Introduction

The continuous improvement of artificial intelligence (AI) technology with the rapid development of technology makes services better and more intelligent. The coexistence of human and intelligent robots is an inevitable trend (Blut et al., 2021; Lu et al., 2020). Since the outbreak in 2020, various AI robots have aided humans in a variety of tasks, including delivering goods, performing simple medical procedures, and caring for the elderly. In some dangerous and embarrassing situations, AI service robots outperform human employees (Pitardi et al., 2021). Applications for anthropomorphic artificial intelligence robots have been expanded into education, business services (hotel services, catering services, banking services, shopping guides), medical care, and other service industries (Čaić et al., 2018). AI technology will play an increasing number of roles in today’s complex social environment. More and more anthropomorphic and humanoid features are being incorporated into the design of service robots in order to promote human-computer service interaction. This design progresses from an initial industrial appearance and behavior design to an internal emotional interaction. As a result, it is critical to discuss the anthropomorphic characteristics of service robots in both theory and practice.

Anthropomorphism of robots refers to the attribution of human traits and emotions possessed by robots (Epley et al., 2007; Seo, 2022). Although the existing research has explored the anthropomorphism of service robots (Seo, 2022; Yam et al., 2021), the relevant empirical results have not reached a consistent conclusion. For example, some studies believe that anthropomorphism of service robots has a positive impact on customer satisfaction (Yam et al., 2021), but others believe that it will bring some negative results (De Visser et al., 2016; Mende et al., 2019). In a word, when discussing anthropomorphic features of service robot, most previous studies focused solely on some human features possessed by robots, ignoring how matching between different features affects consumers’ cognition, understanding and judgment of service robots. Therefore, in order to better understand the influence process behind the anthropomorphism of service robots, this study divides the anthropomorphism of service robots into two dimensions according to the theory of mind perception (Gray et al., 2007), and discusses its impact on consumer psychology from the perspective of dual matching. In cognitive science and social psychology, competence and warmth are considered as the basic dimensions of social cognition, that is, the basic social judgments between people (Fiske et al., 2007). People who are considered competence and warmth will trigger consistent positive emotions (Fiske et al., 2007), usually more favored, and engage in more positive interactions with peers. Previous studies have attempted to explore human-robot interaction from these two dimensions and found that competence and warmth are the most important predictive factors for human preferences between different robot behaviors (Scheunemann et al., 2020). Therefore, this study divided anthropomorphism of service robots into anthropomorphic competence and anthropomorphic warmth. Specifically, anthropomorphic competence refers to the degree to which the service robots is similar to human in ability, such as self-control, memory, planning, thinking, etc. Anthropomorphic warmth refers to the degree to which service robots are similar to humans in emotional experience, such as emotional recognition and communication. Based on this, this study uses experimental research methods to analyze the impact of the overall anthropomorphic degree of service robots in terms of competence and warmth on consumers’ emotional attachment to robots, and then affects their recommendation intention for service robots. Then, using the questionnaire survey method, this study conducts polynomial regression and response surface analysis technology to explore how the matching of anthropomorphic competence and anthropomorphic warmth affects consumers’ recommendation intention for service robots, and further discusses the mediating role of emotional attachment to service robots and the moderating role of paradox mindset.

To sum up, this study intends to make some contributions in the following aspects. First, this study divides anthropomorphism of service robots into two core dimensions (anthropomorphic competence and anthropomorphic warmth), which not only discusses the impact of overall anthropomorphism of service robots on consumers’ recommendation intention, but also discusses the different impact of different matching of the two dimensions on consumers’ psychological process. The results show that the overall degree of anthropomorphism of service robots has a significant positive impact on consumers’ recommendation intention for robots, and the more consistent the anthropomorphic competence and warmth is, the stronger the positive impact of anthropomorphism of service robots will be. This conclusion not only supplements the research on anthropomorphic service robots, but also enriches the research on antecedents of consumers’ recommendation intention. Secondly, this study found that the mediating mechanism of anthropomorphism of service robots affecting consumers’ recommendation intention is emotional attachment to service robots. This study proves that consumers have the strongest attachment to the robots when the service robot is consistent in anthropomorphic competence and warmth, thus increasing consumers’ recommendation intention for robots. The relevant results have opened the internal black box of the service robot’s influence on consumers’ recommendation intention, and made certain theoretical contributions. Third, this study also found that paradox mindset is an important boundary condition from the perspective of consumer characteristics. Specifically, for those consumers with high paradox mindset, they are more sensitive to the matching of service robots in terms of anthropomorphic competence and warmth, so the matching of anthropomorphic competence and warmth can have a more positive impact on the recommendation intention through consumers’ attachment to robots.

