Zum Verhältnis von Geistwahrnehmung und sozialer Unterstützung von Chatbots

TYPE Original Research PUBLISHED 06 March 2024 DOI 10.3389/fpsyg.2024.1282036 On the relationship between OPEN ACCESS mind perception and social support of chatbots EDITED BY Runxi Zeng, Chongqing University, China REVIEWED BY Hio Nam Io, Inju Lee and Sowon Hahn * Macau University of Science and Technology, Human Factors Psychology Lab, Department of Psychology, Seoul National University, Seoul, Macao SAR, China Republic of Korea Zijun Mao, Huazhong University of Science and Technology, China Weihong Dou, The social support provided by chatbots is typically designed to mimic the Lanzhou University, China way humans support others. However, individuals have more conflicting *CORRESPONDENCE attitudes toward chatbots providing emotional support (e.g., empathy and Sowon Hahn encouragement) compared to informational support (e.g., useful information swhahn@snu.ac.kr and advice). This difference may be related to whether individuals associate a RECEIVED 23 August 2023 certain type of support with the realm of the human mind and whether they ACCEPTED 14 February 2024 PUBLISHED 06 March 2024 attribute human-like minds to chatbots. In the present study, we investigated whether perceiving human-like minds in chatbots affects users’ acceptance CITATION Lee I and Hahn S (2024) On the relationship of various support provided by the chatbot. In the experiment, the chatbot between mind perception and social support posed questions about participants’ interpersonal stress events, prompting of chatbots. them to write down their stressful experiences. Depending on the experimental Front. Psychol. 15:1282036. doi: 10.3389/fpsyg.2024.1282036 condition, the chatbot provided two kinds of social support: informational COPYRIGHT support or emotional support. Our results showed that when participants © 2024 Lee and Hahn. This is an open-access explicitly perceived a human-like mind in the chatbot, they considered the article distributed under the terms of the support to be more helpful in resolving stressful events. The relationship Creative Commons Attribution License (CC BY). The use, distribution or reproduction between implicit mind perception and perceived message effectiveness in other forums is permitted, provided the differed depending on the type of support. More specifically, if participants did original author(s) and the copyright owner(s) not implicitly attribute a human-like mind to the chatbot, emotional support are credited and that the original publication in this journal is cited, in accordance with undermined the effectiveness of the message, whereas informational support accepted academic practice. No use, did not. The present findings suggest that users’ mind perception is essential distribution or reproduction is permitted for understanding the user experience of chatbot social support. Our findings which does not comply with these terms. imply that informational support can be trusted when building social support chatbots. In contrast, the effectiveness of emotional support depends on the users implicitly giving the chatbot a human-like mind. KEYWORDS chatbot, social support, mind perception, human-like mind, user experience, human-computer interaction (HCI) 1 Introduction We face many stressful events in our lives and tend to share them with others who provide various support including helpful advice or empathy. All valuable resources for dealing with stress events provided through social interactions and relationships are referred to as social support (Lin et al., 1979; Hobfoll, 1988; Cohen, 2004). Social support is divided into several subtypes based on resource characteristics, including informational support and emotional support (Cobb, 1976; House et al., 1985; Cohen, 2004). Informational support refers to the provision of useful advice or information that helps resolve stressful events, while emotional support refers to expressing empathy, encouragement, and care, making individuals feel loved Frontiers in Psychology 01 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 and valued by the support provider. Social support promotes Our research makes several contributions. First, it provides individuals’ psychological well-being by encouraging them to handle empirical evidence that users’ acceptance of chatbots’ support differs their stressful events in more successful ways (Cohen and Wills, 1985; depending on whether they explicitly perceive human-like minds in Cohen, 2004). the chatbots. The importance of implicit mind perception varies Chatbots, artificial intelligent (AI) agents that interact with users depending on specific types of social support. Specifically, the relation through written texts (Rapp et al., 2021), can be another source of between implicit mind perception and user attitude toward chatbots’ social support. Social chatbots, or chatbots that perform social support is stronger when the chatbots provide a type of support that functions, are designed to provide users with social support while requires a human-like mind (i.e., emotional support). Our findings having daily