advantages of both competition and cooperation as research in HHI suggested [19].Without exaggerating the fndings, it can only be highlighted that cooperative competition tends to afford the robot greater agency when compared to pure competition, without causing a signifcant decline in performance in comparison to a pure cooperative setting.
Nevertheless, the issue to consider is how to make sense of the absence of fndings regarding engagement and enjoyment.One interpretation of these fndings is that -particularly having the successful manipulation check in mind -the interaction setting might not be relevant for enjoyment and engagement.However, in our view, the most compelling explanation for the absence of fndings is that we may have made an incorrect assumption about a medium to large efect size.This could have resulted in our study being underpowered in the end.As stated by [12], if no information is available before conducting an HRI user study, researchers should anticipate small efects and strive for substantial sample sizes.Yet, constrained resources frequently make this impractical.Hence, we advocate considering the predominantly medium efect sizes of enjoyment and engagement uncovered in this experiment as well-informed grounds for future power analysis.Therefore, we would argue that much work remains to be done before a robust understanding of the role of interaction setting in HRI can be established.

CONCLUSION
In summary, our research delved into the subjective and behavioral consequences of various interaction settings in social HRI, extending beyond cooperation to encompass competition and cooperative competition.The results add to the accumulating evidence suggesting that incorporating competitive elements may boost performance.Notably, we found that cooperative competition fosters a higher perception of agency compared to pure competition.This implies that embracing cooperative competition could be a promising avenue for designing edutainment robots as intentional agents.

Figure 1 :
Figure 1: Depiction of the game Quick Tap during a go-trial that was played with the robot Cozmo.