Theory and hypotheses

Literature review for anthropomorphism of service robots

Through systematic and adaptable interfaces, artificial intelligence robots interact, communicate with customers, and deliver services. They are intelligent robots that are socially assisted (Van Doorn et al., 2017; Wirtz et al., 2018). Intelligent robots, in comparison to humans, play a more stable role that is free of fatigue, emotional exhaustion, and role conflict. They can be competent for tedious, repetitive, and dangerous tasks as long as the program is set appropriately. In comparison to self-service technologies (SSTs), intelligent robots’ unstructured interface and process design can guide customers as needed and even correct their mistakes. It is more consistent and dependable than SSTs. The functional adaptability of AI robots and SSTs is a significant distinction (Huang & Rust, 2018). Anthropomorphism refers to the incorporation of human behavior, emotion, and thinking characteristics into non-human objects/organisms (Blut et al., 2021; Seo, 2022), which can be manifested as anthropomorphic appearance and anthropomorphic behavior. Anthropomorphism of AI robots, according to the definition, is to inject a series of unique human characteristics into the robots, which is a method of design or control. In this study, anthropomorphism of AI robots refers to all design/control methods that endow intelligent machines with human characteristics/capabilities.

Although existing research has conducted a discussion on the AI robots in the service situation, no consistent conclusion has been drawn on the impact on the service robots. On the one hand, the anthropomorphic robots’ automated social presence will improve customer satisfaction and participation via the mediating role of social cognition (perception of agency and experience) (Yam et al., 2021). As a result, intelligent robots can be compared to humans in some cases. There is no doubt that socially assisted service robots exhibit anthropomorphic characteristics at the internal, external, and social levels. Their primary positive role is to encourage friendly interactions between humans and robots. In the hotel service, for example, in addition to replacing human service functions (such as welcome, service guidance, goods delivery, and other simple service functions), service robots can have human-like interactions with customers (interesting interactions, reducing service waiting anxiety, responding to customers’ expressions, and so on), thereby reducing stress and loneliness, improving interest in service experience and stability of service quality, and increasing customer satisfaction (Van Doorn et al., 2017; Yam et al., 2021).If the machine is personalized, it can improve the friendly relationship, cooperation, and willingness to participate in service contact.

On the other hand, anthropomorphism of service robots may have negative consequences in service situations. Automation has been shown to reduce interpersonal contact, raise human concerns about privacy disclosure, and diminish the sense of control over human autonomous behavior (Jörling et al., 2019). Customers are more critical of the food and restaurants that use artificial intelligence to provide services (Nozawa et al., 2022). Customers are more likely to attribute failure to highly anthropomorphic robots, especially in the case of service failure, and this raises customers’ doubts about enterprises and intelligent services (De Visser et al., 2016). Furthermore, humanoid robots interaction is more likely to compensate for consumption and increase sales opportunities than human servers, but this must come at the expense of customer anxiety (Mende et al., 2019).

Anthropomorphism of service robots and consumers’ attachment

In the marketing context, attachment is a psychological bond with cognitive, affective and conative characteristics that connects consumers with specific consumption objects (e.g. enterprises, products, brands) (Park et al., 2010). In this study, the attachment object is expanded to service robots, that is, the psychological bond generated by consumers to service robots with cognitive, affective and conative characteristics. According to previous studies, human beings have a higher degree of identification with things that are consistent with or similar to their own characteristics and behaviors (Aggarwal & McGill, 2012), which can meet the human desire for the unknown and social contact needs, and improve perceived fluency and cognitive ability (Kim & Moon, 2009). The level of anthropomorphism has been discussed as the key factor for consumers to adopt service robots (Wirtz et al., 2018). Previous study also found that a high level of anthropomorphism will lead to customer familiarity, participation and positive evaluation (Lu et al., 2020). Therefore, this study speculates that when consumers think robots are more like humans, they are more likely to have an attachment to service robots. Based on this, this study proposes the following hypothesis:

Hypothesis 1

Anthropomorphism of service robots has positive effect on consumers’ attachment to service robots.

