Subgroup Formation in Human–Robot Teams

Loading...

Taking too long?

Reload document

| Open in new tab

Subgroup Formation in Human-Robot Teams

Subgroup Formation in Human–Robot
Teams
Completed Research Paper

Sangseok You
HEC Paris
1 Rue de la Liberation, Jouy-en-Josas,
France
you@hec.fr

Lionel P. Robert Jr.
University of Michigan
105 S. State St., Ann Arbor, Michigan,
USA
lprobert@umich.edu

Abstract

Subgroup formation is vital in understanding teamwork. It was not clear whether
subgroup formation takes place in human–robot teams and what the implications of the
subgroups might be for the team’s success. Therefore, we conducted an experiment with 44
teams of two people and two robots, where each team member worked with a robot to
accomplish a team task. We found that subgroups were formed when team members
identified with their robots and were inhibited when they identified with their team as a
whole. Robot identification and team identification moderated the negative impacts of
subgroup formation on teamwork quality and subsequent team performance.

Keywords: Robots, subgroup, human–robot collaboration, teamwork quality, performance

Introduction

The use of physical robots to support teamwork continues to increase (Gombolay et al. 2014). Robots are
employed in human–robot teams to accomplish teamwork (You and Robert 2018). The National
Aeronautics and Space Administration (NASA) uses remote-controlled robots paired with humans to work
alongside astronauts on space missions (Hoffman and Breazeal 2007). Construction sites and urban search-
and-rescue teams are employing robots for dangerous and labor-intensive tasks. Robots are often viewed
as unique and distinct from other team technologies because of their physical embodiment (You and Robert
2018). Physical embodiment can cause individuals to project identities and personalities onto robots and
view them as humans (Groom and Nass 2007). As a result, individuals can develop strong emotional bonds
with robots (Hiolle et al. 2012).

Emotional bonds with technologies have received much attention from scholars in several domains
including information systems (IS) and human–computer interaction (HCI) (Suh et al. 2011; Vincent 2006;
Zhang 2013). Generally, these emotional bonds have been viewed as beneficial because they promote the
adoption and the continual use of technology (Kim et al. 2010; Robert 2017). Emotional bonds with
technology have also been associated with more engagement, enjoyment, and better performance with
using the technology (Li et al. 2006). Research suggests that individuals develop strong bonds with robots
(Hiolle et al. 2012). Yet, IS scholars have paid little attention to the impacts of emotional bonds with robots.

Although much attention has been directed at the positive outcomes associated with humans’ emotional
bonds with robots and other technologies, little attention has been paid to understanding the potential
drawbacks in human–robot teams. The formation of faultlines may be one such drawback. Faultlines are
hypothetical divisions that can split teams into smaller subsets (Lau and Murnighan 2005). Research on
faultlines has consistently shown that strong bonds within a subdivision of the team relative to the bonds
among all team members can lead to subgroups (see Carton and Cummings 2012 for a review). Subgroups
can divide the team and create discord among team members (Bos et al. 2010). This explains why the
emergence of smaller groups (i.e. subgroups) within teams has been found to undermine teamwork
(O’Leary and Mortensen 2010).

Fortieth International Conference on Information Systems, Munich 2019

1

Subgroup Formation in Human-Robot Teams

Is it possible that humans can form bonds with robots that are strong enough to create subgroups?
According to the literature on robots, people develop emotional bonds with their robots in much the same
way they do with other people (Hiolle et al. 2012; Robert 2018; You and Robert 2018). This suggests that
the human–robot relationships within a team could act as the basis for a faultline. If this is true, would the
emergence of subgroups be detrimental to the success of human–robot teams? Although no study has
explicitly examined faultlines in human–robot teams, one study has found evidence that attachment to
robots can be detrimental to the success of human–robot teams. In explosive ordnance disposal (EOD)
teams, attachment to a robot made operators hesitant to deploy the robot to dangerous missions, and, as a
result, the effectiveness of the EOD team was decreased (Carpenter 2013). These findings suggest that
strong bonds with robots do not always result in positive outcomes.

