It Moves! It Talks! It’s Alive?!
How Robot Characteristics Influence Psychological Responses and Robot Acceptance
Marieke Wieringa
10857591
Master Thesis
Graduate School of Communication
Communication Science (Research Master)
University of Amsterdam
Supervisor: Dr. R.J. Kühne
Abstract
The primary task of robots so far has been to assist humans in industrial and other
professional settings. However, robots are increasingly being designed for the purpose of
communication with humans in the home environment; to provide company for our elderly,
entertain our children and even serve as therapists. This has raised the need to investigate how
people respond to robots and what factors lead to their acceptance. The goal of this study is to
test the effects of robot characteristics on psychological responses to the robot and the
acceptance of the robot. The extent to which people had control over the robot as well as the
vocal expressions of a robot were manipulated in an experiment on human-robot interaction
(N = 92). The results showed that having high control over the robot during the interaction
led to higher perceived task performance and higher perceived ease of use compared to
having low control over the robot. However, having low control over the robot during the
interaction resulted in higher mind attribution than having high control. Furthermore,
interacting with a robot with the capability of vocal expression led to higher perceived
animacy, higher mind attribution and higher perceived human likeness compared to
interacting with a robot without this capability. Finally, perceived ease of use and perceived
human likeness were related to robot acceptance. These findings contribute to the
understanding of how robots are evaluated during human-robot interaction. Furthermore, it
contributes to knowledge on which factors play a role in the acceptance of social robots.
Keywords: social robots, human-robot interaction, control, vocal expressions, perceived task performance, perceived ease of use, animacy, anthropomorphism, robot
It Moves! It Talks! It’s Alive?! – How Robot Characteristics Influence Psychological Responses and Robot Acceptance
Robots are increasingly being used in settings other than just industry, such as
education, therapy and even in the home environment. These robots do not include merely
functional robots such as robotic vacuum cleaners, but also include robots that are designed to
communicate with humans in order to entertain, emotionally engage and even serving as pets
or companions (Fong, Illah & Nourbakhsh, 2002). These robots are usually referred to as ‘social robots’. According to Libin and Libin (2004), communication between humans and social robots can lead to several benefits, such as elevated mood and well-being. For example,
robotic seal Paro has been shown to improve mood and communication of both children and
elderly (Wada et al., 2005).
These benefits are only enjoyed when people frequently interact with the robot (Libin
& Libin, 2004). This has raised the need to investigate how individuals respond to robots and
what factors might lead to their acceptance (Beer, Prakash, Mitzner & Rogers, 2011). This
paper aims to contribute to such knowledge by investigating how robot characteristics affect
utilitarian and hedonic responses to the robot, and how these responses eventually influence
robot acceptance. Specifically, this paper investigates the influence of two robot
characteristics on psychological responses in human-robot interaction: the robot’s
controllability and vocal expression. Utilitarian responses include the perceived task
performance and perceived ease of use of the robot. Hedonic responses include perceived
animacy and anthropomorphism. Accordingly, we aim to answer the following research
question:
RQ: How do the level of control over a robot and the vocal expressions of a robot during human-robot interaction affect utilitarian and hedonic responses to the robot, and how do these responses influence the acceptance of the robot?
This study is relevant for three reasons. First, robot designers try to design their robots
as user-friendly as possible (Beer, Prakash, Mitzner & Rogers, 2011). Other than that,
designers often also try to design their robots in a way that it seems lifelike, in order to
emotionally engage its users (Bartneck, Kulic & Croft, 2009). It is therefore of interest to
robot designers to know which robot characteristics contribute to user-friendliness as well as
the apparent animacy and human likeness of robots. Secondly, even though there are a lot of
different ways to exert control over a robot, studies on how different control methods
influence factors of robot acceptance are currently lacking (Beer, Prakash, Mitzner & Rogers,
2011). Third, in current studies on responses towards a robot (such as, for example,
anthropomorphism), people do not actually interact with a robot (e.g Eyssel & Kuchenbrandt,
2011; Eyssel, Kuchenbrandt & Bobinger, 2011; Eyssel et al., 2012). We therefore still know
little about the role that these responses play in actual human-robot interaction and eventually
in the acceptance of the robot. This paper aims to close this research gap by studying
utilitarian and hedonic responses to a social robot in an experimental study including actual
human-robot interaction.
The next section gives a detailed description of social robots, the different types of
social robots, their functions within current society and important characteristics of the social
robot. Then, we describe how the level of control over robots are expected to influence the
utilitarian responses perceived ease of use and perceived task performance. Then, the
expected influence of controllability and vocal expression on the hedonic responses perceived
animacy and anthropomorphism will be described. Finally, we will discuss how perceived
ease of use, perceived task performance, perceived animacy and anthropomorphism are
Robot Types
The term robot is defined by the Merriam-Webster dictionary as “a machine that looks
like a human being and performs various complex acts of a human being, such as walking or talking”. In scientific literature, there is agreement that there are broadly two types of robots. The first type of robot, most often referred to as ‘industrial robots’ or ‘professional service robots’, operate in industrial and other professional settings such as the military. These robots are usually fully computer controlled and do not necessarily resemble a human (Thrun, 2004;
Libin & Libin; 2004). The second type of robots, usually referred to as ‘social robots’, are
specifically designed to communicate with humans in more domestic settings, and more often
resemble a human being (Zhoa, 2006). The latter type of robot is the main focus of this paper.
Social robots function in a variety of settings where they fulfil various tasks. These
settings include amongst others education and therapy (Libin & Libin, 2004). In education,
social robots are being used to teach children and young adults basic programming skills
(Fong, Illah & Nourbakhsh, 2002). In therapy, social robot KASPAR resembles a child and
aids in therapy for autistic children (Höflich, 2013), while robotic seal Paro is designed to
stimulate interaction amongst elderly (Wada et al., 2005).
The specific type of social robot that is the focus of this paper however, are those
robots that are designed for domestic settings where they serve as pets or companions. An example of such a robot is Sony’s AIBO, a robotic dog that learns through interaction with humans (Fong, Illah & Nourbakhsh, 2002) and humanoid robots RoboSapien and MiP
(Behnke, Müller & Schreiber, 2005). The main goal of these robots is to entertain and provide
company at home (Libin & Libin, 2004). This has important implications for these robots.
First, the fact that these robots are designed for the purpose of entertainment makes
these robots not just utilitarian, but also hedonic products. In other words, these robots are not
2011). Therefore, these robots do not only evoke utilitarian responses, but also by hedonic
responses (Libin & Libin, 2004). De Graaf and Allouch (2011) define utilitarian responses as
those responses tied to the utility of the robot, whereas hedonic responses relate to the
experience of the user while using a robot. Hedonic responses have no obvious relation to
task-related goals such as the utilitarian responses.