The four different scenarios of anthropomorphism of service robots

This study identified four different matching scenarios, as shown in Table 1, based on the level of anthropomorphic competence and anthropomorphic warmth: high-high; low-low; high-low; low-high. The first two are congruent, while the latter two are incongruent. Specifically, this study will investigate whether consumer attachment to service robots is higher in congruence scenarios than in incongruence scenarios.

Table 1 The four different scenarios of (in)congruence in anthropomorphic competence-warmth of robots
Full size table

Cognitive dissonance theory believes that human cognition includes attitude, intention, behavior and other components, which emphasizes that human cognition always seeks a balanced and coordinated state (Festinger, 1962; Hinojosa et al., 2017). Individuals have a tendency to maintain self-cognitive consistency. When individuals perceive that the attitudes, intentions or behaviors conveyed by the information providers are not balanced or contradictory, they will bring negative emotions and emotional experiences to individuals, that is, cognitive dissonance (Festinger, 1962). Dissonance is an unpleasant motivational state that occurs when the individual’s behavior and attitude are inconsistent (Levy et al., 2018). At this time, the individual experience generates certain psychological pressure, and there is no active behavior (Joule & Beauvois, 1997). However, this perceived cognitive structure can only be restored to a balanced cognitive structure if it is recombined.

Specifically, this study believes that when consumers’ cognition of service robots in terms of anthropomorphic competence and anthropomorphic warmth is inconsistent, it may lead to cognitive dissonance and thus have a negative impact on consumers’ psychological processes and cognitive judgments. Consumers’ cognition of the service robot is consistent when the service robots have the same anthropomorphic competence and warmth. Consumers’ cognition of the service robot is inconsistent when the service robot have different anthropomorphic competence and warmth. According to previous research, cognitive congruence will help consumers maintain cognitive balance, whereas cognitive dissonance will result in a series of negative emotions (Harmon-Jones, 2000; McGrath, 2017) and stress reactions (Maertz Jr et al., 2009; Menasco & Hawkins, 1978), which will be detrimental to consumers’ positive evaluation of service robots, thus weakening their attachment to service robots. As a result, when service robots are consistent in competence and warmth, consumers have a higher evaluation of the service robots, resulting in a higher attachment to service robots; when service robots are inconsistent in competence and warmth, consumers have a low attachment to service robots. Based on this, this study proposes the following hypothesis:

H2: The higher the congruence a service robot’s anthropomorphic competence and warmth level are, the higher the consumer’s attachment to service robots.

The mediating effect of attachment to service robots

Loyalty is one of the company’s primary marketing objectives, and the company frequently manages repurchase intention directly as a loyalty indicator. Recommendation intention is commonly used as a factor contributing to repurchase intention (Kato, 2019) and as a typical attitude indicator (Kato, 2022). Therefore, it is very important to manage consumers’ recommendation intention. According to the preceding conclusion, the consistency of service robots in competence and warmth can increase consumers’ attachment to service robots. The study then hypothesizes that consumers who have a stronger attachment to service robots will be more willing to recommend them. Specifically, customer emotional attachment has a significant positive impact on customer loyalty and satisfaction (Aldlaigan & Buttle, 2005; Levy & Hino, 2016), and plays an important role in maintaining the relationship between consumers and enterprises. Therefore, this study also believes that when consumers have a high level of attachment to service robots, they will show more positive feelings towards service robots, thus generating higher recommendation intention. Based on this, this study proposes the following hypothesis:

H3: Consumer attachment to service robots mediates the relationship between anthropomorphic competence-warmth congruence of service robots and consumer recommendation intention for service robots.