conversations and building social relationships with them imply that users’ implicit mind perception should be considered to (Shum et al., 2018; Ta et al., 2020). Mental health care chatbots are enhance the positive effects of support when designing chatbots for built to offer a variety of social support to help users deal specifically emotional support. In contrast, chatbots’ informational support is with their mental health problems, such as depression and anxiety more reliable in inducing positive user reactions. disorders (e.g., Fitzpatrick et al., 2017; Inkster et al., 2018). Previous studies have shown that social support from chatbots can improve psychological health (Fitzpatrick et al., 2017; Ly et al., 2017; Inkster 2 Background et al., 2018; Mehta et al., 2021; Meng and Dai, 2021). The social support provided by chatbots is usually designed to 2.1 Mind perception and its effect on mimic that provided by humans. When individuals share stressful human-chatbot interactions events, the chatbot not only provides useful information but also sends messages expressing empathy, as humans normally do (e.g., Whether a chatbot is considered a mindful entity depends on the Fitzpatrick et al., 2017; Inkster et al., 2018; Ta et al., 2020). However, observers’ perception. In other words, for the same chatbot, some human-like social support from chatbots does not always guarantee people may believe it has its own mind, while others may not (e.g., Lee positive user responses. Previous research showed users have et al., 2020). In addition, their perception can change depending on conflicting attitudes toward a particular type of social support— the context. For example, people are more likely to attribute emotional support. Some users acknowledge and appreciate chatbots’ humanness to AI agents when they have a more human-like empathic expression, whereas others feel discomfort when the chatbot appearance (Disalvo et al., 2002). expresses emotional support (e.g., Liu and Sundar, 2018; Urakami Individuals can attribute human-like minds to chatbots through et al., 2019; Bae Brandtzæg et al., 2021). Better understanding of users’ implicit and explicit processes. Implicit processes are automatic/ opposing attitudes is needed to optimize the effectiveness of spontaneous and made without awareness, whereas explicit processes chatbot support. are controlled and made with awareness (Nosek, 2007; Low and In the present study, we explored the conflicting user experience Perner, 2012). Chatbots are computers and do not fundamentally have of chatbot support, considering the characteristics of support, human-like properties. Nevertheless, sometimes people implicitly especially whether it is associated with the realm of the human mentalize them while explicitly recognizing that they lack human-like mind. When an AI agent appears to have the abilities that they minds. Reeves and Nass (1996) demonstrated that people exhibited ultimately lack, people may experience feelings of discomfort (Gray social responses to computers and treated them as if they were human and Wegner, 2012). People generally consider emotion as unique to beings, even when they explicitly acknowledged that computers are humans (Gray and Wegner, 2012) and perceive AI agents as lacking not humans. Additionally, previous studies have shown that explicit it (Gray et al., 2007; Jacobs et al., 2022). People may expect the and implicit mentalizing do not always go together (Banks, emotional support provider to have and feel emotions; therefore, 2020, 2021). they may reject emotional support from chatbots because chatbots Implicit mentalizing has been explored within the computers are do not have such capabilities. Individuals can only acknowledge the social actors (CASA) paradigm. This paradigm assumes that, when ability of chatbots to provide emotional support when they are computers or machines display enough social cues to users, users perceived to have a human-like mind. On the contrary, people may instinctively treat them as though they are humans (Nass et al., 1994; accept informational support from chatbots even if they do not Nass and Moon, 2000). Implicit attribution of mindfulness to humanize them. Any computer systems, including chatbots, are computers elicits similar responses from users as when they interact expected to be proficient in searching and providing useful with other people. For example, people tend to assess a particular information, as confirmed in previous studies (Brandtzaeg and computer’s performance more positively when they are asked to Følstad, 2017; Kim et al., 2018). answer an evaluation questionnaire on the same computer compared Although previous studies have investigated the importance of to independent sources (e.g., a paper questionnaire or a different mind perception in understanding chatbots’ social support (Liu and computer), which implicates that people perceive it as a direct Sundar, 2018; Urakami et al., 2019; Bae Brandtzæg et al., 2021), none evaluation of the computer, leading them to show politeness to the have directly measured users’ mind perception of a specific chatbot computer (Nass et