Furthermore, in order to obtain their goal of communicating with and entertaining
humans, these robots (and social robots in general) usually require two things. First, the social
robot requires at least some level of autonomy rather than being fully controlled by the human
like industrial robots (Thrun, 2004). Autonomy is defined as the extent to which a robot can
sense its environment, plan and act based on that environment with the intent of reaching
some task-specific goal without external control. The opposite of autonomy is therefore full
human control. (Beer, Fisk & Rogers, 2014). The second characteristic social robots require is
an interface through which they can communicate with humans, for example speech and vocal
expression.
This paper poses that these robot characteristics (the level of control over the robot and the robot’s vocal expressions) influence utilitarian and hedonic responses towards the robot. We furthermore pose that utilitarian and hedonic responses in turn influence robot acceptance.
The next section focuses on explaining how the robot characteristics are expected to influence
utilitarian and hedonic responses toward robots. Then, we will explain how utilitarian and
hedonic responses influence robot acceptance.
Utilitarian Responses
Perceived task performance. One of the most applied models used to explain the acceptance of technology is the Technology Acceptance Model (TAM: Davis, 1989). The
TAM posits that acceptance will be predicted by two utilitarian considerations: perceived
the degree to which a person believes that using a particular system would enhance his or her
performance (Davis, 1989). In the case of robots, perceived usefulness could therefore refer to
how well they themselves perform on a task (Beer, Prakash, Mitzner & Rogers, 2011).
According to Davis (1989), perceived ease of use and perceived usefulness will be
influenced by specific characteristics of the technology. In the case of robots, the level of
control over the robot will determine which tasks a robot is able to perform (Beer, Prakash,
Mitzner & Rogers, 2011). Methods of control where the human is in control over the robot,
such as control by remote, are used extensively with robots in industrial and professional
settings, such as in planetary exploration and search and rescue actions and is therefore
currently an important default mode of control for robots (Lampe & Chatila, 2006). Humans
generally feel motivated to be feel in control over their environment (White, 1959).
Experiencing a lack of control can therefore lead to several negative outcomes such as stress,
frustration and poor performance (Burger, 1985). Even though very little studies tested the
effect of the controllability of a robot on its perceived task performance, a series of studies did
show that the controllability of characters in a video game influenced feelings of competence
within the game. Specifically, feelings of competence in the game were enhanced when
participants could easily control the characters compared to when the characters were more
difficult to control (Ryan, Rigby and Przybylski, 2006). In these studies, competence was
defined as feelings of effectance, and is therefore similar to performance. Accordingly, we
pose the following hypothesis:
H1: Having high control over a social robot will lead to better perceived task performance than having low control over a social robot
Perceived ease of use. The second utilitarian variable, perceived ease of use, refers to the degree that a person believes that using a system would be free of effort (Davis, 1989).
difficult to use and therefore feel apprehensive to use it. The effect of the level of control over
a robot on perceived ease of use has not yet been studied (Beer, Prakash, Mitzner & Rogers,
2011). However, as stated previously, feeling out of control can lead to several negative
outcomes (Burger, 1985). The reason that lacking control can lead to negative outcomes is
that people feel generally motivated to function effectively within their environment by
reducing uncertainty about it. This motivation to feel in control over your environment is
referred to as effectance motivation (White, 1959). Lacking control however creates feelings
of uncertainty. According to Davis (1989), perceived ease of use relates to the effort put into
using the technology. If using certain technology increases feelings of uncertainty, it takes
more effort to use the technology compared to when people feel in control of the technology
(Luczak, Roetting & Schmidt, 2003). Therefore, we pose the following hypothesis:
H2: Having high control over a social robot will lead to increased perceived ease of use compared to having low control over a social robot
Hedonic Responses
Animacy. Robot designers often try to make their robots as lifelike as possible (Bartneck, Kulic & Croft, 2009). According to the Merriam-Webster dictionary, “animate” means “the state of being alive”. Detecting animate entities has been necessary for survival in order to distinguish prey and predator (Pratt, Radulescu, Guo & Adams, 2010). Infants start
distinguishing between animate and inanimate objects when they are only nine months old
(Poulin-Dubois, Lepage & Ferland, 1996). In scientific literature, there are two contrasting
hypotheses that explain the attribution of animacy to objects.
The first hypothesis, the Newtonian violation hypothesis, poses that animacy is
attributed to objects whose motion violates Newtonian laws of motion (Scholl & Tremoulet,
2000). This means that people will attribute animacy to an object if it stops or starts moving
(1982) show that certain movements can indeed influence the perception of animacy (in:
Scholl & Tremoulet, 2000, p. 304). These motions include a start from rest, a change of
direction or moving in a direct path towards an object. Furthermore, Tremoulet and Feldman
(2000) showed in an experiment including a single rigid object moving across a uniform field
that perceptions of animacy were significantly influenced by change in speed and direction.
The second hypothesis, the intentionality hypothesis, states that animacy is attributed when
intentionality is perceived, such as when an object responds to its environment (Tremoulet &
Feldman, 2006). For example, an object is perceived as more animate when it changes
direction in order to avoid an obstacle (Blythe, Miller & Tod, 1999).
The most important condition for the perception of animacy in both the Newtonian
violation hypothesis and the intentionality hypothesis is that motion has to be self-propelled,
rather than driven by an external force (Poulin-Dubois, Lepage & Ferland, 1996). This
assumption would imply that when the movements of a robot are fully controlled by the
human during human-robot interaction, it should be perceived as less animate than when the robot controls its’ own movements. Furthermore, according to the Newtonian violation hypothesis, attribution of animacy should further increase if the robot is also able to stop,
start, turn and accelerate independently without external force (Stewart, 1982), in other words,
without being controlled by the human. Perceived animacy should also increase if the robot is
able to respond to its environment without external control, as stated by the intentionality
hypothesis (Tremoulet & Feldman, 2006). Based on these arguments, we pose the following
hypothesis:
H3a: Having low control over a social robot during human-robot interaction leads to higher perceived animacy than having high control over the social robot.
Besides independent motion, the robot’s vocal expressions should also affect
humans (Hargie, 2010) and most animals also have the capability of vocal expression: the lion
roars, the dog barks, the cat meows etc. Therefore, the capability of vocal expressions is an
important characteristic that increases animate perceptions when applied to robots (Fink,
2012). However, we pose that in a similar way that motion has to be self-propelled in order to
be perceived as animate (Poulin-Dubois, Lepage & Ferland, 1996), vocal expressions should
also be caused by the robot itself rather than an external force, such as a human pressing a
button. Therefore, we pose the following hypothesis:
H3b: Interaction with a low controllable social robot with the capability of vocal expression leads to higher perceived animacy compared to interaction with a low controllable social robot without the capability of vocal expression
Perceived animacy is the degree to which a person perceives the robot as being alive.