Despite the importance of faultlines and the subgroups they create in traditional teams, the consequences
of faultlines in human–robot teams remain relatively unexplored. Yet, as robots are increasingly being
designed to elicit social and psychological responses, this becomes vital to understanding how such
reactions are likely to manifest themselves in human–robot teams. We believe this calls for IS scholars to
broaden our view of the role of technology. In this paper, we began to explore the concept of subgroup
formation in human–robot teams with the following research question.

RQ) How does a bond between a human and his or her robot influence the formation of human–
robot subgroups and the subsequent team outcomes?

Our goal is to determine whether the bonds between team members and their robots can act as a faultline
and have negative effects on team outcomes as they do in traditional teams. We examined 44 human–robot
teams. Results offer evidence for both the existence and impacts of subgroups in human–robot teams.

Background

Robots in the Current Study

Although definitions of robots vary widely across disciplines, we highlight the most common characteristic
for this study: the physical embodiment. The physical embodiment distinguishes robots from other
technological agents, such as chatbots, recommendation agents, and avatars (Ziemke et al. 2015). Due to
robots’ physical embodiment, interactions with these robots are qualitatively different from those with
other technological agents because they allow for more visceral and tangible experiences, such as touching
and hitting (Lee et al. 2006). Existing in the same physical space enables people to interact with a robot
“real-time and real-space, here and now” (Dourish 2001, p. 235). We believe that the physical embodiment
is a key property of robots that creates strong bonds between an individual and a robot. Thus, the robots in
this paper were chosen primarily to manifest physical embodiment rather than intelligence, autonomy, or
human-likeness.

Bonds with Technology and Robots

Research has shown that strong bonds with a technology artifact can have important implications for
understanding technology use. The basic premise behind the importance of bonds with technology is that
the stronger the bonds individuals feel to a technology artifact, the more they prefer to use it and enjoy
using it. Typically, the consequences of the bonds with technology are positive. For instance, emotional
attachment has been positively related to individuals’ willingness to use and continue to use their
technology, such as a mobile phone and an online avatar (Kim et al. 2010; Suh et al. 2011). Emotional bonds
with technology can also enhance the quality of interaction. Individuals who built an emotional bond with
avatars felt higher levels of social presence in videogames and a better shopping experience in virtual worlds
(Kim et al. 2015; Suh et al. 2011). Also, emotional bonds with robots were shown to increase the
performance and viability of human–robot teams (You and Robert 2018). However, emotional bonds with
technology can also have adverse effects. People sometimes hesitate to adopt new technology when they
develop strong bonds with their current technology (Read et al. 2011).

Emotional bonds have been crucial to understanding the interaction between humans and robots because
of the embodied presence of robots (Groom and Nass 2007). Physically embodied objects (as opposed to
digital objects) tend to elicit visceral interactions (Schifferstein and Zwartkruis-Pelgrim 2008). Individuals
are often more engaged with physical objects and exert more effort to maintain relationships with them

Fortieth International Conference on Information Systems, Munich 2019 2

Subgroup Formation in Human-Robot Teams

(Lee et al. 2006). This explains, in part, why people can build emotional bonds with physically embodied
agents like robots more easily than virtual avatars (Groom and Nass 2007; Lee et al. 2006).

The importance of human–technology bonds has prompted both IS and HRI researchers to investigate the
conditions that promote these bonds. The degree to which people believe that a technology artifact is a part
of them or represents them has been found to be a facilitator of emotional bonds with technology (Lee and
Sundar 2015). This can come in the form of perception of shared identity between people and their
technology in terms of characteristics (Vincent 2006), personality (Govers and Mugge 2004), and
appearance (Suh et al. 2011). Such bonds can be facilitated when users build, personalize, or customize their
technology artifacts and robots (Lee and Sundar 2015). This is because people often view the artifacts they
create or personalize as representations of themselves (Ahuvia 2005).

Subgroup Formation in Teams

Although the implications of faultlines and subgroup formation have not been studied in the context of
human–robot teams, both have an extensive literature base (Thatcher and Patel 2012). Next, we review the
literature on faultlines and subgroups within teams. We also introduce the idea of faultlines and subgroup
formation in human–robot teams. Specifically, we discuss how the human–robot relationship can act as a
faultline in human–robot teams and lead to subgroup formation.

