Abstract
Keywords
Introduction
The term empathy tends to be used to refer to the ability to share the feelings and thoughts of other people and take another person’s perspective, in order to facilitate the process of social interactions (Azevedo et al., 2013; Håkansson & Montgomery, 2003; Kalisch, 1973). It is defined as “a combination of interrelated components of emotion recognition (in oneself and others), affective responsiveness (sharing the emotional experience of others), and perspective taking (cognitively assuming the perspective of others)” (Rodriguez, 2013, p. 494) or as “the capacity to (a) be affected by and share the emotional states of another, (b) assess the reasons for the other’s state, and (c) identify with the other, adopting his or her perspective” (Aaltola, 2014, pp. 243–244; De Waal, 2008, p. 281). We also differentiate between cognitive empathy, as the capacity to perceive and predict the feelings of others, and affective empathy, as the capacity to recognize emotional reactions when witnessing the suffering of another person (Patient & Skarlicki, 2010).
Empathy is associated with both positive and negative feelings. On the one hand, it can be triggered by feelings of happiness, achievement, excitement, and celebration (Keen, 2006). On the other hand, empathy is often experienced by an observer of another person in a distress situation or in pain. It can also arise while observing antisocial behaviors and aggression (Batson et al., 1987; Clark et al., 2019; Stanger et al., 2012; Sterzer et al., 2007). There is some terminological confusion with respect to the terms of empathy and sympathy. Sympathy is interpreted as the ability to pity and understand the suffering of others, that is to care for the suffering of others, where it can have a reaction tied to negative emotion (Eisenberg & Strayer, 1990). In contrast, empathy, as a broader concept, is the possibility of experiencing positive and negative situations or emotions as if they are one’s own (Davis, 2018; Keen, 2006). Although the term sympathy is older than empathy, both terms are associated with compassion, which is the understanding of the situation of the other.
It is important to realize that some people may be more predisposed to empathic reactions than others (Adams, 2019; Eisenberg et al., 1994). Dispositional empathy indicates the manner in which a person tends to respond toward the experiences of other people in general (Konrath et al., 2011). It is a multidimensional construct that includes both cognitive and affective facets (Davis et al., 1994, p. 370).
In general, empathy is understood as the ability to “put yourself in someone else’s shoes” (Faulkner, 2018, p. 218), however, “the others” is typically expected to be a fellow human being rather than a technological object (e.g., a robot) or a virtual character (Matravers, 2017, p. 86). An interesting issue that can be studied in the context of simulation games is to investigate what happens when we replace (virtual) humans by any (other) kind of character including virtual animals. Non-human characters in the form of virtual animals are mostly met with violence in computer games, where people can hunt animals as bears, deer or sharks. It has been shown that the use of violent video games can stimulate aggression, decrease pro-social behavior, increase mood changes, and impulsivity in young learners (Greitemeyer, 2018, 2019; Sherry, 2001; although for a summary of an opposing view, see the discussion in Kühn et al., 2019). Arguably, they can also lead to apathy towards the fate of similar characters in the real world.
Referencing empathic virtual agents, Ochs et al. (2008) proposed that empathy in human-machine interaction can be developed in two ways: (1) virtual agents (in this case, virtual characters) can manifest empathic emotions toward a player (also called “empathetic virtual characters” that foster immersion by McQuiggan et al. (2008, p. 1511)), and (2) players express empathic emotions toward a virtual agent (which Paiva et al. (2005) named “empathic synthetic characters”). In our study, we focus on the second type of relation. In line with Paiva et al. (2005, p. 265), we make use of “distress-inducing stimuli” (Belman, 2016, p. 64; Eisenberg et al., 1989, p. 42) to create conditions favorable to experiencing possible empathic reactions toward virtual characters. We vary the game character appearance (expressiveness and artificiality) to investigate its influence.
