Keywords

1 Introduction

Anthropomorphic design refers to the endowments of human-like characteristics, motivations, intentions, or emotions in information systems [1]. Along with the wide application of anthropomorphic systems, researchers have examined the effects of anthropomorphic interface of information systems on user experiences in various contexts, including e-commerce [2, 3], decision support systems [4], and online education [5]. Most research finds that anthropomorphic designs have positive impacts on users’ perceptions. For example, it can increase users’ trust and use intention towards the systems [2, 6]. Meanwhile, researchers have also found that the effects of anthropomorphic interface are highly contingent upon the concrete design of various avatar features, such as gender [7], ethnicity [8], appearance [9], facial expressions [10], and representations [11]. However, little research so far has investigated whether or not the age of an avatar may influence users’ perceptions of an anthropomorphic system.

Based on the baby schema (Kindchenschema) theory [12], we propose that avatar age is also an important design element that can affect users’ perceptions. When the avatar is a child, it might induce different reactions as compared to the scenario when the avatar is an adult. Because people unconsciously interact with anthropomorphic systems in the way they interact with humans [13], and they usually evaluate children and adults with different criteria. Specifically, users may feel a stronger perception of cuteness when interacting with an embodied avatar with child-like appearances and behavioral characteristics [14]. Social psychology studies have revealed that cuteness perceptions can trigger care-taking behaviors and promote social engagement [15]. Marketing researchers have also found that cute product designs can not only help consumers form a warm impression of a brand but also build up brand attachments [16]. However, the factor of cuteness has received relatively little attention in HCI research.

In addition, the usage scenarios also matter in evaluating anthropomorphic designs. In this paper, we focus on the scenario when errors occur in human-computer interaction. Various system errors are almost inevitable in system use, which often lead to negative user experience [17]. However, there is little research on how to improve user experience through the design of error messages or prompts for correction [18]. Service marketing literature has suggested that appropriate interaction design can effectively alleviate users’ negative feelings when service failure occurs [19]. Therefore, we chose to explore the effects of avatar cuteness in this particular scenario of system error. In summary, this paper attempts to answer the following research question: can avatar cuteness in anthropomorphic systems decrease users’ negative perceptions when an error occurs?

2 Theoretical Foundation and Hypotheses Development

2.1 Anthropomorphic Design in Information System

Originated from the interaction design between humans and robots, studies of anthropomorphic designs date back to the 1970s [20]. An information system can be anthropomorphized through various methods [21], such as representing the system with an avatar, demonstrating the personality of the system by texts and voice, and establishing attachments between users and the system by expressing emotions [22].

Previous studies have shown that anthropomorphic interface design generates positive user reactions because they enable users to interact with the system in the same way they interact with human beings [23], which can stimulate more social interactions between a user and the system [24]. Previous research have suggested that anthropomorphic interface can increase users’ trust in the system [6]. Users find interacting with anthropomorphic systems more joyful [25] and they feel more immersed in virtual environments [26] and become more willing to use such systems [2]. Users’ learning abilities are also improved when interacting with systems with avatars [27].

Users’ perceptions of anthropomorphic systems are heavily influenced by the appearance and behavioral characteristics of the avatars used. For example, users prefer a system whose avatar shares similar demographic characteristics with their own, including gender [7] and ethnicity [8]. Besides, they would infer the system performance based on avatar appearance. For example, a system is perceived as more powerful when the avatar being dressed more professionally [9].

This being said, the age of an avatar, which is another important demographic characteristic, has received relatively less attention compared with other demographic factors. Among the few papers examining the effect of avatars’ age, Baylor and Plant (2005) found that middle-aged avatars were more likely to stimulate girls’ interests in mathematics and science than younger avatars. Rosenberg-Kima et al. (2008) found that interacting with young cool avatars can improve female learners’ self-efficacy than old cool avatars.

2.2 Cuteness

In this paper, we propose that a key difference between a “child” avatar and an “adult” avatar is the perception of baby schema and cuteness. Proposed by Lorenz (1943), the baby schema (Kindchenschema) theory summarizes a series of physical characteristics of infants, including large round eyes, round and protruding cheeks, protruding foreheads, round faces, big heads, and thick extremities [31]. People would perceive a higher level of cuteness when an object scores higher on baby schema [14].