Faultlines — which represent potential breaks within teams — have been used to explain the formation of
subgroups within teams. These potential breaks can be the basis for division within teams (Lau and
Murnighan 2005). Faultlines have been based on variables such as race, gender, age, and occupation
(Thatcher and Patel 2012). Theories related to faultlines suggest that team members are likely to form
stronger bonds with those they are more similar to (Lau and Murnighan 2005; Li and Hambrick 2005). For
example, in a team with two engineers and two accountants, occupations could potentially be a faultline.
Faultlines increase in strength as the number of similar attributes between members within a subdivision
increases relative to the number of differences across subdivisions (Polzer et al. 2006). In the previous
example, if the two engineers were men and the two accountants were women, the strength of the potential
faultline would increase because it would include gender and occupation.

Subgroups can form when faultlines are activated (Bezrukova et al. 2009; Thatcher and Patel 2012).
Subgroups are a subset of the team consisting of two or more members (Carton and Cummings 2012). The
presence of subgroups has been associated with negative implications. Subgroup formation has been found
to reduce variables that promote teamwork, such as trust and satisfaction, while increasing variables that
are harmful to teamwork, such as conflict (Cronin et al. 2011; Pearsall et al. 2008; Robert 2016a).
Traditionally, subgroup formation has been studied in collocated teams, but recent studies have found the
evidence in virtual teams (Robert 2016b). In many cases, subgroups are formed in geographically
distributed teams when team members form stronger bonds with collocated team members than with
dispersed members (Polzer et al. 2006). Subgroups in virtual teams have been associated with more conflict
and coordination problems, lower trust, lower team identification, and lower-functioning transactive
memory systems (O’Leary and Mortensen 2010; Spell et al. 2011).

Although there are many ways to measure faultlines, the underlying core logic remains similar. When team
members form stronger bonds within subdivisions than with the team as a whole, the team is likely to
fracture into subgroups. The word likely should be emphasized because faultlines do not always lead to
subgroup formation (Homan et al. 2007). Faultlines do, however, provide the basis for which subgroup
formation occurs within teams (Lau and Murnighan 2005). There is still much ongoing research into what
activates faultlines from potential to actual subgroup formation (Jehn and Bezrukova 2010; Pearsall et al.
2008). Most scholars seem to agree that once activated, subgroups can be harmful to effective teamwork
(Jehn and Bezrukova 2010; Thatcher and Patel 2012).

Despite the rich evidence of the effects of subgroups in teams, researchers are only beginning to examine
how robots can contribute to the formation of subgroups and how this can influence team outcomes.
Faultlines are often anchored on the perception of relatively stronger bonds with one than another. Thus,
it is possible that a team of two humans each working with a robot can fracture into two human–robot
subgroups based on the team members’ strong bonds with the robots. We learned from prior research that
people often develop strong emotional bonds with their robots (Friedman et al. 2003; You and Robert
2018). The bonds are as strong as or stronger than the ones people develop with other humans and pets

Fortieth International Conference on Information Systems, Munich 2019 3

Subgroup Formation in Human-Robot Teams

(Krämer et al. 2011; Melson et al. 2009). Therefore, we believe that human–robot bonds can act as a
faultline and, when triggered, lead to subgroup formation. Thus, we investigated the existence of human–
robot subgroups and the implications associated with the outcomes of human–robot teams.

Hypotheses Development

To understand the implications of subgroup formation in human–robot teams, we developed a theoretical
model (Figure 1). The model posits that the relationship between an individual and a robot can be the basis
of a faultline. In particular, this faultline can become activated when individuals in teams establish stronger
bonds with their robots than with a human teammate. The model draws from social identity theory (Hogg
et al. 2004) to describe how identity-based bonds can result in subgroups in human–robot teams. Robot
identification (i.e. identifying oneself with a robot) is likely to engender subgroups based on human–robot
bonds, while team identification (i.e. identifying oneself with the whole team) is likely to reduce the
formation of subgroups. The model also illustrates the moderation effects of both robot and team
identification for teamwork quality. Specifically, subgroup formation decreases teamwork quality when
robot identification is higher than low, whereas it increases teamwork quality when team identification is
higher than low.