Past research showed that simulation games can be used as powerful tools to develop learning and social skills through virtual characters (Hofstede et al., 2010; Ke & Moon, 2018). Moreover, simulation games are typically considered well-suited to support educational programs fostering empathy, because they allow players to adopt new perspectives in an immersive way (Bachen et al., 2012; Belman & Flanagan, 2010). Research on human-computer interaction (HCI) (Pan & Hamilton, 2018; Scassellati et al., 2018), human robot interaction (HRI) (Ishiguro & Nishio, 2018), and video games (Wu et al., 2018) has led to a proliferation of studies using human characters. In contrast, recent research, such as that conducted by Schwind et al. (2018), has shown a lack of systematic studies on the effect of different design features for non-human artificial characters, such as animal-like characters or virtual robots, despite their growing popularity in simulated environments including learning applications.
A few studies analyzed empathy toward non-human characters like virtual pets. Virtual pets are defined by Tsai (2008) as the simulations of “life-like agents” and could possibly be used to build relationships with players (pp. 49). Tsai and Kaufman (2014, p. 149) used a “real-time pet simulation videogame” called Nintendogs and reported a positive effect of interacting with virtual pets on users’ relationship to pets in real life, suggesting that children playing the game developed an empathic disposition toward the virtual characters. This result supports the conclusion that empathic feelings toward a virtual animal are, in principle, possible. However, non-human characters often have human characteristics, such as movement, expressiveness, behavior, and personality (Tinwell et al., 2011; Yamane et al., 2010) and it is not clear what the effect of these properties is on the user. For example, does the “uncanny valley effect” (Ho & MacDorman [2017, p.129]; Mori, 1970) apply to the same extent to virtual animals? Which design features are necessary for the user to establish an emotional bond with a virtual animal?
According to the uncanny valley theory, an object with human-like traits can fail to elicit affinity when it is very realistic to its human counterpart (Mori, 1970, 2012). Aesthetic and other features of the inanimate objects can have a direct effect on their acceptation or rejection by contributing to the uncanny valley effect (Hodgins et al., 2010; Misselhorn, 2009; Riek et al., 2009). Here, a clear link to empathy can be found as it originally stems from the German word “Einfühlung,” i.e., the appreciation of the aesthetics of an object as an observer projection (Eisenberg & Strayer, 1990, pp. 18-20). One of the few investigations that have been carried out on virtual animals in relation to the uncanny valley effect concerned the appearance of the character, in particular the effects of realism, stylization, and facial expressions (Schwind et al., 2018). According to the outcomes of the study, the naturalness of the virtual animal was fundamental to create the sensation of realism with the user.
An important aspect of simulated environments involving virtual characters concerns immersion. Even though immersion is one of the most significant concepts in game design, little is known about the relationship between immersion and the empathic disposition toward a virtual character in a game environment. Arguably, empathy-centered design can improve the affective experience and emotional response of the player with respect to the virtual game character (Belman & Flanagan, 2010; Brown & Cairns, 2004). Next to that, several studies convincingly show that the affective experience increases immersion in the game (Jennett et al., 2008; Nacke & Lindley, 2008; Nacke et al., 2011). Given the link between emotional response and empathy on the one hand, and immersion on the other hand, we might expect empathy and immersion to be strongly related to each other.
The relationship between dispositional empathy and situational empathy plays a critical role in the measurement of empathy toward a virtual character. Eisenberg et al. (1994) describes that situational and dispositional measures can be correlated occasionally. It depends on the hypothesis of the study and their different methods of vicarious emotional responding (e.g., emotions and sensations experienced through the stories of others by watching or reading). The dispositional empathy can describe the current level of empathy of the participant, but cannot describe whether a stimulus affects the measure of empathy in the person. Likewise, the situational empathy can help to clarify if the stimulus (e.g., virtual character) can affect the empathy of the participant and possible relationships. For this reason, many studies used dispositional and situational empathy measures to identify relationships between both sub-constructs of empathy to have a better measure of empathy (Adams, 2019; Eisenberg et al., 1994; Holmgren et al., 1998; McQuiggan et al., 2008; Rosenthal-von der Pütten et al., 2013). To date, no previous studies have investigated whether the situational empathy toward a virtual character can be affected by the dispositional empathy of the users.