However, a second important hedonic response takes things further by describing how we not
only have the tendency to view inanimate objects as being alive, but that we may also
perceive them as being human.
Anthropomorphism. People have the tendency to treat technological devices as if they were human (Luczak, Roetting & Schmidt, 2003). This tendency to attribute human
characteristics, intentions and emotions to nonhuman agents is referred to as
anthropomorphism (Epley, Waytz & Cacioppo, 2007). Examples of anthropomorphism
include attributing a humanlike appearance to nonhuman agents such as deities, or to believe
that computers possess mental capacities or minds of their own and can therefore conspire
against you (Luczak, Roetting & Schmindt, 2003). Gray, Gray & Wegner (2007) showed that
perceiving something as possessing “mind” includes perceiving something as being able to
experience emotions and being conscious of its environment, while at the same time being
capable of making its own decisions. These dimensions of mind attribution are important
anthropomorphism is perceiving nonhuman agents as if they were human (Epley & Waytz,
Akalis & Cacioppo, 2008). Anthropomorphism therefore goes beyond merely describing the
actions of a nonhuman agent as humanlike, but refers to the process of attributing a nonhuman
agent with human characteristics, emotions and a mind of its own. For example,
anthropomorphism occurs when a pet owner goes beyond describing the behaviour of his dog as “affectionate” to infer that “my dog loves me” (Epley, Waytz & Cacioppo, 2007).
Anthropomorphism relates to animacy in the sense that anthropomorphic inferences may
include perceptions of animacy. However, since animate life is not a uniquely human quality,
anthropomorphism goes beyond perceptions of animacy by inferring some nonhuman agent
possesses uniquely human qualities (Epley, Waytz, Akalis & Cacioppo, 2008).
The strength of anthropomorphic inferences people make can differ in different
contexts. Strong forms of anthropomorphism include not only behaving as if a nonhuman
agent, such as a deity, possess human characteristics, but also includes actively endorsing the
belief that the deity possesses these characteristics. Weaker forms of anthropomorphism, such
as cursing at your computer, are more immediate responses and may not necessarily include
the active endorsement of the belief that the nonhuman agent actually possesses human
characteristics (Epley, Waytz, Akalis & Cacioppo, 2008).
Anthropomorphism serves as a mechanism through which uncertainty about
technology is reduced (Luczak, Roetting & Schmidt, 2003) and communication with robots is
facilitated (Duffy, 2003; Fong, Nourbaksh, & Dautenhalm, 2002). Anthropomorphising helps
people rationalise the real or imagined behaviour of nonhuman agents such as robots, by
treating the robot as if it were a rational agent whose actions are governed by choices and
desires (Duffy, 2003). What motivates people to anthropomorphise in a given situation is
described by Epley, Waytz and Cacioppo (2007), who hypothesized several key psychological
earlier in this paper, namely, effectance motivation.
Specifically, effectance motivation is defined as the motivation to interact effectively with one’s environment by understanding it and reducing uncertainty about it (White, 1959). Interacting with nonhuman agents such as technology can lead to feelings of uncertainty,
especially when technology is not functioning properly (Luczak, Roetting & Schmidt, 2003).
According to Epley, Waytz and Cacioppo (2007), knowledge about humans in general, and
about the self in particular, serve as a readily available heuristic for rationalizing the
behaviour of nonhuman agents. Since self-knowledge is developed in childhood before
knowledge about others, it is more readily accessible and more detailed than other-knowledge
(Epley, Waytz, Akalis & Cacioppo, 2008). Epley, Waytz and Cacioppo (2007) pose that this
readily available information about humans in general, and self-knowledge in particular is
used to reduce uncertainty while interacting with nonhuman agents in order to rationalise their
behaviour. Anthropomorphism should therefore be increased when people are faced with
uncertainty, thereby increasing effectance motivation.
Specific characteristics of nonhuman agents such as robots can increase effectance
motivation. One of these characteristics is the apparent predictability of the nonhuman agent
(Epley, Waytz & Cacioppo, 2007). Therefore, interacting with a low controllable agent could
increase effectance motivation and subsequent anthropomorphism compared to interacting
with a highly controllable agent, since lacking control creates uncertainty about their
behaviour (Epley, Waytz, Akalis & Cacioppo, 2008). The findings of a study by Waytz,
Heafner and Epley (2014) support the idea that lacking control over an agent can lead to
increased anthropomorphism. In their experiment, participants using a driving simulation
drove either a normal car or an autonomous car, able of controlling its own speed and
steering. They found that people were significantly more likely to perceive the autonomous
studies on the influence of effectance motivation on anthropomorphism showed that people
were significantly more likely to perceive a robot as possessing humanlike traits (Eyssel &
Kuchenbrandt, 2011) as well as having a mind of its own (Eyssel, Kuchenbrandt & Bobinger,
2011) when they expected to interact with an unpredictable robot compared to a predictable
robot. Therefore, we pose the following hypothesis:
H4: Having low control over a social robot during human-robot interaction will lead to higher levels of anthropomorphism than having high control over the robot, meaning that it will lead to a) higher levels of mind attribution and b) higher perceived human likeness
The second robot characteristic discussed in this paper, the robot’s vocal expressions,
are also expected to influence anthropomorphism for several reasons. As stated previously,
speech and vocal expression is the main mode of communication for humans (Hargie, 2010).
Therefore, it is perceived as a humanlike feature when applied to robots (Fink, 2012).
Furthermore, the ability of a robot to vocally express itself can give an impression that the
robot possesses some level of intelligence, independent thought and even emotions (Beer,
Fisk & Rogers, 2011). The idea that vocal expression can influence anthropomorphism has
also been supported by scientific research. For example, Eyssel et al. (2012) conducted an experiment in which they manipulated the robot’s voice in such a way that is sounded either humanlike or robotlike and found that people were more likely to attribute mind to a robot
with a humanlike voice than a robotlike voice. Furthermore, Waytz, Heafner and Epley
(2014) found that an autonomous vehicle was anthropomorphised more when it had a female
voice than when it had no voice. In these studies, the effect of vocal expression on
anthropomorphism was tested for both a robot that seemed to be functioning autonomously
and an autonomous driving car. Vocal expression in these studies were therefore not caused
motion has to be driven by an internal rather than an external force in order to create
perceptions of animacy (Poulin-Dubois, Lepage & Ferland, 1996), vocal expressions might
influence anthropomorphism under the assumption that it is caused by an internal rather than
an external force. Accordingly, we pose the following hypothesis:
H5: Interacting with a low controllable social robot with the capability of vocal expression will lead to higher levels of anthropomorphism, including a) more mind attribution and b) higher perceived human likeness, compared to interacting with a low controllable social robot without the capability of vocal expression.