The moderating effect of consumer paradox mindset

Paradox mindset refers to the degree to which a person accepts tension and is motivated by it (Miron-Spektor et al., 2018). Individuals who engage in contradictory thinking will not avoid tension or compromise between competing elements. On the contrary, they are at ease with conflict and strive to embrace and transcend their opposing elements in order to achieve a higher level of learning and discovery (Lewis & Smith, 2014). Because people who engage in paradoxical thinking have a strong desire to successfully overcome tension, research shows that they will have a sense of optimism about their abilities even in the face of uncertainty and negative feedback (that is, an overall positive view) (Shepperd et al., 2015). For example, when individuals embrace and work under tension, they tend to focus on their positive aspects (Lomranz & Benyamini, 2016).The inconsistent competence and warmth of service robots will cause cognitive dissonance, which will have a negative impact on consumer attachment to robots. As a result, this study takes consumer paradox mindset into account as a moderating variable.

Specifically, for those consumers with high paradox mindset, they are willing to experience and accept tension, so they have a relatively high tolerance for cognitive dissonance in order to avoid the negative impact of cognitive dissonance. That is, when consumers’ paradox mindset is relatively high, they can accept service robot inconsistency in terms of anthropomorphic competence and warmth, resulting in relatively high attachment to service robots. Even so, this study still believes that consumers will have higher attachment to the service robots when the service robot is consistent in anthropomorphic competence and warmth. On the contrary, consumers with a low paradox mindset are unwilling to experience and accept tension, so cognitive dissonance is more negative for them. In other words, when consumers’ paradox mindset is low, they will be negatively impacted by service robot inconsistency in terms of anthropomorphic competence and warmth, resulting in low attachment to service robots. Based on this, this study proposes the following hypothesis:

H4: The relationship between anthropomorphic competence-warmth congruence of robots and consumer attachment to the robots will be moderated by paradox mindset of consumer. Specifically, for consumer with a high paradox mindset, consumer attachment to the robots will be positively predicted by increasing congruence between anthropomorphic competence and warmth of robots.

Study 1

The purpose of Study 1 is to verify Hypothesis 1 through experimental methods, that is, to explore the causal relationship between the anthropomorphic degree of service robots and the degree of consumers’ attachment to service robots. Since attachment is the core mediation mechanism proposed in this study, this study will also further explore the mediation role of attachment between anthropomorphism of service robots and consumers’ recommendation intention.

Methods

Participants and procedures

This experiment is single-factor-two level (anthropomorphic degree of service robots: high vs. low) between-subject design. The subjects participate online experiment on the Credamo platform. The subjects are undergraduate from a university in western China. After the researcher made the questionnaires for the two groups, the students from the two classes of the marketing course were asked to fill in the questionnaires. They will be randomly divided into two groups through the system and fill in the corresponding questionnaires. Finally, 85 valid questionnaires were obtained, with 23 males and 62 females. The age range is 18 ~ 25 years old (M = 20.39, SD = 1.25). There are 43 people in group of high anthropomorphic degree of service robots and there are 42 people in the group of low anthropomorphic degree of service robots.

Measures

First, in order to initiate anthropomorphic degree of service robots, this research requires the two groups to complete different reading and transcribing tasks (Seo, 2022; Yam et al., 2021). In the high anthropomorphic degree group, the participants are asked to read the following sentence: “Suppose you plan to go to a place for two days. After arriving at the hotel, you found a humanoid service robot named Lily, which has a human like face, eyes, mouth and arms. You find Lily can help you find a reservation, complete the check-in, and deliver drinks or food to your room. Lily can even make facial expressions and gestures like a person, making you feel warm and relaxed.” In the low anthropomorphic degree group, the participants are asked to read the following sentence: “Suppose you plan to go to a place for two days. After arriving at the hotel, you found a humanoid service robot named Lily, which has a human like face, eyes, mouth and arms. You find that Lily is unable to help you find a reservation, complete the check-in procedures, and deliver drinks or food to your room. Lily can’t make facial expressions and gestures like a person to make you feel warm and relaxed.” The subjects were asked to type two sentences, that is, sentences in italics in the above situations, describing Lily’s anthropomorphism information respectively.

Then, the subjects filled in their evaluation of the anthropomorphic degree of this service robot, which is used for manipulation test. The specific item is: “Please rate the degree of anthropomorphism of the service robot (Lily) in the above scenarios?” (1 = Not humanized, 5 = very much humanized). Finally, the subjects answered their robot attachment, recommendation intention and demographic information. Specifically, this study used the four-item scale to measure consumers’ attachment to robots and an example item is “To what extent do you feel emotionally bonded to Lily?”. Consumers rated their recommendation intention for the robots using the two-item scale and an example item is “Would you recommend Lily to others (including your family and friends)?”.