One of the most well-known approaches for assessing dispositional empathy (however, see Eisenberg & Strayer, 1990, for a discussion of potential weaknesses of this approach) is to use a self-reported questionnaire, such as Davis’s Interpersonal Reactivity Index (also known as the IRI questionnaire), which is the most frequently used (Davis, 1983; Garcia-Barrera et al., 2017; Hojat, 2016; Otterbacher et al., 2017; Rivers et al., 2016). Measurement of dispositional empathy may be affected by self-representation concerns, but the IRI questionnaire has an acceptable validity and reliability confirmed in studies with Dutch (De Corte et al., 2007), Chinese (Siu & Shek, 2005), and French population (Gilet et al., 2013). This questionnaire appears to be suitable for experiments concerning fictional characters, due to the inclusion of a fantasy measure scale.
Contrary to empathy as a trait, situational empathy (i.e., context-dependent empathic reactions) can be induced through specific stimuli or situations. Many researchers have utilized situational empathy to measure empathy-related responses (Eisenberg et al., 1994; Holmgren et al., 1998). To capture the concept of empathy in its entirety, self-reported measures based on existing questionnaires should be combined with measurements of psycho-physiological responses.
Situational empathy in the context of simulated environments appears to be very closely related to the concept of engrossment and presence (or, total immersion), both sub-concepts of immersion. Immersion is the illusion of experiencing a virtual environment as if it were akin to the real world (Hou et al., 2012; Mäyrä & Ermi, 2007) and can be enhanced by realistic elements within the virtual game experience. In the same way, the term immersion can be defined as a mental process (Mäyrä & Ermi, 2007) or intrinsic human characteristic (Hou et al., 2012), which can be affected by different elements of game design (e.g., size screen, viewing angle, audio, character’s aesthetics, or story) (Domsch, 2017; Hou et al., 2012; Mäyrä & Ermi, 2007).
One of the most well-known ways of assessing immersion within a computer interface is a questionnaire developed by Jennett et al. (2008). The questionnaire contains 31 questions on both game elements and experience. Brown and Cairns (2004) proposed that game qualities can be described in three distinct levels of immersion: engagement, engrossment, and total immersion. Whereas engagement can be measured using factors that include time, attention, and energy required from the player, engrossment, is associated with gamers’ emotions. Total immersion is associated with “presence”. At this stage, the impact of player’s feelings and thoughts are affected by their level of empathy and game’s atmosphere (e.g., graphics, plot, and sounds). Brown and Cairns (2004) offer anecdotal support for the idea that empathy and immersion have a strong relationship that is fostered by the game character and its interaction with the environment (character appearance and first-person player perspective).
To sum up, existing literature suggests that empathy and immersion are two strongly related concepts (engrossment and engagement) that are both affected by the game character’s appearance. Likewise, the game character’s appearance (artificiality/expressiveness) influences the level of self-reported situational empathy. Finally, the empathic tendency of the participants (dispositional empathy) and their possible empathetic reactions toward the virtual character (situational empathy) can influence to different levels of immersion. However, the empirical evidence supporting these conclusions is currently mainly based on anecdotal accounts. The expected relations are shown in Figure 1, based on reports found in the literature discussed above.
Expected relationships between the game character’s appearance, empathy, and immersion.
The purpose of the current study is to explore the effect of different design features pertaining to the virtual game character (artificiality and expressiveness) on situational empathy and immersion of the user in the simulated environment. In terms of artificiality, the body appearance of the virtual character in our experiment was manipulated using a distinction made by Coeckelbergh (2011, p. 199), who proposed that animals are “natural” and “biological” entities, a living organism in an ecosystem, while robots are “artificial” and “technological” articles and objects. Referring to a game character appearance, a natural appearance is understood to be closer to the biological animal in the real world in terms of important characteristics such as the animal’s color or body shape. An artificial appearance, on the other hand, has characteristics akin to the appearance of a robot. With respect to characters’ expressiveness, virtual characters can display human-like facial expressions (Beer et al., 2015; Paiva et al., 2005; Thomas & Johnston, 1997). According to Dyck et al. (2008), it is possible to enrich virtual faces of an avatar with expressions of happiness, sadness, fear, disgust, anger, and a neutral state. Dyck and colleagues found that recognition of the virtual expressions was comparable to natural facial displays of emotion. Interestingly, in their study, neutral expressions were most frequently confused with the emotion of sadness (both in human and virtual faces) but were also chosen when participants were uncertain about the emotion displayed. In contrast to Dyck et al., Hoffmann et. al., (2010), studying human expressions only, considered the neutral face to be a facial expression of its own right, but not an expression of emotion. Likewise, Tinwell et al. (2011) used basic universal emotions (happiness, anger, fear, sadness, and disgust) as described by Ekman (1992) and used the neutral expression as a kind of control state in virtual character (91.47%) and human images (89.92%), with surprisingly high recognition rates. In our study, we made use of three basic facial expressions: happiness, sadness, and neutral. These facial expressions were chosen due to their high and consistent recognition rates for virtual faces (Dyck et al., 2008).