So far, we have discussed how the controllability of a robot is expected to affect
utilitarian responses perceived ease of use and perceived task performance, and utilitarian
responses perceived animacy and anthropomorphism. Furthermore, we have discussed how the robot’s capability of vocal expression is expected to affect perceived animacy and anthropomorphism. The next section will now explain how these different responses are
related to the acceptance of the robot.
Influences on Acceptance of the Social Robot
The Technology Acceptance Model (TAM: Davis, 1989) defines acceptance as a
combination of attitudes, intentions and behaviours towards technology. Heerink et al. (2008)
pose that the acceptance of robots consists of their functional acceptance, but also their
acceptance as conversational partners with whom humans could build a potential relationship.
The section below describes how the utilitarian and hedonic responses discussed above are
expected to influence robot acceptance.
Influences of Utilitarian Responses. The TAM predicts that perceived usefulness and perceived ease of use of some form of technology predict the attitude towards using the
technology and eventually the intention to use the technology (Davis, 1989). These
Theory describes important mechanisms that influence whether people will adopt certain
behaviour. According to Bandura (2004), one of the important mechanisms that influences
whether behaviour is adopted are the outcome expectancies. Specifically, people will be more
likely to adapt new behaviour when they believe the behaviour will result in positive
outcomes, such as enhanced performance. According to Davis (1989), perceiving a system as
useful results in positive outcome expectancies about using the system, and will therefore
increase the likelihood someone will use a system.
The TAM has been widely applied and proven to be effective in predicting technology
acceptance in various fields (Lee, Kozar & Larsen, 2003). Since the perceived usefulness is
defined as the believe that using certain technology would enhance performance of the task, in
the case of robots, perceived usefulness relates to how well the robot performs on various
tasks (Beer, Mitzner, Prakash & Rogers, 2011). Findings of scientific research support the
importance of perceived usefulness in predicting robot acceptance. For example, de Graaf and
Allouch (2011) found that the perceived usefulness of a robot indeed significantly predicted
the attitude towards use. Furthermore, Heerink et al. (2010) found that usefulness predicted
both the intention and the actual use of a social robot. We therefore pose the following
hypothesis:
H7: Perceived task performance has a positive effect on robot acceptance
The second predictor of acceptance in the TAM is the perceived ease of use, and has
also been successfully used in scientific research to predict robot acceptance. This prediction
that ease of use positively affects acceptance is based on the concept described in the Social
Cognitive Theory as self-efficacy. Self-efficacy is defined as the belief in one’s ability to
perform a behaviour (Bandura, 2004). According to Social Cognitive Theory, self-efficacy is
one of the most important mechanisms that determine whether someone will adopt certain
system as easy to use increases self-efficacy beliefs, and should therefore have a positive
effect on the acceptance of the technology.
The importance of perceived usefulness in the acceptance of robot has also been
shown by scientific research. A survey by Ezer, Fisk and Rogers (2009) showed that
perceived ease of use of a robot was able to predict the attitude towards accepting the robot in
the home environment for both younger and older adults. Furthermore, a series of experiments
by Heerink et al. (2009; 2010) showed that perceived ease of use significantly predicted the
intention to use the social robot I-cat. Accordingly, we pose the following hypothesis:
H6: Perceived ease of use has a positive effect on robot acceptance
Although the TAM mainly thanks its popularity to the fact that it can be applied in
various fields, it has been criticised for focusing only on two utilitarian factors of acceptance
(Beer, Mitzner, Prakash & Rogers, 2011). In reality, hedonic responses also play a role in the
acceptance of social robots (de Graaf & Allouch, 2011). These responses are discussed below.
Influences of Hedonic Responses. The two hedonic responses discussed in this paper are perceived animacy and anthropomorphism. As discussed previously, anthropomorphism
serves as a function through which uncertainty about nonhuman agents such as robots can be
reduced and communication with them can be facilitated (Luczak, Roetting & Schmidt, 2003;
Epley, Waytz & Cacioppo, 2007). It allows people to establish a humanlike connection with
robots (Epley, Waytz, Akaliz & Cacioppo, 2008). According to Höflich (2013), perceiving a
robot as more humanlike increases familiarity, and familiarity increases liking. Findings of
scientific research support this idea. For example, Waytz, Heafner and Epley (2014) found
that mind attribution to an autonomous vehicle significantly predicted trust in that vehicle.
They also found that people were less likely to blame the car for a crash when they attributed
the car with mind. Furthermore, research has found that people are willing to spend more time
Kuchenbrandt, 2012; Eyssel, Kuchenbrandt & Bobinger, 2011; 2012). Furthermore, de Graaf
and Allouch (2013) found that perceived human likeness significantly predicted whether
people viewed the robot as a potential friend. Overall, these studies show that
anthropomorphism can result in more positive attitudes towards robots. We therefore pose the
following hypothesis:
H8: Anthropomorphism has a positive effect on robot acceptance, meaning that a) mind attribution and b) perceived human likeness positively affects robot acceptance Finally, even though robots may not always be designed as humanlike (for example
robotic dogs), robot designers do often try to design their robots as lifelike as possible. This is
because lifelike creatures have the power to involve people emotionally (Bartneck, Kulic &
Croft, 2008). Studies show that a robot being perceived as animate can lead the robot to also
be perceived as more intelligent, and that this holds true even if the robot does not resemble a
human (Bartneck, Kanda & Mubin, 2009; de Graaf & Allouch, 2011). Libin and Libin (2004)
furthermore found that lifelike robots were perceived as friendly companions. Since perceived
animacy can lead to emotional engagement, it may facilitate acceptance in a similar way as
anthropomorphism: by creating a connection between the human and robot that is perceived
as real (Epley, Waytz & Cacioppo, 2007). We therefore pose the following hypothesis:
H9: Perceived animacy has a positive effect on robot acceptance
In sum, we expect that high control over a robot will lead to higher perceived task
performance and perceived ease of use than low control. However, we expect that having low
control over the robot will lead to higher perceived animacy and anthropomorphism than
having high control. Finally, we expect that perceived task performance, perceived ease of
use, perceived animacy and anthropomorphism will positively affect robot acceptance. These
hypotheses will be tested in an experiment involving the manipulation of robot characteristics.
Method Participants and Design
This study was conducted in April and May 2016 at the University of Amsterdam. 92
students (62 women, 30 men) were recruited at the University of Amsterdam to participate in
a laboratory study on the evaluation of a small humanoid robot. Participants ranged in age
from 18 to 34 years (M = 23.14, SD = 2.95). The study employed a 3x1 design (high
controllability of robot/low controllability of robot/low controllability with vocal
expressions). Participants were randomly assigned to one of these three conditions. The study
was granted ethical approval by the Ethical Committee of the Amsterdam School of
Communication Research at the University of Amsterdam.