al., 1994). People also apply gender stereotypes to and examined the relationship between this perception and various computers while automatically treating them like human beings (Nass types of social support. In the current study, we examined whether et al., 1994; Nass and Moon, 2000). Therefore, based on the CASA explicit or implicit mind perception is related to users’ attitudes paradigm, chatbots’ human-like social support is expected to have a toward chatbots’ social support. Additionally, we explored whether the positive effect on human-chatbot interactions similar to how support strength of the relation differs depending on the type of from humans does in human-human interactions. Previous studies chatbot support. have shown that users perceive the empathic expression of computers Frontiers in Psychology 02 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 positively when these agents display sufficient social cues (Bickmore Hypothesis 1-1: When users implicitly perceive human-like minds and Picard, 2004; Brave et al., 2005). in chatbots, they perceive the chatbots’ messages as more effective. Explicit mentalizing has been investigated through theories of mind perception. Theories of mind perception explain that people Hypothesis 1-2: When users explicitly perceive human-like minds perceive the mind of a particular entity in dimensions of agency in chatbots, they perceive the chatbots’ messages as more effective. and experience (Gray et al., 2007; Gray and Wegner, 2012). Agency is the ability to think and act volitionally, whereas experience is the ability to feel sensations and emotions. Researchers have explored 2.2 Importance of mind perception the general perception of various entities and found that adult depending on the type of social support humans are perceived as having both high agency and experience, whereas AI agents, including chatbots, are perceived as having a Theories of mind perception explain how the perception of low-middle level of agency but lacking experience (Gray et al., entities’ minds is related to the way people perceive their behaviors 2007; Jacobs et al., 2022). The higher the rating of an entity’s agency and interact with them. For example, when an agent is perceived as and experience, the more human-like people perceive it to be. The having agency, it is attributed with the responsibility for its behavior perception of AI agents can change through anthropomorphism, (Waytz et al., 2010). Additionally, when individuals dehumanize other which is the process of attributing human-like characteristics (e.g., people, or deny others’ human mind abilities, they tend to have more thought, intention, desire, and emotion) to nonhuman entities negative attitudes toward others (Hodson and Costello, 2007; Esses (Epley et al., 2007). The perceived agency and experience of the AI et al., 2008). agent can increase when people anthropomorphize the agent (Yam When AI agents behave as though they have mental abilities that et al., 2021). they essentially lack, people have a negative impression toward them. Explicitly attributing human-like minds to AI agents usually has In general, people assume that experience is fundamental to human a positive effect on the interactions between humans and the agents. beings (Gray and Wegner, 2012) and perceive AI agents as lacking it For example, Lee et al. (2020) found that mind perception in a chatbot (Gray et al., 2007; Jacobs et al., 2022). Consequently, people feel increased co-presence and closeness with the chatbot. Yam et al. uncanniness when AI agents appear to have experience (Gray and (2021) showed that when users anthropomorphized a service robot, Wegner, 2012). Relatedly, Stein and Ohler (2017) showed that people they perceived higher agency and experience in the robot, which perceived the empathetic and emotional expression of an AI agent enhanced their overall satisfaction with the hotel where the robot more negatively when they acted as though they had mental abilities, worked. In addition, when users perceive an intelligent personal and suggested that it was because people might not expect them to assistant to sense, think, and act autonomously (i.e., ascribe mind have those capabilities. attributes to the agent), they perceive the agent as more competent, The concept of mind perception may explain why conflicting user which leads to a higher intention to continuously use the agent (Hu attitudes frequently occurred for chatbots’ emotional support et al., 2021). compared to informational support. Different types of social support Based on previous research, we expected implicit and explicit offer different recourses (Cobb, 1976; House et al., 1985; Cohen, mind perception to be related to a more positive user attitude toward 2004); therefore, what mental abilities the support provider is expected chatbots’ social support. We examined perceived message effectiveness to possess can vary depending on the type of support provided. On (i.e., the extent to which users perceive chatbots’ messages as effective) the one hand, informational support abilities, such as understanding for user attitude in the context of chatbots providing social support, stress situations and exploring useful information to manage these, are as previous research did (Liu and Sundar, 2018). Furthermore, related to agency. On the other hand, emotional support abilities, such we measured mind perception as a variable rather than manipulating as understanding stress situations and the thoughts and feelings of it because users’ mind perception is influenced by a variety of factors, support-seekers, are related to agency, but also include sharing similar making it challenging to manipulate, even when incorporating feelings with support-seekers and feeling a desire to comfort them, anthropomorphic cues in chatbots (Epley et al., 2007; Waytz et al., which are related to experience. 