Our hypothesis was that an expressive virtual character with animal-like features (i.e., low artificiality) would lead to a higher level of subjective immersion and situational empathy compared to a robotic character without emotional facial expressions. The main research questions addressed in the experimental design were as follows: (1) Does game character appearance (artificiality and expressiveness) influence the level of self-reported situational empathy? (2) Does game character appearance (features pertaining to artificiality and expressiveness) influence the level of immersion?, and (3) Does self-reported situational empathy correlate with immersion?
Methods
We used a 2x2 between-participant design and a control condition. The four experimental conditions of the game character appearance were: Natural (virtual animal) with expressiveness (emotional facial expressions), natural (virtual animal) with non-expressiveness (without emotional facial expressions), artificial (virtual robotic animal) with expressiveness (emotional facial expressions), and artificial (virtual robotic animal) with non-expressiveness (without emotional facial expressions). The control condition contained a baseline amorphous game character.
Participants were videotaped while accomplishing the experimental task. The task consisted of taking care of a virtual character by supplying it with energy, break time, and fun in a simulation game lasting 10.16 minutes (609.6 seconds). Before playing the game, the participants filled out the IRI questionnaire (Davis, 1983). After playing the game the participants filled out an “Immersion Questionnaire” (Jennett et al., 2008, p. 644), and a situational empathy question about the experiment.
Participants
We recruited 100 participants between 18 to 29 years old (M=22.47, SD=2.914). The participant sample was balanced for gender, with 50 females (50%) and 50 males (50%), drawn from a university student population in the Netherlands. The participants were randomly assigned to one of five experimental groups. Each condition had 20 participants (10 female and 10 male). The participants received a course credit for their participation and their participation was voluntary. They originated from several countries: Aruba (1), China (1), Colombia (3), Finland (1), France (1), Germany (1), Greece (2), Iceland (1), India (1), Iran (1), Ireland (1), Italy (3), Jamaica (1), Latvia (1), Morocco (3), Netherlands (70), Poland (1), Romania (2), Spain (1), Thailand (1), Turkey (1), United States (1), and Vietnam (1). The participants reported playing videogames: daily (10%), several times per week (19%), several times per month (19%), several times per year (31%), and never (18%).
Stimulus
The beaver is not considered a worldwide charismatic wild species, compared to other species such as the panda, polar bear, wolve, tiger, dolphin, whale, or ape (Albert et al., 2018; Ducarme et al., 2013). Since the beaver is not perceived as very attractive, cute, and charming by default, we expected that it would not directly stimulate a sense of empathy in the participants of the study. The beaver was depicted in a virtual setting with two different body design features; a natural body (virtual animal) and an artificial body (virtual robotic animal). Both designs are shown in Figure 2. We chose an amorphous figure as a control character, since it does not contain a body easily recognized by the user, such as any geometric figure used in other studies (Heider & Simmel, 1944).
Modeling in 3D design of the natural beaver (left side) and artificial, robot beaver (right side).
Artificiality and expressivity were combined into four experimental conditions and one control condition (as shown in Figure 3):
Five conditions of the game character.