Procedure
Participants registered for the study online, and were welcomed in a waiting room
upon arrival. Here, participants were informed that they would take part in a study about the
evaluation of a small humanoid robot. After signing the informed consent form, the
participant was accompanied by the researcher to the lab where the interaction with the robot
took place. The experimenter made sure that the robot was already switched on in the
appropriate mode before the participant entered the lab. In the lab, the experimenter explained
to the participant that the robot works on a self-balancing mechanism, meaning that the robot
would not fall over even if it was pushed. The experimenter demonstrated by giving the robot
a push. The participant was then invited to do the same. By inviting the participants to touch
the robot, this study followed other studies on robot interaction, which also invited
participants to touch the robot (de Graaf & Allouch, 2013; Nomura et al., 2008).
After the participant touched the robot, the experimenter explained to the participant
that he/she would perform a task with the robot, namely making the robot follow a short
the high controllability condition, participants were allowed to practice with the controls, so
that they would feel in control of the robot while performing the task. Participants in the low
controllability conditions were not allowed to practice with the controls. After participants
completed the task with the robot, they were guided to a separate room where they filled in a
questionnaire.
Stimulus Material
In this study, participants interacted with the entertainment robot MiP (short for
Mobile Inverted Pendulum). MiP is a small humanoid robot and has six pre-programmed
control modes, including gesture control mode, free roaming mode and remote controlled
mode. The robot is also equipped with an IR sensor, which allows it to detect obstacles. MiP
furthermore has the ability to produce vocal expressions such as whistling sounds, happy
sounds, surprised sounds and sad sounds. Figure 1 shows an illustration of the robot.
In order to manipulate controllability of the robot, MiP was switched to different
operating modes in the high and the low controllability condition. In the high controllability
condition, participants controlled the robot with an I-pad. This allowed participants to have full control over the robot’s movements. Participants were able to make the robot go forward,
backwards and turn in different directions at their will.
In the two low controllability conditions on the other hand, the robot was switched to ‘free roaming mode’. This mode allows MiP to explore its environment without being
controlled by an external force. When switched to this mode, the robot will move forward and
turn in other directions independently. Its IR sensors allow it to detect and respond to
obstacles in the environment. Upon detection, the robot stops and turns into another direction. However, the direction in which the robot moves and turns can’t be controlled. This means that in the low controllability conditions, participants could only manipulate the robot’s movement by placing their hand in front of the robots’ IR sensors to make it turn in another
direction, but they had no control over the direction in which the robot would turn and drive
towards as a consequence. Furthermore, the robot also changed directions independently at
random moments, making it more difficult for participants to control the direction in which
the robot was driving. A pre-test (N = 15) showed that those in the low controllability
condition (N = 7) felt significantly less in control over the robot than those in the high
controllability condition (N = 8), t(13, 1) = 2.4, p =.032, d = 1.24.
In order to manipulate vocal expression, two different low controllability conditions were employed. In the first low controllability condition the robot’s vocal expressions were turned off (as in the high controllability condition), whereas the robot’s vocal expressions were turned on in the second low controllability condition. In the latter condition, the robot
would make a whistling sound at the start of the interaction and would make content
humming sounds every now and then throughout the interaction. If the robot detected an object (such as when participants blocked the robots’ IR sensors to make him turn), it would make a surprised sound while turning. Furthermore, if MiP would fall (for example if
sound. All these vocal expressions were missing in the other low controllability and I the high
controllability condition.
Measures
After participants completed the interaction with the robot, they filled out a
questionnaire. The questionnaire started with thanking the participants again for their interest
in the study and asking some demographic questions, namely age, gender and nationality.
Then, the utilitarian responses task performance and ease of use were measured. The hedonic
responses animacy, mind attribution and finally human likeness were measured next. Finally,
the acceptance of the robot was measured. How these concepts were operationalized will be
described below. For the complete measurement instruments, see the appendix.
Perceived task performance. To measure the extent to which participants perceived the robot to have performed well on the task (following a certain parkour), we asked them to
what extent they believed the robot succeeded in completing the task. The scale consisted of four items, including “I think the robot performed well on the task” and “I think the robot succeeded in performing the task”. All items were rated on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Because the scale was newly created, a
principal component analyses was conducted in order to check if the items formed a
dimensional scale. The analysis showed that the four items indeed formed a single
uni-dimensional scale: only one component has an eigenvalue above 1 (eigenvalue 3.21) and there
was a clear cut-off point in the scree plot after one factor. Internal consistency of the scale
was good (Cronbach’s alpha = .92).
Perceived ease of use. Perceived ease of use was measured using a scale by Heerink et al. (2010). This scale was developed to measure the perceived ease of use of robots
specifically. The scale consists of five items, including “I think I will know quickly how to use the robot” and “I find the robot easy to use”. Items were measured along a 7-point scale
ranging from 1 (strongly disagree) to 7 (strongly agree). Internal consistency of the scale was good (Cronbach’s alpha = .80).
Animacy. Animacy was measured using a semantic differential scale developed and validated by Bartneck, Kulic & Croft (2009). The scale was developed to measure the
animacy of robots specifically. The scale consisted of six items, including dead/alive and
artificial/lifelike. Participant rated the items along a 7-point scale. Internal consistency of the scale was sufficient (Cronbach’s alpha = .74).
Mind attribution. Mind attribution was measured using an adapted version of the mind attribution scale by Kozak, Marsh and Wegner (2006). Participants rated the robot on 10 mental capacities such as “this robot can experience pain”, “this robot is capable of emotion” and “this robot has the capacity to plan actions”, “this robot is capable of doing things on purpose”. Participants rated the items along a Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). The original scale consists of two subscales. However, this study
follows previous research where the scales were combined into one (Eyssel et al., 2012;
Waytz, Heafner & Epley, 2014). The internal consistency of the overall scale was excellent (Cronbach’s alpha = .90).
Human likeness. Human likeness was measured using a semantic differential scale developed and validated by Bartneck, Kulic & Croft (2009), which was developed
specifically to measure the human likeness of robots. Examples of items include fake/natural
and machinelike/humanlike. Participants rated the items along a 7-point scale. The complete
original scale consisted of five items and had a Cronbach’s alpha of .75. However, the
reliability analysis showed that removing one item (moving rigidly/moving elegantly)
increased reliability of the scale. This item was removed, since the robot also moved slightly
controlled manually (high controllability condition). The final scale therefore consisted of four items, internal consistency of the scale was reliable (Cronbach’s alpha = .76).
Acceptance of robot. Heerink et al. (2008) propose that robot acceptance includes functional and social acceptance. Therefore, we measured both the attitude towards using the
robot (functional acceptance), as well as the extent to which the robot was perceived as a
potential friend or companion (social acceptance). Attitude towards use of the robot was
measured using a scale by Heerink et al. (2010). The scale consisted of four items, including “I think it’s a good idea to use the robot” and “It’s good to make use of the robot”. Items were measured on a Likert scale, ranging from 1 (strongly disagree) to 7 (strongly agree). Internal consistency of the scale was good (Cronbach’s alpha = .88).