Researchers have found that most people show strong love in cute infants [32] because the cuteness of infants is a primary elicitor of caregiving behaviors from adults [12]. Mammalian infants-particularly human infants-are incapable of taking care of themselves, cuteness evolved as an adaptive mechanism for attracting nurturing behaviors from adults, thereby increasing their survival chance [32, 33]. The ability to recognize and differentiate cute entities from non-cute ones reflects an innate human motive to nurture and care for the offspring. Given the evolutionary role of cuteness, people’s responses to cute entities are biologically hardwired [34]. People’s preferences for cute entities are not limited to living things. Non-living objects with cute designs can also trigger cuteness perception. For example, Mickey Mouse, with its supernormal large head and big round eyes, is adored by the audience [35]. Products with cute features, such as rounded appearance and soft materials [36], small size [37], and soft colors [38], are very popular among consumers.

People’s responses to cuteness are both affective and inferential. In terms of affect, cuteness could generate sensory pleasure [34, 39] as well as complex emotions, including tenderness [15] and empathy [40, 41]. These emotions would elicit prosocial behaviors [23]. In terms of inferential responses, cute entities are associated with physical weakness and vulnerability [42], warmth [30], honesty [43], as well as youth and vitality [44]. These inferences can motivate people’s nurturing and caring-giving behaviors [32], and foster them to pay more attention to cute entities [45].

2.3 Hypotheses Development

When using an anthropomorphic system, users will perceive a higher level of cuteness if the avatar looks and behaves like a young child rather than an adult [14]. As discussed above, we expect that this cuteness perception can lead to a prosocial mindset of care-taking and helping. When an error occurs, this mindset will make the user become more helpful and more willing to resolve the problem. When a user would like to solve the problem rather than criticize the system, he or she is likely to perceive the error as less severe.

Besides, a user may apply different schemes when evaluating avatars of different age. In general, people tend to believe that mistakes made by children are not as fatal as those made by adults due to their lack of capabilities. As a result, the same mistake made by a child is more forgivable and less severe than that by an adult. Therefore, we posit:

H1: Avatar cuteness decreases users’ perceived severity of a system error.

Prior research has shown that cuteness elicits warmth, kindness, and attractiveness perceptions [30, 32, 42], and triggers prosocial behaviors and social engagements [32]. Feelings like warmth and kindness are associated with perceived social closeness [46]. Berry (1991) found that people with warm faces have higher social closeness scores than those with faces that are judged as less warm. In the same vein, a cute avatar will make the system be perceived as socially closer to the users as it triggers higher warmth perception than a less cute avatar. Therefore, we posit:

H2: Avatar cuteness increases the users’ perceived social closeness with the system.

3 Methods

3.1 Experiment Stimuli

A single factor (cuteness: low vs. high) between-subject laboratory experiment was conducted to test these hypotheses. We developed a mock-up system that claimed to evaluate a subject’s word processing capability by asking the subject to count the frequency of certain keywords from several pre-stored textual snippets. Subjects were asked to interact with the system and complete the evaluation task as fast as they can. Based on their performance, the top 25% of the subjects could get additional cash rewards. The system was designed with an intentional error so that the subject had to redo one-round of counting.

Avatar cuteness was manipulated by varying the look of the avatar and the font of the instructions which represented the conversation. As shown in Table 1, subjects in the low cuteness group saw a male adult avatar and the on-screen texts were presented in a formal Chinese fontFootnote 1, while those in the high cuteness group saw a baby boy avatar and the on-screen texts were presented in a childish style Chinese fontFootnote 2. A pretest was first conducted to ensure that the two designs can effectively manipulate subjects’ perception of cuteness. Thirty-seven subjects were recruited for the pretest. They were asked to browse the user interface of the two groups one by one (with the presentation order counterbalanced) and then assess the two avatars’ age, cuteness, and physical attractiveness respectively. The results showed that the perceived age of the baby boy was significantly younger than the male adult avatar (M adult = 27.38 vs. M child = 8.51, p < 0.001), and the avatar of baby boy was perceived as significantly cuter than the adult avatar (M adult = 4.51 vs. M child = 5.70, p < 0.001). Meanwhile, there was no significant difference in perceived attractiveness between the two avatars (M adult = 4.59 vs. M child = 4.95, p > 0.237). These results showed that our manipulation of avatar cuteness is effective.