Procedure
For the purposes of our experiment, we designed an experimental game called “Justin Beaver” (as shown in Figure 4). This game environment was inspired by the virtual pet game Tamagotchi, which also had an autonomous virtual character, but in a 2D environment (Higuchi & Troutt, 2004). Our game has a virtual simulation environment in 3D graphics, where the virtual character (animal or robot-animal beaver) explores a natural habitat. The virtual character moves randomly in the game environment. The player is instructed to take care of the character by supplying it with energy, break time, and fun during 10.16 minutes (609.6 seconds) through a “drag & drop” system. The players’ performance was indicated by increasing or decreasing level bars on the screen.
Screenshot of the game environment used in this study.
On the initial screen of the game, five virtual characters were displayed which allowed the researcher to choose with which character the participant would play. Then, to start playing the game, the participant had to press the “F9” key which saved an initial timestamp on the computer used later to synchronize different streams of data. The virtual characters in the experimental conditions (1) and (3) (the beaver or robot beaver) displayed facial expressions in the following order: 3 minutes of sadness, 3 minutes of neutral, 3 minutes of happiness, and 1 minute and 16 seconds of neutral (as shown in Figures 5 and 6). As the virtual characters in the other two experimental conditions (2) and (4) did not show emotional facial expressions, their appearance did not change during the game.
Timeline of emotional facial expressions of the natural beaver (virtual animal).
Timeline of facial expressions for the artificial beaver (virtual robotic animal).
After 10 minutes, a “distress-inducing stimuli” was evoked in the game when another character (a hunter) came out from behind a bush and shot the beaver. After the distress situation, there was a delay of 16 seconds and then a screen appeared with the message: “Game Over”.
Our control group (5) played with an amorphous figure as their virtual character. This virtual character had no emotional facial expressions, artificial or natural properties, or any particular kind of character. Its shape was indeterminate, and lacking a definite form (somewhat comparable to a marshmallow). The player took care of the character by supplying it with energy, break time, and fun during 10.16 minutes and they watched the same distress event as in the experimental situations and a message “Game Over”.
Ethics Approval
Ethics approval was obtained from the Research Ethics Committee of the Tilburg School of Humanities with the reference REC#2017/01. We found that this experimental game did no harm to the participants, other than feeling perhaps a little uncomfortable when the virtual character found itself in the distress situation. However, this situation was necessary for the successful induction of a distressful event in the experiment. After the experiment, the researcher explained to the participant that it was a virtual character and that no harm was actually done to any characters in the real world. Data and experimental game are publicly available on the Dataverse platform at Tilburg University (see https://dataverse.nl/dataset.xhtml?persistentId=hdl:10411/NJJATW).
Instrumentation
The participants took part in the task individually in front of a computer screen. Before of the game, IRI questionnaire was administrated online. After playing the game, the participants filled out an online Immersion Questionnaire and answered the Self-reported situational empathy about the character. Finally, the researcher conducted a semi-structured individual interview with the participant, taking between 3 to 7 minutes. In total, the experiment took approximately 1 hour.
Self-Reported Situational Empathy
We measured self-report situational empathy with a single item: “To what extent did you really empathize with the character (animal/robot-animal/amorphous figure)? The participants indicated the answer on a 5-point scale where the extremes were labeled (1=Not at all and 5=Very much).
Dispositional Empathy
To measure dispositional empathy, the participants filled out the IRI questionnaire (Davis, 1983, p. 1) which measures empathy on four sub-scales:
The principal component analysis (PCA) of the IRI questionnaire indicates a possible reduction from a larger number of variables of the IRI questionnaire to a smaller number of factors. The Keiser-Meyer-Olkin measure of sampling adequacy was 0.706 and the Bartlett test of Sphericity was highly significant, suggesting the data were suitable for PCA. The scree plot (as shown in Figure 7) of the PCA shows that four groups (factors) appear stacked and separated from the rest. Factor 1 was labeled “fantasy”, factor 2 “personal distress”, factor 3 “perspective taking”, and factor 4 “empathy concern” (see supplemental material #1). The outcomes of the PCA were in line with the sub-scales of the IRI questionnaire as they have been used in the past. In the data analysis below, we, therefore, treated dispositional empathy both as a single complex concept (validated by the Cronbach’s alpha value) and in terms of the four sub-scales (validated by the PCA).