To measure the extent to which people accepted the robot as a companion, we asked
participants whether they could view the robot as a potential friend. To do so, a scale used by Lee et al. (2006) was used. The scale consisted of three items, including “I think I could spend a good time with this robot” and “I think this robot could be a friend of mine”. Participants rated the items along a 7-point Likert scale, ranging from 1 (strongly disagree) to 7 (strongly agree). Internal consistency of the scale was sufficient (Cronbach’s alpha = .75).
Results Analytical Approach
As a first step, ANOVA’s and a chi-square test were conducted to check if the random
assignment of the participants to the three experimental conditions was successful in terms of
age and gender. Then, a one-way MANOVA was performed to see whether controllability
and vocal expressions of the robot affected robot acceptance directly. Using Helmert
contrasts, we first compared the high control condition was to the two low control conditions
together to test the effect of controllability of the robot. Then, the two low control conditions were compared with each other to test the effect of the robot’s vocal expressions. This same
procedure was used to test the effects of controllability and vocal expressions of the robot on
the utilitarian and hedonic response factors in a MANOVA. Additional analyses tested if
effects changed after controlling for age and gender. Finally, the whole model including the
effects of the robot characteristics on the utilitarian and hedonic responses as well as all
effects on robot acceptance was tested using path analysis.
Randomization Checks
Randomization checks showed that there were no significant differences between the three conditions in terms of gender distribution, χ²(2) = 1.19, p t= .553, and participants’ age, F(2, 89) = 1.41, p = .249. These results indicate that the random assignment of the
participants to the three conditions was successful in terms of age and gender.
Effects of Robot Characteristics on Robot Acceptance
First, an exploratory analysis tested whether controllability and vocal expressions
influenced the acceptance of the robot directly. A one-way MANOVA was conducted using
the attitude towards use and the extent to which participants perceived the robot as a potential
companion as dependent variables. The condition variable functioned as the grouping
variable. Using Pillai’s Trace, there was a significant effect of the robots characteristics on robot acceptance (Pillai’s Trace = .16, F(4, 178) = 3.75, p = .006, η² = .08).
The univariate analysis found moderate but only marginally significant differences
on the companionship scores between the groups, F(2) = 2.82, p = 0.065, η² = .06. However,
there were no significant differences between the groups on the attitude towards use, F(2) =
2.29, p = .107. Helmert contrasts were used to check if controllability and vocal expression of
the robot affected the attitude towards use and the extent to which the robot was perceived as
a potential friend or companion. To check the possible effect of controllability, the first
contrast compared the scores of the high controllability condition to the score of the two low
condition scored significantly higher on attitude towards use than those in the low control
conditions (p = .038). However, there was no significant difference in companionship
between the high control and the two low control conditions (p = .848).
The second contrast compared the first low control condition (not including vocal
expressions of the robot) with the second condition of low control (including vocal
expressions of the robot), leaving the high control condition out of the analysis. This contrast
revealed that there was no significant difference in the attitude towards use between the two
low control conditions (p = .732). However, there was a significant difference in
companionship between the two low control conditions: those in the condition without vocal
expression scored significantly lower on the companionship scale than those in the condition
including vocal expression of the robot (p = .02).
Effects of Robot Characteristics on Utilitarian and Hedonic Responses
The same procedure was used to test the effects of controllability and vocal
expression on the utilitarian and hedonic responses. A one-way MANOVA was conducted
using perceived task performance, perceived ease of use, perceived animacy, mind attribution
and perceived human likeness as the dependent variables. The condition variable was used as the grouping variable. Using Pillai’s Trace, there was a strong significant effect of robot characteristics on the utilitarian and hedonic responses (Pillai’s Trace = .74, F(10, 172) = 10.03, p < .000, η² = .37). Table 1 shows the mean scores of the group for each variable.
Hypothesis 1 predicted that having high control over the robot would lead to higher
perceived task performance than having low control. The effect size of the univariate analysis
indeed revealed large significant differences between the groups on perceived task
performance (F(2, 89) = 67.10, p < .000, η² = .60), In support of hypothesis 1, the first
contrast (comparing the high control condition to the two low control condition showed that
than those in the low controllability conditions (p < .000), while there was no difference in
perceived task performance between the two low controllability conditions (p = .643).
Hypothesis 2 predicted that having high control over the robot would lead to higher
perceived ease of use than having low control over the robot. The univariate analysis found
large significant differences between the groups on perceived ease of use (F(2, 89) = 16.10,
p < .000, η² =.27). The first contrast comparison showed that those in the high controllability condition scored significantly higher on perceived ease of use than those in the low
controllability conditions (p < .000). There was no significant difference in perceived ease of
use between the two low controllability conditions (p = .155). Hypothesis 2 is therefore
supported. Hypothesis 3a predicted that having low control over the robot would lead to
higher perceived animacy than having high control. Furthermore, Hypothesis 3b predicted
that interaction with a low controllable robot with the capability of vocal expression would
lead to higher perceived animacy than interaction with a low controllable robot without this
Table 1
Means Scores with Standard Deviations of High Control and Low Control Conditions
Condition
High control Low control Low control (vocal expression) Dependent variable M (SD) M (SD) M (SD) Task Performance 5.85 (.89) 3.44 (.98) 3.33 (1.00) Ease of Use 5.72 (.86) 4.30 (.95) 4.67 (1.19) Animacy 4.03 (1.04) 3.82 (.70) 4.30 (.77) Mind attribution 1.91 (.70) 2.16 (.99) 2.79 (1.12) Human likeness 2.45 (.89) 2.29 (.95) 2.90 (.88)
Attitude towards use 4.96 (1.16) 4.45 (1.20) 4.32 (1.33)
capability. However, the univariate analysis found only small and marginally significant
differences between the groups on perceived animacy (F(2, 89) = 1.75, p = 0.94, η² = .05).
The first contrast comparison found no significant difference in perceived animacy between
the high control and the low control conditions (p = .868). Hypothesis 3a must therefore be
rejected. Results of the second comparison revealed that the first condition of low control, not
including vocal expression of the robot, perceived the robot as significantly less animate than
in the second condition of low control including vocal expression (p =.030). This result
supports hypothesis 3b1.
Hypothesis 4 predicted that having low control over a robot would lead to higher
anthropomorphism than having high control over a robot. Hypothesis 4a specified that having
low control over the robot would lead to higher mind attribution than having high control over
the robot. Furthermore, 5a predicted that vocal expressions of the robot would increase mind
attribution. The univariate analysis found large significant differences between the groups on
mind attribution (F(2, 89) = 6.35, p = .002, η² = .14). Results of the first contrast comparison
showed that those in the low control conditions scored significantly higher on mind attribution
than those in the high control conditions (p = .009). Hypothesis 4a is therefore supported.