Robot
Identification

+ H1

– H3

Human
Robot
Subgroup

– H2

+H4

Team
Identification

Human Robot
Teamwork
Quality

H5

Human Robot
Team
Performance

Figure 1. Research Model
Figure 1 Research Model

Robot Identification and Subgroup Formation

Robot identification is associated with increases in subgroup formation in human–robot teams. Robot
identification can be defined as the extent to which individuals believe their robot is a part of the self (You
and Robert 2018). This can happen because individuals believe that an object and themselves share the
same qualities (Connell and Schau 2013). Robot identification can be viewed as a specific instance of
material identification. Material identification can be represented by the concept of self-extension and
occurs when an individual becomes attached to a material object (Mugge et al. 2009). Self-extension, as it
pertains to material identification, literally means to extend one’s self to include a material object (Belk
2013). Self-extension has been used to explain brand identification (Kim et al. 2001), digital goods (Belk
2013), and avatars (Suh et al. 2011; You and Sundar 2013).

Robot identification represents a strong attachment to the robot when individuals have extended
themselves to include the robot (You and Robert 2018). In the first place, robot identification speaks to the
individual-level psychological process based on the individual’s relationship with his or her robot. However,
in teams working with multiple robots, team members use and control their own robot as part of the
collaboration with other teammates (Yanco and Drury 2004). Thus, the shared experience of identification
with a robot can also be viewed as a team-level phenomenon. In such cases, subgroup formation is likely to
occur when individuals have a stronger relationship within their subdivision than across the team as a whole
(Cronin et al. 2011). In a human–robot team, the relationship between an individual and his or her robot
engenders the subdivision of the whole team. All things being equal, the likelihood of subgroup formation
increases as an individual’s identification with his or her robot increases.

Hypothesis 1: Robot identification increases subgroup formation in human–robot teams.

Fortieth International Conference on Information Systems, Munich 2019 4

Subgroup Formation in Human-Robot Teams

Team Identification and Subgroup Formation

Team identification is associated with decreases in subgroup formation in human–robot teams. Team
identification is defined as “the extent to which members are psychologically identified with a group” (Scott
1997, p. 120). Team identification is derived from social identity theory (Hogg et al. 2004). Social identity
theory helps to explain who we are in comparison to others (Hogg et al. 2004). Team identification occurs
when individuals believe their identity overlaps with the team’s identity (Van Der Vegt and Bunderson
2005). When this occurs, their membership in the team becomes self-defining (Janssen and Huang 2008).

Team identification has been described as the glue that binds team members together (Bezrukova et al.
2009). This is because team identification promotes strong inter-team relationships, in part by leading team
members to view their teammates more positively (Gibson and Vermeulen 2003). For example, team
identification has been found to increase trust toward team members (Han and Harms 2010). Trust is
associated with strong interpersonal bonds among all team members while team identification also
facilitates inter-team relationships by reducing conflict and promoting cohesion and a sense of a shared
faith (Robert and You 2018; You and Robert 2018).

Team identification represents a strong bond to one’s team. Teams with members who are psychologically
bonded to the team as a whole have less chance of fracturing into subgroups (Ren et al. 2014). Conversely,
fractures are likely to occur when individuals have a stronger relationship within their subdivision than
across the team (Dovidio and Gaertner 2000). Thus, when human–robot teams are high in team
identification, there is a greater chance that the relationship between an individual and his or her robot will
be offset by the relationship between the individual and the team. As a result, increases in team
identification should reduce the likelihood of subgroup formation in human–robot teams.

Hypothesis 2: Team identification decreases subgroup formation in human–robot teams.

Moderation Effects of Robot Identification and Team Identification

In this study, we built and tested a theoretical model to explain the impacts of human–robot subgroups in
human–robot teams. We drew on theories related to intra-group bias, subgroup formation, and dual
identification. Theories on the intra-group bias, subgroup formation, and dual identification suggest that a
subgroup’s strength and the presence of a collective identity might help dictate the influence of a subgroup.
Empirically, several meta-analyses offer strong evidence to support these assertions (see Carton and
Cummings 2012 for a review). Taken together, both theory and empirical evidence suggest that the presence
of a collective identity alters how subgroup formation influences teamwork and performance.