Scree plot showing four groups (factors) of the IRI test.
Immersion
We measured immersion using a questionnaire that was originally developed by Jennett et al. (2008). Participants indicated their answers on a 5-point scale where the extremes were labeled (see supplemental material #2). The test had 31 items. The items (1-5, 7, 11-17, 19, and 21-31) was scored: 1 to 5. The reversed-scored items of this test were: 6, 8, 9, 10, 18, and 20. The Cronbach’s alpha for the Immersion questionnaire was reliable with alpha=0.906.
The principal component analysis (PCA) of the Immersion questionnaire indicates a possible reduction from a larger number of variables of the Immersion questionnaire to a smaller number of factors. The Keiser-Meyer-Olkin measure of sampling adequacy was 0.787 and the Bartlett test of Sphericity was highly significant, suggesting the data were suitable for PCA. The scree plot of the PCA (illustrated in Figure 8) shows that two groups (factors) appear stacked and separated from the rest. Factor 1 was labeled “engrossment”, and factor 2 “engagement” (Brown and Cairns (2004), pp. 1297–1298; see supplemental material #2).
Scree plot showing two groups (factors) of the Immersion test.
The outcomes of the PCA of the results of the Immersion Questionnaire was associated with two labels, which were used on the construct of immersion developed by Brown and Cairns (2004). Engagement was considered the first phase of immersion in which the players show their “effort, invest time, and attention” (p. 1298). Engrossment was considered the second phase of immersion in which the players show their “emotional investment in the game (p. 1299)”. Total immersion was considered the total score of the immersion which the players are totally involved in the game.
Semi-Structured Interview
During the debriefing the participants answered three open question in the form of a short semi-structured interview. This was used to collect additional information about experiment. The questions were: (1) Did you feel any emotional attachment to the character (animal/robot-animal/amorphous figure)? Why? (2) How did you feel when the character was in a distress situation?, and (3) What aspect could we improve in the video game to improve the emotional link to the character?
We analyzed whether the level of immersion of participants in the game correlated with their level of situational empathy answer. In addition, we compared the results of the IRI questionnaire with the Immersion Questionnaire and self-reported situational empathy. To identify whether the virtual character’s artificiality and expressiveness influenced the level of immersion and situational empathy of a participant, we compared the average results of the Immersion Questionnaire and self-reported situational empathy of the participants in the experimental conditions using two-way ANOVA. Moreover, we compared the results with the control group consisting of 20 participants. Data collected in the semi-structured interviews with open questions was used to gain qualitative insight and to acquire additional information about the experiment. Data preprocessing and analysis was performed using SPSS 24.0.
Results
Interaction Effect Between Artificiality and Expressiveness Toward Self-Report Situational Empathy
The first question of this study aimed to explore if the game character appearance (artificiality/expressiveness) influenced the self-reported situational empathy of the participants. A two-way ANOVA showed that there was no statistically significant interaction between the effect of artificiality and self-reported situational empathy F(1, 95) = .698,
Differences between mean values of self-reported situational empathy for artificiality/naturality and expressiveness/non-expressiveness (game character appearance). An interaction effect was found between game character appearance and self-reported situational empathy. Standard errors are represented in the figure by error bars attached to each column.
Interaction Effect Between Artificiality and Expressiveness Toward Immersion
The second question of this study explored whether game character appearance (artificiality/expressiveness) influenced the immersion level of the participants. A two-way ANOVA showed no statistically significant effect of artificiality: F(1, 95) = .116,
Mean difference values representing immersion for artificiality/naturality and expressiveness/non-expressiveness (game character appearance). Standard errors are represented in the figure by error bars attached to each column.
Relationship Between Dispositional Empathy, Self-Reported Situational Empathy and Immersion
The third question of this study was to explore whether dispositional empathy and self-reported situational empathy were correlated with immersion. A Pearson product-moment correlation coefficient was computed to assess the relationship between the dispositional empathy, self-reported situational empathy, and immersion. There was a positive correlation between the dispositional empathy and immersion level,
Bivariate Correlations Between the Measured Variables.