Table 1. Stimuli design
Full size table

3.2 Measures

Measures for perceived severity is adapted from Pronk et al. (2010). Subjects respond to two questions, “How serious do you think the system error is?” and “How serious do you think the consequences of this mistake are?” on 7-point scales. Measures for social closeness is adapted from Ward and Broniarczyk (2011). Subjects rated “How close do you think the relationship between you and the system is, comparing the system to a person?” on a 7-point scale. Subjects’ gender and age were measured as control variables.

3.3 Procedures

The experiments were conducted in a behavioral lab. All experiment instructions, stimuli, and questionnaires are stored and presented through a self-administrated online survey system. A total of 77 undergraduate students were recruited to participate in the experiment in exchange for a cash reward. The average age is 21.8 years, and 54.54% of them were female.

Upon arrival, the subjects were randomly assigned to one experimental condition and were asked to finish two word-counting tasks. In each task, the subject was asked to count the frequency of the five specific keywords. The task could only finish when all five answers were correct. In the second task, the system would intentionally display a “save failure”, so the subjects had to finish an additional count in order to complete the task. After completing the two tasks, subjects were required to complete a questionnaire containing the measures for all variables. Upon the completion of the questionnaire, subjects are debriefed, thanked, and dismissed.

4 Results

Table 2 summarizes the group means and the standard deviations of perceived error severity and social closeness. The Cronbach alpha for perceived error severity is 0.947.

Table 2. Descriptive Statistics of Dependent Variables
Full size table

We then performed a one-way ANOVA to test the effects of avatar cuteness on perceived error severity and social closeness. Participants in the high cuteness group perceived the error as significantly less severe (M low = 4.74 vs. M high = 3.91, p < 0.017), thus H1 is supported. However, there is no significant between-group difference in terms of perceived social closeness (M low = 3.11 vs. M high = 3.03, p > 0.758). Therefore, H2 is not supported.

5 Discussion

5.1 Summary of Findings

The results of our experiment suggest that avatar cuteness can significantly decrease users’ perception of error severity as expected. Nevertheless, it fails to lead to higher perceived social closeness. There could be two possible explanations. First, social closeness can be effectively activated as long as the interaction partner is anthropomorphic [49]. Second, the perception of social closeness can be shaped by interacting with the system for an extended period of time, which was not the case in our experiment.

5.2 Theoretical Contributions and Practical Implications

The present research makes three theoretical contributions. First, our research addresses the research gap of avatar age. Specifically, we found that people interacting with a child avatar tends to perceive the error as less severe when a service failure takes place. This finding extends our understanding of people’s responses to anthropomorphic design in human-computer interaction.

Secondly, we introduce the concept of cuteness into the field of human-computer interaction, and employ it to explain the effect of avatar age. In recent years, with the growth of a new generation of young consumers, the word “Moe” (a Japanese word for cuteness) has attracted much attention, especially in the East Asian countries [50]. However, most of early studies on cuteness have been conducted by scholars in the fields of social psychology, marketing, and journalism. This study not only introduces cuteness in the HCI field, but also examines its applicability in the design of anthropomorphic systems.

Third, we take system error as a research scenario and broaden the boundaries of HCI research. Error handling is a very typical application scenario of HCI; however, there is little research focusing on this particular scenario. These research took the perspectives of system builders such as programmers rather than ordinary users [19]. Some studies have explored the role of error message design in improving user experience. For example, Linderman and Fried (2004) proposed several design principles. Seckler et al. (2012) focused on the layout of error messages. Our research is aimed at end users. We examine the role played by the avatar cuteness in the specific scenario of system errors. By transplanting theories of service failure and recovery in service marketing, we enrich the scope of IS research.

The research also has important implications for practitioners. Our results show that system designers can consider adding cuteness design elements in interactions when software errors are likely to occur. It could alleviate users’ negative perceptions of the error and thus reducing dissatisfaction. The appearance and interaction style of the avatar both work effectively in cuteness design.

5.3 Limitation and Future Research

This research has some limitations. First, we manipulated cuteness by varying avatar age and fonts. Future research can examine other forms of cuteness manipulation such as language style and avatar voice. Second, we created one particular system error in the experiment. Subsequent research can use other experimental tasks and error scenarios to verify our findings. Third, we adopted a lab experiment in this study, and all participants were university students. Future studies can use field experiments or natural experiments so as to improve the external validity.