Negative Impacts of Subgroup Formation on Teamwork Quality

Generally, the subgroup formation has been found to reduce teamwork quality by increasing divisiveness
within teams (Robert 2016a). Divisiveness degrades teamwork quality by negatively influencing members’
perception of their teammates and the work experience with them. Subgroup formation can lead team
members to make negative attributions about members who are perceived to be outside the subgroup. The
same actions taken by outgroup teammates is likely to be viewed more negatively when subgroups are
formed (Cronin et al. 2011). This reduces intergroup relationships among team members and heightens the
potential for conflict (Bezrukova et al. 2009). When this occurs, team members are likely to experience
more frustration, anxiety, and discomfort (Lipponen et al. 2003; Polzer 2004). Increases in these factors
should degrade teamwork quality.

Subgroup formation is also likely to reduce the motivation of team members. Team members are less likely
to exert much effort when they do not trust the actions taken by outgroup teammates (Kameda et al. 1992).
Research on effort-withholding in teams has repeatedly shown that negative attributions to one’s teammate
or poor inter-team relationships are largely associated with members putting less effort toward the team’s
objectives (Srinivasan et al. 2012).

In the following, we build on the prior literature on identity theory and subgroup formation to explain the
moderation effect of subgroup formation for the relationship between different identification mechanisms
and team outcomes in Hypotheses 3 and 4. Specifically, although subgroup formation generally decreases
teamwork quality, the adverse effects of subgroup formation can be harnessed by different identification

Fortieth International Conference on Information Systems, Munich 2019 5

Subgroup Formation in Human-Robot Teams

mechanisms in teams working with robots. As discussed, the negative effects of subgroup formation have
been observed in research on teamwork (see Carton and Cummings 2012 for a review). The evidence from
research on teamwork offers an explanation to the negative relationship between the subgroup formation
and performance in human–robot teams (Homan et al. 2007; O’Leary and Mortensen 2010).

Robot Identification, Subgroups, and Teamwork Quality

Scholars have identified the strength of subgroups as a critical element in understanding the relationship
between subgroups and teamwork. The logic is simple. As the strength of subgroups increases, so does the
divisiveness in the team. This divisiveness manifests itself in many forms, such as conflict and discord
(Goyal et al. 2008). The subgroup strength has also been shown to alter variables like satisfaction and
performance (Cronin et al. 2011). Subgroups can have positive impacts on teamwork quality by providing
emotional and psychological support to its members. This view of subgroups was originally put forth by
Gibson and Vermeulen (2003). They found empirical evidence that moderate levels of subgroup strength
facilitate learning in teams. They argued that moderate levels of subgroup strength provide team members
with a safe space to engage in the learning process. This was in contrast to when subgroup formation was
weak or strong. When subgroup strength was weak, subgroups had little or no relationship with team
learning. When subgroup strength was strong, subgroups actually decreased team learning by creating a
divisive environment. Other scholars have found evidence that subgroup formation can have positive
impacts on teamwork in the form of social integration, open communications, and perceptions of fairness
(Robert 2016b; Spell et al. 2011).

In a similar vein, robot identification should moderate the relationship between subgroup formation and
teamwork quality in human–robot teams. Specifically, subgroups should hurt teamwork quality when robot
identification is high. When subgroups are formed, increases in robot identification result in increases in
the strength of the subgroup. High levels of robot identification coupled with subgroup formation lead to
the divisiveness associated with subgroups by reinforcing the subgroup boundaries. As many studies have
shown, divisiveness is likely to be associated with decreases in teamwork quality (Thatcher and Patel 2012).
We believe this condition best represents the high levels of subgroup formation observed by Gibson and
Vermeulen (2003). Studies have shown that individuals can build a strong bond with a robot that surpasses
teamwork among human teammates. For instance, because of strong bonds with a bomb disposal robot,
soldiers sometimes do not want to deploy the robot to dangerous missions that expose the robot to the risk
of destruction (Carpenter 2013).

Conversely, subgroups should have a positive impact on teamwork quality when robot identification is low.
We believe this condition represents the low levels of subgroup formation observed by Gibson and
Vermeulen (2003). In their study, subgroup formation revealed positive effects on the team learning
experience. We expected a similar mechanism to be at play in teams with subgroups and low levels of robot
identification. Subgroups based on human–robot bonds might provide a comfortable and enjoyable
interaction with robots without interfering with the team’s collective, collaborative effort. Therefore, when
robot identification is low, increases in subgroup formation should lead to increases in teamwork quality.