Furthermore, the second contrast comparison showed that those in the low control condition
with vocal expressions of the robot scored significantly higher on mind attribution than in the
low other low control condition not including vocal expression (p = .011). This supports
hypothesis 5a2.
Finally, hypothesis 4b predicted that having low control over a robot would lead to
higher perceived human likeness than having high control over a robot. The univariate
analysis found moderate significant differences between the groups on perceived human
1 These results remained significant after removing outliers and controlling for age and gender. 2 The results for mind attribution remained the same after controlling for age and gender
likeness (F(2, 89) = 3.13, p = .026, η² =.08). However, the results showed that people did not
perceive the robot significantly more humanlike in the low control conditions compared the
high control condition (p = .469). Hypothesis 4b must therefore be rejected. Hypothesis 5b
predicted that vocal expressions would increase perceived human likeness. The second
contrast showed that those in the low control condition including vocal expression of the
robot perceived the robot as significantly more humanlike than those in the low control
condition not including vocal expression of the robot (p = .009). This result supports
hypothesis 5b3.
Full Model
In order to test how the utilitarian and hedonic factors as well as controllability and
vocal expressions of the robot influenced the attitude toward using the robot and the extent to
which the robot was perceived as a potential friend, we employed a path analysis using
AMOS 23. Because of the sample size, the variables for the utilitarian and hedonic responses,
as well as the variables for attitude towards use of the robot and the extent to which the robot
was perceived as a friend were entered in the model as observed rather than latent variables.
In order to distinguish between the effects of controllability and the effect of vocal expression,
the low controllability not including vocal expression was used as the reference category. For
the other two groups (high controllability, low controllability with vocal expression), two
dummy variables were created and entered into the model. The first dummy variable
contained all participants that had high control over the robot. The second dummy variable
contained all participants that had low control over the robot with the capability of vocal
expression. Zero-order correlations between all variables entered in the model can be found in
3 After removing one outlier and controlling for age and gender, the univariate ANOVA for perceived human likeness became only marginally significant, F(2, 84) = 2.07, p = .066. The effect of the condition variable remained significant, F(2.84) = 3.30, p = .042. Results of the planned contrasts remained the same (contrast 1: p = .424, contrast 2: p = .018). Furthermore, a significant main effect of gender was found, F(1, 84) = 5.32, p = .023. Women were more likely to perceive the robot as humanlike (M = 2.64, SD = .89) than men (M = 2.26, SD = .84).
Table 2. The data had no problems with multivariate normality, Mardia’s coefficient stayed below the critical cut off value of 1.96 (Mardia’s coefficient = -3.10, critical ratio = -1.06). Therefore, we could proceed with the analysis.
First, we estimated a model with all hypothesized paths from the dummy variables
representing high controllability and low controllability including vocal expression to the
mediators, and from the mediators to the indicators of robot acceptance attitude towards use
and companionship using maximum likelihood estimation. We furthermore estimated direct
paths from the dummy variables to the indicators of robot acceptance. We allowed the error
terms of the mediators to covary as well as the two error terms of the dependent variables.
The model had acceptable fit, although the values for the RMSEA were still above the desirable threshold (χ²(2) = 3.87, p = .144, χ2/df = 1.94, CFI = .99, RMSEA = .10, 90% CI [.00, .25]). The model showed that all direct paths between the dummy variables representing
the robot characteristics were non-significant, except for the direct path between the dummy
variable for vocal expression and the attitude towards use (b* = -.21, p = .050). We then used
a nested model to test whether the three non-significant direct effect of the robot
Table 2
Means, Standard Deviations and Zero-Order Correlations
M SD 1. 2. 3. 4. 5. 6. 7. 8. 9. 1. High control 2. Vocal exp. .50** 3. Task perf. 4.19 1.50 .78** .41** 4. Ease of use 4.89 1.17 .50** -.13 .65** 5. Animacy 4.05 .86 -.02 .21* .13 .26* 6. Mind attr 2.29 1.01 -.26* .35** -.23* -.05 .25* 7. Humanlike 2.55 .94 -.07 .27** -.003 .16 .64** .53** 8. Attitude use 4.57 1.25 .22* .19 .33** .36** .38** .14 .39** 9. Companion 3.29 1.28 .02 .20 .16 .40** .43** .27** .48** .52** Note. * p < .05. ** p < .01. p < .000.
characteristics on the attitude towards use and companionship could be dropped. The
chi-square difference test showed that dropping these paths did not significantly deteriorate model
fit (χ²difference(3) = 2.01, p = .570). These paths were therefore removed from the model. The
final model is shown in figure 2. The final model had good model fit (χ²(5) = 5.88, p = .318, χ2/df = 1.18, CFI = 1, RMSEA = .04, 90% CI [.00, .16]). Figure 2 shows the final model.
The final model found the same significant effects of the robot’s vocal expression on
perceived animacy (b* = .23, p =.037), mind attribution (b* = .28, p = .013), and perceived
human likeness (b* = .28, p = .014) as were found in the ANOVA’s. However, since the full
model compared the high control condition to the first low control condition (not including
Figure 2. Final model with standardized estimates and significance level (covariances between error terms not displayed for reasons of parsimony). χ²(5) = 5.88, p = .318, χ2/df =
vocal expressions) only, the effects of the robot’s controllability differ in the final model compared to the ANOVA’s. For this reason, the model does not show the significant effect of the level of control over the robot on mind attribution (b* = -.13, p = .259), as was found in
the ANOVA. This is however not a problem when interpreting the effects of the utilitarian
and hedonic responses on robot acceptance predicted in the hypotheses.
Hypothesis 6 predicted that perceived task performance has a positive effect on robot
acceptance. However, the model showed now significant effect of perceived task performance
on either attitude towards use (b* = .11, p = .355) or companionship (b* = -.09, p = .444).
Hypothesis 6 is therefore rejected. Hypothesis 7 predicted thatperceived ease of use has a
positive effect on robot acceptance, but ease of use did not have a significant effect on the
attitude towards use (b* = .17, p = .142). However, it did have a significant positive effect on
the extent to which the robot was perceived as a potential friend (b* = .38, p < .000), showing
partial support for hypothesis 7.
Hypothesis 8 predicted that anthropomorphism has a positive effect on robot
acceptance. Hypothesis 8a specified that mind attribution has a positive effect on acceptance.
The model showed that mind attribution did not significantly affect the attitude towards use,
(b* = .07, p = .530) or the extent to which the robot was perceived as companion, (b* = .08,
p = .42). Hypothesis 8a is therefore rejected. However, perceived human likeness of the robot had a significant positive effect on both the attitude towards using the robot (b* = .28, p =
.027) and the extent to which the robot was perceived as a friend (b* =.28, p = .026). These
results support hypothesis 8b, which predicted that perceived human likeness has a positive
effect on robot acceptance.