Discussion
The general aim of this study was to investigate if the appearance of the virtual animal character can be adapted to foster empathy and immersion. The game character appearance was systematically manipulated in terms of (1) artificiality (robotic or natural), and (2) expressiveness (with or without emotional facial expressions). Next to that, we measured dispositional empathy, situational empathy, and immersion in the game.
The first aim of this study was to explore whether game character appearance (manipulated in terms of its artificiality and expressiveness) can influence the level of self-reported situational empathy toward the character. Referring just to expressiveness, the present study set out with the aim of assessing whether emotional facial expressions of the virtual character can foster empathy in the user. In reviewing the literature, Ochs et al. (2008) described that a virtual character or agent can be called empathic when there are two situations: (1) users can feel empathy toward the virtual character/agent, or (2) the virtual character/agent shows empathic emotions concerning the users. For situation (1), previous studies have used facial expressions onto virtual characters to develop empathetic towards virtual agents or characters (Niewiadomski et al., 2008; Prendinger et al., 2005; Vinayagamoorthy et al., 2006). For instance, Ochs et al. (2008) used facial expressions in the virtual human characters to simulate the perception of emotions in agents. This means that a virtual character was perceived as being more empathetic when it had positive and negative emotions than when it was non-expressive. Can an emotionally expressive virtual character also foster the user’s empathy towards itself? The aim of the present study was to examine if a user would display more empathetic reactions towards virtual agents with facial expressions. We found that emotional facial expressions of the virtual character by itself cannot foster the empathy of users, as there was no statistically significant effect of expressiveness on self-reported situational empathy.
Another important aspect of the appearance of the game character is artificiality and naturality, with the theoretical construct of artificiality being based on the work of Coeckelbergh (2011), who defined a robot as an “artificial” and “technological objects” object (p. 199). In robotics, artificiality plays a very important role in realism and its effects with the emotional connection with the users. Andrews (2013) has suggested some concerns about artificiality of the robots when he described that modern robots look more like machines than live animals in natural environments such as a zoo. He noted that due to advances in technology in a virtual environment, it is easier to design realistic robot animals that simulate fine motor movement, appearance, and unpredictable behavior of a live animal. Previous studies investigating the uncanny valley have shown how artificiality (more or less human-like) of robots can affect the familiarity or affinity of the users (Mori, 1970, 2012). Riek et al. (2009) showed that appearance of a robot can affect the empathy of the users. They found that human-like robots foster more empathy in the users compared to mechanical-looking robots. However, they did not associate the mechanical looks of robots with artificiality but with the absence of anthropomorphic (human-like) traits. In reference to zoomorphic (animal-like) traits, there is a study that shows how a video where a dinosaur robot was tortured can affect self-reported empathy (Rosenthal-von der Pütten et al., 2013). This study showed that participants had a significant main effect on self-reported empathy related to “Pity for robot/angry at torturer”, but not “empathy with the robot” (pp. 24–25). This dinosaur robot had a more biological appearance than mechanical, which was called naturality in this study. For this study, the “natural” condition was designed through a virtual animal with a “natural” and “biological” body appearance (Coeckelbergh, 2011, p. 199). Contrary to expectations, this study did not find that the virtual character was perceived with more empathy when it was manipulated only in terms of its artificiality or naturality as it did not increase situational empathy in users. The most interesting finding of the current study was that the impact of artificiality/naturalness depended on the expressiveness of the character in that congruent appearances gave rise to the highest levels of situational empathy. We found that an artificial (virtual robotic animal) body appearance with absence of expressiveness and a natural (virtual animal) body appearance with expressiveness appeared to generate most empathetic reactions. We assume that these effects were due to participants preferring congruent virtual characters. In fact, similar choices can be found outside of the research domain. For example, in the movie WALL·E (film produced by Pixar Animation Studios), the animators designed a robot character called “Eve” with harmonic characteristics where the eyes of a robot were more mechanical (two blue moving lights) than human (e.g., an iris and a pupil).