Results

Prior to testing the hypotheses, this study conducted t tests to verify the utility of the anthropomorphic degree manipulation. Results indicated that participants in the high anthropomorphic degree group (M = 3.64, SD = 0.93) rated anthropomorphism higher than those in the low anthropomorphic degree group (M = 2.12, SD = 1.10), t (83) = 6.90, p < 0.01, indicating that the manipulation was successful.

Then, this study tested hypothesis 1. The results showed that participants in the high anthropomorphic degree group reported higher attachment to service robot Lily (M = 2.67, SD = 0.92) than participants in the low anthropomorphic degree group (M = 2.24, SD = 0.95), t (83) = 2.15, p < 0.05, supporting Hypothesis 1.

In addition, this study further examined the mediating role of attachment to service robot. This study conducted a bootstrap analysis with a process model (Preacher & Hayes, 2004). The results showed that the mediating effect of attachment to service robot is significant (indirect effect = 0.17, 95%LLCI = 0.035, 95%ULCI = 0.401). Therefore, the anthropomorphism of service robots can enhance their recommendation intention by increasing their attachment to robots.

Study 2

The purpose of study 2 is to explore how different anthropomorphic matching of service robots in terms of competence and warmth affect consumers’ recommendation intention. Specifically, using the questionnaire method, this study further explored the mediating role of attachment to service robots and the moderating role of consumers paradox mindset.

Methods

Participants and procedures

Through an online platform (Credamo), this study collected a questionnaire on consumers’ perceptions of service robots. These customers volunteered to take part in the study and received RMB 3 in exchange for their time. Eventually, 215 of whom returned complete questionnaires, constituting the final sample of study 2. Among the consumers, approximately 54.42% were male (n = 117), 77.21% had a college degree or better (n = 166), the average age was 24.42 years (SD = 8.48), and the number of robot types consumers previously seen was 4.61 years (SD = 10.86).

Measures

According to the translation/back translation procedure (Brislin, 1980), the measurement item of this study is Chinese and the consumer answers based on the five-point Likert format (1 = strongly disagree; 5 = strongly agree).

Anthropomorphism of robots. Anthropomorphism of robots was measured from two dimensions, namely, anthropomorphic competence and anthropomorphic warmth. These items were developed from previous studies (Seo, 2022; Yam et al., 2021). Specifically, the guiding words are “Please express your views on the following items according to some service robots you have observed recently”. The anthropomorphic competence of service robots was measured by four items: “Service robots can think”, “Service robots can plan their actions”, “Service robots can remember things”, and “Service robots can complete my commands” (α = 0.80). The anthropomorphic warmth of robots was measured by four items: “Service robots can communicate with me”, “Service robots can recognize my emotions”, “Service robots can make me happy”, and “Service robots can feel my pain” (α = 0.82).

Attachment to robots. This study used the four-item scale of brand attachment developed by Park et al. (2010) to measure consumers’ attachment to service robots, and changed the “brand” in these items to “service robot”. An example item is “To what extent do you feel that you are personally connected to service robots?” (α = 0.92).

Paradox mindset. Consumers rated their paradox mindset using the nine-item scale developed by Miron-Spektor et al. (2018). An example item is “I often experience myself as simultaneously embracing conflicting demands.” (α = 0.89).

Recommendation intention for robots. Consumers rated their recommendation intention for the service robots using the two-item scale developed by Altunel and Erkurt (2015). The measurement items are as follows: “Would you recommend service robots to others (including your family and friends)?” and “Would you say positive things about service robots to other people?” (α = 0.90).

Control variables. This study controlled for consumer’s gender, age, education and the number of types of robots they have previously seen.