Hypothesis 3: Robot identification moderates the impact of subgroup formation on teamwork
quality. Subgroup formation decreases teamwork quality when robot identification is high and
increases teamwork quality when robot identification is low.

Team Identification, Subgroups, and Teamwork Quality

Theories related to dual identification posit that individuals seek to be a part of smaller subgroup while
being a part of a larger collective identity (Hogg et al. 2004; Richter et al. 2006). According to these theories,
a collective identity could alter the effects of a subgroup. A collective identity can act as a unifying force that
can bridge the divide between subgroups (Lau and Murnighan 2005; Ren et al. 2014). The bridging can
allow both subgroups to exist along with a broader collective identity without the negative effects
traditionally associated with subgroups (Ren et al. 2014). Several studies examining the influence of
subgroups have found evidence of the benefits of a collective identity in teams with subgroups (Thatcher
and Patel 2012).

It is important to note that, in this paper, we go further than simply stating that team identification can
reduce the negative impacts of the subgroup formation. We propose that team identification determines

Fortieth International Conference on Information Systems, Munich 2019 6

Subgroup Formation in Human-Robot Teams

when subgroup formation can have a positive or negative impact on teamwork quality. To do so, we draw
attention to the literature on dual identification. Theories around multiple identifications in organizations
assert that identification with a workgroup can have a positive impact when that identity is nested within a
larger superordinate organizational identity. Richter et al. (2006) examined the impact of workgroup
identification on inter-work group conflict and productivity. They found that when employees identified
with their organization, work identification was associated with decreases in conflict and increases in inter-
work productivity. Van Dick et al. (2008) found that workgroup identification only leads to more job
satisfaction and extra-role behavior when employees also identify with their organization.

Richter et al. (2006) and Van Dick et al. (2008) did not examine subgroup formation. However, given these
findings, we expect that team identification should moderate the impact of subgroup formation in human–
robot teams. Subgroup formation should increase teamwork quality when team identification is high. When
this occurs, teams should benefit from both the emotional support of a subgroup member (i.e. robot) and a
collective identity. However, when team identification is low, the subgroup formation represents the
divisiveness associated with subgroups.

Hypothesis 4: Team identification moderates the impact of subgroup formation on teamwork
quality by increasing teamwork quality when high and decreasing teamwork quality when low.

Teamwork Quality and Human–Robot Team Performance

Last, we posit that teamwork quality increases the performance of human–robot teams. Teamwork quality
is a team’s perception of the interactions involved during the teamwork. Teamwork quality, thus, includes
the degree to which team members enjoy and are satisfied with teamwork and the perceived support from
teammates (Dayan and Di Benedetto 2008; Easley et al. 2003). Prior literature demonstrates the positive
link between the quality of teamwork and the team’s performance (Hoegl and Gemuenden 2001).

Prior studies have not examined the impacts of teamwork quality on performance in human–robot teams
(Oriz et al. 2010). However, they at least provide evidence by examining constructs similar to the perception
of teamwork quality. Hoffman and Breazeal (2007) reported perceived efficiency of collaboration with a
robot as a measure of successful teamwork.

We believe that the general mechanism of teamwork quality increasing team performance can also apply to
human–robot teams. Specifically, positive teamwork quality can lead team members to be more committed
to the teamwork to maintain the ongoing positive experience. Also, good teamwork quality often indicates
good interaction, coordination, and communication among teammates and the effective use of the robots,
all of which contributes to better team performance (Hoegl and Gemuenden 2001).

Hypothesis 5: Teamwork quality increases team performance in human–robot teams.

Method

As part of a larger research project, we conducted a 3 x 1 between-subjects lab experiment with individuals
recruited at a large university in the United States. We randomly assigned 88 individuals (mean age = 23.6,
standard deviation [SD] = 4.1, 42 females) to 44 teams, each of which consisted of two humans and two
robots. Teams were randomly assigned to one of the three conditions in the experiment: robot identification
(15 teams), team identification (14 teams), and control (15 teams). Each participant was paired with a robot
to accomplish a team task. Participants were paid $20 and received additional compensation based on their
team performance against other teams in the experiment.

Experimental Task and Robots

The objective of the task was to move water bottles from one point to another point (Figure 2). There were
marks on the experimental setting area, indicated as Points A, B, and C. Participants were asked to control
their robots to move five water bottles from Point A t