The second aim of this study was to determine how the game character appearance (features pertaining to artificiality and expressiveness) influenced the level of immersion. The results for immersion were comparable to those for situational empathy: Immersion levels were higher for expressiveness when the game character’s appearance was natural (virtual animal). However, immersion was higher for non-expressiveness when game character’s appearance was artificial (virtual robotic animal).
The third aim of this study was to identify the relationship between empathy and immersion produced by the game character’s appearance. Our findings were consistent with the expected relationship between game character appearance, empathy, and immersion presented in the literature review. We found a moderately positive correlation (see Figure 11) between self-reported situational empathy and immersion, in line with the conclusions of Brown and Cairns (2004) who observed that players who were not completely immersed, experienced a lack of empathy with respect to some game design features (such as the appearance of the virtual character). Self-reported situational empathy was significantly correlated with both sub-concepts of immersion: engrossment and engagement. Interestingly, engrossment had a greater correlation with situational empathy (
Results of the expected relationship between game character appearance, empathy, and immersion.
Finally, we examined if dispositional empathy had any relation to situational empathy and immersion. The results of our study further support the idea that dispositional empathy is likely related to situational empathy towards a virtual character in simulated environments. This finding is in line with the proposal of Eisenberg et al. (1994). Participants with a high level of dispositional empathy are more likely to empathically respond to the virtual stimuli, though the response may differ depending on the stimulus or circumstances suitable for it (Adams, 2019). With respect to immersion, the results of our study showed a low significant correlation between dispositional empathy and immersion. This finding, while preliminary, suggests that there is a relationship between dispositional empathy, situational empathy and immersion, which is an important aspect for future research.
Limitations and Recommendations
In our research, we found that the appearance of a virtual animal character can be manipulated to foster empathic and immersion reactions in users of simulated environments. However, there are other aspects of these environments than need to be examined in future research, such as possible negative effects of empathy on players. An interesting example was discussed by Happ et al. (2013) in relation to a game character who was considered to be a victim of circumstances, generating empathy and justifying a high level of acceptance towards violence perpetrated by the character. One should thus be cautious when employing empathy in simulated environments as there is something inherently dangerous to people’s emotions being manipulated through game characters.
From a methodological point of view, our measure of situational empathy relied on a single-item instrument. A single item can give us a general idea about participants’ empathic feelings towards the character in an exploratory manner, however only one item is not sufficient to check the reliability and validity of the concept. For future studies, we intend to develop a multi-item measurement of situational empathy in simulated environments and to test its construct validity. Next to that, we aim to include socio-demographic variables such as the participants’ age, gender, and cross-cultural background, in the experimental design. Finally, in future research, we will examine more closely the links between self-reports of situational empathy and psychophysiological measurements (e.g., heart rate and facial expressions) to externally validate the construct.
Conclusion
The purpose this study was to determine how a virtual character’s visual appearance affects players’ experiences of empathy and immersion in virtual environments. In our experiment, the body appearance and facial expressions of the virtual character were manipulated in terms of (1) artificiality (robotic or natural), and (2) expressiveness (with or without facial expressions). The major finding of this study was the discovery that the interaction between the artificial/natural appearance and the expressiveness of a virtual character – particularly, their congruence - affects self-reported situational empathy of a player, as well as the level of immersion experienced in a simulation game. We found a positive correlation between dispositional empathy, self-reported situational empathy, and immersion, thereby empirically confirming the link between empathy and immersion previously proposed on conceptual grounds. The findings shed new light on studies of non-human characters (e.g., virtual animal/virtual robotic animals) and their effect on user experience. Further research might explore other types of situational empathy reactions. These may include but are not limited to measuring the emotions of participants using facial recognition software or psychophysiological measures during the same virtual interactive experience developed in this study.
Supplemental Material
Supplemental_Material_S_ang_G – Supplemental material for The Influence of Game Character Appearance on Empathy and Immersion
Supplemental material, Supplemental_Material_S_ang_G for The Influence of Game Character Appearance on Empathy and Immersion by Alexandra Sierra Rativa, Marie Postma and Menno Van Zaanen in Simulation & Gaming
Footnotes
Declaration of Conflicting Interests
Funding
ORCID iD
Supplemental Material
Author Biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