Analytic strategy

First, polynomial regression and response surface methods were used to test Hypothesis 1 (Edwards & Parry, 1993). Specifically, attachment to robots was regressed on the control variables, as well as the five polynomial terms, that is, anthropomorphic competence of service robots (AC), anthropomorphic warmth of service robots (AW), anthropomorphic competence of service robots squared (AC2), anthropomorphic competence of service robots times anthropomorphic warmth of service robots (AC*AW), and anthropomorphic warmth of service robots squared (AW2). Estimated equation is as follows and control variables are omitted.

$$\begin{aligned} {\rm Attachment}\; {\rm to} \; {\rm robots} &= {\rm b}_{0}+{\rm b}_{1}{\rm AC}+{\rm b}_{2}{\rm AW}+{\rm b}_{3}{\rm AC}^{2}\\&\quad+{\rm b}_{4}{\rm AC}*{\rm AW}+{\rm b}_{5}{\rm AW}^{2}+{\rm e}\end{aligned}$$
(1)

In the above equation, the congruence test involved the slope (b1 + b2) along the congruence line (AC = AW), and the slope (b1-b2) and curvature (b3-b4 + b5) along the incongruence line (AC = -AW).

Second, in order to test the impact of robot competence-warmth consistency on recommendation intention through attachment to service robots (Hypothesis 2), this study uses the block variable method proposed by Edwards and Cable (2009). In order to get the joint action of five polynomials (AC, AW, AC2, AC * AW, AW2), these five polynomials are combined into a Block Variable. In order to evaluate the mediating effect, this study used R software and Monte Carlo repeated sampling 20,000 times to estimate the 95% confidence interval of the mediating effect.

Third, Chow test (Chow, 1960) was used to test the moderating effect of paradox mindset and directly promote the comparison of high and low group coefficients of paradox mindset. This test is similar to testing coefficient equality in a multi-sample structural equation model.

In summary, the analysis software mainly used in this study includes Stata and R. In particular, we performed the regression and Chow test in Stata 15 and evaluate the mediating effect in R.

Results

Confirmatory factor analyses

The data of this study is collected from the self-evaluation of consumers, and the common method bias needs to be considered. Therefore, this study followed the method of Harman’s (1976) approach for single factor statistical test. The results show that the first factor accounts for 36.98% (less 50%) of the total variance, indicating that common method bias would not affect the follow-up analysis (Podsakoff et al., 2003).

Then, this study further conducted confirmatory factor analyses to examine the distinctiveness of the five variables (Anthropomorphic competence of service robots, Anthropomorphic warmth of service robots, Consumer paradox mindset, Consumer attachment to service robots, Consumer recommendation intention). The results revealed that the three-factor model (χ2 = 591.09, df = 220, SRMR = 0.07, CFI = 0.87, TLI = 0.85) was superior to all plausible alternative models (see Table 2).

Table 2 Model fit results for confirmatory factor analyses
Full size table

Correlation analyses

Table 3 presents the means, standard deviations, and correlations of all study variables. Anthropomorphic competence of service robots (r = 0.37, p < 0.01) and anthropomorphic warmth of service robots (r = 0.50, p < 0.01) were significantly related to consumer attachment to service robots. And consumer attachment to service robots was also significantly related to consumer recommendation intention for the service robots (r = 0.54, p < 0.01).

Table 3 Descriptive statistics and correlations
Full size table

Hypotheses testing

Hypothesis 2

proposed that the higher the congruence anthropomorphic competence and warmth levels of service robots are, the higher the consumer attachment to service robots. As shown in Model 2 of Table 4, the surface along the incongruence line was significantly negatively curved downward (Curvature = -0.72, p < 0.01), indicating that the congruence condition has higher attachment to service robots than the incongruence condition. In order to interpret these results holistically, this plotted the overall response surface using the coefficient estimates in Fig. 2. The concave curvature along the AC = -AW line illustrates that consumer attachment to service robots increases as anthropomorphic competence and warmth become more converge compared to dyads where anthropomorphic competence and warmth become more discrepant. Thus, Hypothesis 2 was verified.