Finally, perceived animacy was hypothesized to have a positive effect on robot
acceptance in hypothesis 9. However, no significant effects of perceived animacy on either
a potential friend (b* = .15, p = .185) were found. Hypothesis 9 must therefore be rejected.
Surprisingly, the model also showed a significant negative direct effect of vocal expression on
the attitude towards using the robot (b* = -.24, p = .011).
Discussion and Conclusion
The goal of this study was to investigate the influence of two robot characteristics on
utilitarian and hedonic responses towards the robot and robot acceptance. The study showed
significant effects of the degree the participants could control the robot during the interaction.
Having high control over the robot led to higher perceived task performance and perceived
ease of use than having low control. However, having low control over the robot during the
interaction led to higher mind attribution compared to having high control over the robot. Furthermore, significant effects of the robot’s vocal expressions were found. Interacting with a robot with the capability of vocal expression led to higher perceived animacy, higher mind
attribution and higher perceived human likeness compared to interacting with a robot without
this capability. We furthermore found significant effects of these responses on robot
acceptance: perceived ease of use and perceived human likeness were positively related to
robot acceptance. There was also a significant direct negative effect of vocal expressions on
the attitude towards using the robot.
Practical and Theoretical Implications
The findings of this study have important implications for research on the effects of
robot characteristics on psychological responses towards robots in human-robot interaction.
The effects of controllability and vocal expression found in this study were consistent with
previous research on mind attribution (Waytz, Heafner & Epley, 2014; Eyssel et al., 2012;
Eyssel & Kuchenbrandt, 2011). However, we did not find the expected effects on either
perceived animacy or perceived human likeness. The expectation that the level of control over
self-propelled motion creates perceptions of animacy (Poulin-Dubois, Lepage & Ferland,
1996). Even though this hypothesis has been shown to be valid in research on the perception
of animacy of geometric shapes (Scholl & Tremoulet, 2000), it did not hold up for the
self-propelled motion of a robot. One possible explanation could be that observing the movements
of abstract forms across a solid background in a video leaves more room for imagination and
interpretation about possible intentions of the object than observing the movements of a
concrete object. For example, a study by McAleer et al. (2004) found that people watching a
video where the visual ques of two people dancing were reduced to only white body
silhouettes across a black background felt more aroused compared to watching the actual
video of the two people dancing, as the former video was interpreted as two people fighting
rather than dancing.
The expectation that having low control over a robot would lead to higher perceived
human likeness of the robot than having high control was based on the theory of effectance
motivation (Epley, Waytz and Cacioppo, 2007), which states that people are more likely to
anthropomorphise when faced with uncertainty. However, this study found no effect of the
level of control over the robot on perceived human likeness. An important explanation as to why we didn’t find this effect could be that weaker forms of anthropomorphism only include immediate behavioural reactions towards the non-human agent, treating it as if it were human.
This weak form of anthropomorphism does not include the actual endorsement of the belief
that a non-human agent possesses human qualities (Epley, Waytz, Akalis & Cacioppo, 2008).
The interaction with the robot might have only induced this weak form of anthropomorphism,
which would have led people to behave towards the robot as if it were human, but would have
not led people to actually believe that the robot possesses human qualities. Since this weaker form of anthropomorphism consists of behaviour, it can’t be detected through a questionnaire as was used in this study.
This paper furthermore has implications for research on robot acceptance.One
important finding of this study is that perceived task performance and perceived ease of use
failed to predict the attitude towards use, as is predicted by the Technology Acceptance Model
(Davis, 1989). The TAM has been tested and validated in various fields, but is usually used to
predict the acceptance of pieces of technology functioning as tools to enhance task
performance (Lee, Kozar & Larsen, 2003). Social robots however are primarily hedonic rather
than utilitarian products. They serve hedonic purposes such as enjoyment and company, rather
than utilitarian purposes such as a robotic vacuum cleaner (Lee, Shin & Sundar, 2011). The
fact that perceived task performance and perceived ease of use did not predict the attitude
towards using the robot indicates that people indeed did not view robot as a tool with a
specific utilitarian function.The findings therefore show support for the claim that hedonic
responses play an important role when evaluating acceptance of social robots (de Graaf &
Allouch, 2011).
However, it must be noted that perceived ease of use did predict the extent to which
the robot was perceived as a potential friend. An explanation as to why perceived ease of use
did influence the acceptance of the robot as a friend but not the attitude towards use as
predicted by the TAM could be provided by the self-determination theory (Ryan, Rigby &
Przybylski, 2006). The self-determination theory describes what factors predict intrinsic
motivation (e.g. motivation that is derived from the satisfaction of performing an action),
which is the main motivation underlying play. According to the self-determination theory,
feelings of competence can increase intrinsic motivation. In a series of experiments, Ryan,
Rigby and Przybylski (2006) found that intuitive controls in video games increased feelings of
competence, which in turn increased enjoyment and, most importantly, increased preference
experienced increased feelings of competence, which in turn may have led to increased
enjoyment and preference for continued play with the robot.
Another important finding relating to robot acceptance is that only perceived human
likeness significantly influenced the attitude towards using the robot and the extent to which
the robot was perceived as a potential friend, whereas perceived animacy and mind attribution
did not. One possible explanation for this finding may be that people made their judgements
about the human likeness of the robot solely based on perceptual ques such as its vocal
capabilities and its appearance, which somewhat resembled a human (see figure 1). After all,
perceived human likeness of the robot was influenced by the robot’s vocal expressions but not
by the level of control over the robot. Most designers choose to design the appearance of the
robot in a humanlike way because it increases familiarity and thereby has a positive influence
on acceptance (Fong, Nourbakhsh & Dautenhahn, 2003; Höflich, 2013). Therefore, only
perceptual cues about the robot might have influenced robot acceptance, but not the beliefs
about its animacy or its cognitive capabilities. The assumption that cognitive beliefs might not
have influenced robot acceptance may relate to the strength of those beliefs. As discussed
previously, weaker forms of anthropomorphism do not include the actual believe that a
nonhuman agent has humanlike qualities, such as free will and emotions (Epley, Waytz,
Akalis & Cacioppo, 2008). Bartneck, Kanda, Mubin and Mahmud (2009) also pose that the
perception of animacy moves gradient. They imply that, rather than being either alive or dead, there is a category of “sort of alive”. Perceptions of animacy may therefore also exist in weaker or stronger forms. In the end therefore, the humanlike appearance of the robot might
have increased familiarity and thereby acceptance (Höflich, 2013), but the beliefs about the robot’s cognitive capabilities and its animacy may have been too weak to influence the acceptance of the robot.