Table 4 Polynomial regression results: The effects of anthropomorphic competence-warmth congruence on consumer attachment to service robots
Full size table
Fig. 1
figure 1

Conceptual model

Full size image
Fig. 2
figure 2

The moderating effect of paradox mindset

Full size image
Fig. 3
figure 3

The effects of anthropomorphic competence-warmth congruence on consumer attachment to robots

Full size image

In addition, as shown in Table 4, the slope along the congruence line is significant and positive (Slope = 0.84, p < 0.01), indicating that the high-high congruence condition has higher attachment to service robots than the low-low congruence condition. The response surface in Fig. 2 also indicates that consumer attachment to service robots is higher at the rear corner (high/high congruence) than at the front corner (low/low congruence). However, as for the asymmetrical incongruence effect, the quantity representing the lateral shift is negative but not significant (Slope = -0.16, ns), indicating a shift toward the region where AW is not greater than AC. This asymmetrical effect is also shown in Fig. 2, in which consumer attachment to service robots is not higher at the left corner (AC = 2 and AW= -2) than at the right corner (AC = -2 and AW = 2). This result further explains the research hypothesis that the anthropomorphic competence-warmth incongruence of service robots will lead to cognitive dissonance of consumers, thus reducing their attachment to service robots.

In order to test the mediating effect proposed by hypothesis 3, this study uses the block variable method. After obtaining the block variables, this study conducted a mediating effect test, and the results are shown in Table 5. The block variable has a significant positive impact on consumer attachment to service robots (β = 0.56, p < 0.001), while the consumer attachment to service robots also has a significant positive impact on consumer recommendation intention for the robots (β = 0.52, p < 0.001). The mediating effect of block variables on consumer recommendation intention for the robots through consumer attachment to robots was 0.29, and the 95% confidence interval did not include 0 (95% CI = [0.176,0.421]). Therefore, Hypothesis 3 was verified.

Table 5 The mediating effect of consumer attachment to service robots
Full size table

Turning to Hypothesis 4, which proposed that consumer paradox mindset moderated the relationship between anthropomorphic competence-warmth congruence of service robots and consumer attachment to service robots. The results are shown in Table 6 and Fig. 3. Specifically, this study divided the samples according to the median of the consumer paradox mindset, analyzed the samples above and below the median by structural equation, and compared the difference of coefficients (Lee & Antonakis, 2014). As shown in Table 5, in the high paradox mindset group, the surface along the incongruence line bends down significantly (curvature = -1.68, p < 0.01). However, in the low consumer paradox mindset group, this trend becomes less obvious (curvature = -0.53, p < 0.1).

Table 6 Polynomial regression results: The moderating effects of paradox mindset
Full size table

Next, this study further examined whether the five polynomial terms were different among groups. Using the SUEST command of Stata, this study simultaneously tested the difference of regression coefficients b1, b2, b3, b4 and b5 in the two groups (for AC, AW, AC2, AC * AW, AW2 in Eq. 1). The results show that the difference of the above coefficients were significant (χ2(5) = 13.49, p < 0.05). Therefore, Hypothesis 4 was verified.

Discussion

Service robots are becoming more popular as artificial intelligence technology advances (Blut et al., 2021; Lu et al., 2020). However, no consensus has been reached on the impact of service robots’ anthropomorphism on consumers’ psychological processes and behavioral responses. Based on the theory of mind perception (Gray et al., 2007), this study divides the anthropomorphism of service robot into two dimensions (competence and warmth) and discusses its impact on consumer psychology from the perspective of dual matching. It found that anthropomorphism of service robots has positive effect on consumers’ attachment to robots. Consumer attachment to service robots increased when anthropomorphic competence and warmth were congruent. In addition, consumer attachment to service robots mediated the relationship between anthropomorphic competence and warmth (in)congruence of service robots and consumer recommendation intention, and consumer paradox mindset moderated the effect of anthropomorphic competence and warmth (in)congruence of service robots on consumer attachment to service robots.

Theoretical implications

First, this study divides anthropomorphism of service robots into two core dimensions, namely, anthropomorphic competence and warmth, which enriches the literature on the interaction between service robots and consumers. Although the previous studies have explored the anthropomorphism of service robots (Seo, 2022; Yam et al., 2021), the relevant empirical results have not reached a consistent conclusion. For example, some studies believe that anthropomorphism of service robot has a positive impact on customer satisfaction (Yam et al., 2021), but others believe that it will bring some negative results (De Visser et al., 2016; Mende et al., 2019). This study believes that the reason for this confusion is that there is no effective distinction between anthropomorphic features. This study uses polynomial regression and response surface analysis to discover that the influence of anthropomorphism of service robot on consumer attitudes is dependent on anthropomorphic competence and warmth matching. The anthropomorphic service robot can have a more positive impact on consumers when the anthropomorphic competence and warmth are consistent. These findings have important theoretical implications for the literature on anthropomorphism of service robot.

Second, this study further enriches the mediating mechanism of anthropomorphic service robots influencing consumers’ recommendation intention, and opens the black box. Although previous studies have also explored the internal mechanisms of anthropomorphic service robots affecting consumer attitudes, such as pleasure (Seo, 2022) and perceived agency or experience (Yam et al., 2021), there is a lack of relevant research, which requires more in-depth discussion. Based on the perspective of attachment theory (Park et al., 2010), this study believes that the internal mechanism of anthropomorphic service robots impact on consumers’ recommendation intention is consumers’ attachment to service robots. The research conclusion also shows that both the overall anthropomorphism and the matching of competence and warmth can enhance the recommendation intention through consumers’ attachment to robots. Therefore, this study has made an important theoretical contribution by proposing a new mediating mechanism to link (attachment to service robots) the research of anthropomorphic service robot with the research of consumer recommendation intention.

Third, this study expands the boundary conditions of anthropomorphic service robots by proposing the paradox mindset as a moderator. The personality tendency of consumers has an important influence on their attitudes and behavioral responses (Madrigal & Boush, 2008; Marquis & Filiatrault, 2002). For example, previous study proved that neuroticism, extraversion, openness to experience exerted a more indirect and positive influence on compulsive buying, while conscientiousness and agreeableness showed a stronger direct and negative relationship with hedonistic shopping experiences and compulsive buying (Tarka et al., 2022). Therefore, from the perspective of paradoxical psychology, this study explores whether individuals with different paradoxical psychology have different anthropomorphic cognition of service robots. The study found that consumer will be positively impacted by service robot consistency in terms of anthropomorphic competence and warmth when consumers’ paradox mindset is high, resulting in high attachment to service robots. This study contributes theoretically and provides important boundary conditions for service robot anthropomorphism research.

Practical implications for marketing

First, the results of this study show that the matching of service robots in anthropomorphic competence and warmth has an important impact on consumer attitudes. In order to improve the positive attitude of consumers, the anthropomorphic competence and warmth of service robots should be kept at a high level as far as possible. In addition, even if the two dimensions cannot be maintained at a high level, the level of both can be lowered at the same time to avoid consumers’ negative reactions due to cognitive dissonance. Second, the consistency of service robots in terms of anthropomorphic competence and warmth can enhance consumers’ attachment to robots and thus enhance their recommendation intention. Therefore, it is also very important to enhance consumers’ attachment. For example, it can improve the communication and exchange between consumers and service robots, so that they can have emotional connection with each other. Third, this study also found that the paradox mindset of consumers is an important boundary condition for the anthropomorphic service robot to affect consumer attitudes. Therefore, enterprises need to take different marketing strategies and service attitudes according to the personality characteristics of consumers to maximize the consumer positive experience.

Limitations and future research

This study also has some limitations. First, this study only discusses the mediating role of consumer attachment to robots. Future research can further explore the internal mechanism between anthropomorphism of service robot and consumer attitudes from other theoretical perspectives. Second, this study only discusses the moderating role of paradox mindset, and future research can also explore the differential impact of other consumer personality characteristics on consumer attitudes. Third, this study only explores the anthropomorphic competence and warmth of service robots, but service robots also exhibit anthropomorphism in other aspects. Future research can delve deeper into this area. For example, Zhang et al. (2022) mentioned many anthropomorphic dimensions, including mission completion (core), user sensory experience (external manifestation), artificial intelligence (guarantee), and unique human characteristics (promotion). Fourth, both research samples in this study are from China and both tend to be younger. Future research can further explore other countries and select different sample groups as research subjects to increase the applicability of research conclusions. Fifth, this study only focuses on the impact of the matching of competence and warmth on consumer attitudes, without considering the independent impact of a single dimension. Given the inconsistent conclusions of previous studies (De Visser et al., 2016; Van Doorn et al., 2017; Yam et al., 2021), future research can consider the nonlinear effect (inverted U-shape) of anthropomorphism of service robots on the consumer attitudes to further enrich relevant literature.