2010). The corresponding framework is illustrated in Figure 1 and the The importance of attributing human-like minds therefore may corresponding hypotheses were as follows: vary depending on the type of support provided by chatbots. Chatbots FIGURE 1 Conceptual framework for hypotheses 1–1 and 1–2. Frontiers in Psychology 03 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 are perceived as having agency to some extent, but lacking experience 3 Materials and methods (Gray et al., 2007; Jacobs et al., 2022). Individuals might expect chatbots to provide informational support and acknowledge it, even 3.1 Participants though they do not ascribe human-like minds to chatbots. Previous studies have shown that users usually expect a chatbot to perform A total of 163 individuals participated in this study. We recruited informational analysis and retrieval (Brandtzaeg and Følstad, 2017; participants using online recruitment systems and communities and Kim et al., 2018). On the contrary, people consider chatbots’ emotional rewarded them with either course credits, gift vouchers, or payments. support as artificial and less genuine unless they humanize and Twenty-six individuals were excluded from the analysis owing to attribute experience abilities to the chatbot. In a previous study, users incomplete data. The final sample comprised 137 participants (56.2% denied an AI agent’s empathetic expressions while considering that it female). The mean age was 23.3 (SD = 6.1), ranging from 18 to 49 years. pretended to understand their emotions (Urakami et al., 2019). All the participants were Korean and 43.1% had experience using However, users who appreciated chatbots’ empathic expressions chatbots. We conducted retrospective power analyses (Faul et al., reported feeling as if they were talking to a human (Fitzpatrick et al., 2009) with our sample size at α =0.05. The current study had sufficient 2017; Bae Brandtzæg et al., 2021), which suggests the importance of power (>87%) to detect mid-to-large effects (Cohen’s d = 0.65 or mentalizing chatbots for accepting their empathy. partial η2 = 0.10). The power to detect medium effects (Cohen’s d = 0.50 Implicit attribution of a human-like mind may be sufficient to or partial η2 = 0.06) was as follows: 81% for t-tests and 65% for two-way evoke a positive experience from a chatbot’s emotional support, even analysis of variance (ANOVA) tests. when individuals do not explicitly mentalize them. As previously mentioned, the CASA paradigm, which explains positive user experiences of chatbots’ emotional support, assumes that users 3.2 Experiment design automatically attribute humanness to computers (Nass et al., 1994; Nass and Moon, 2000). We used a between-subjects experimental design, and the type Taken together, we hypothesized that the necessity of perceiving of social support provided by the chatbot was manipulated as human-like minds in chatbots for inducing positive user reactions follows: (1) base condition, (2) informational support condition, varies depending on the type of support provided by the chatbot. (3) emotional support condition. In the base condition, the chatbot Attributing human-like minds might be more crucial for emotional prompted participants to discuss their interpersonal stress events support to elicit desirable effects than informational support. As by asking them several questions. Specifically, participants were we did for Hypotheses 1–1 and 1–2, we measured perceived message asked what the stressful situation was about, what thoughts and effectiveness for user attitude toward chatbots’ social support. The emotions arose from the experience, how they behaved in the conceptual framework is illustrated in Figure 2 and the corresponding situation, how they would change their behavior, and how their hypotheses were as follows: emotions changed after conversing about the stressful event. It would function as a form of social support since previous research Hypothesis 2-1: The relationship between implicit mind perception showed that, when chatbots questioned what type of person the and perceived message effectiveness is stronger when the chatbots user was and what thoughts, feelings, and beliefs the user had, provide emotional support compared to informational support. these questions facilitated introspection and encouraged users to become aware and understand themselves better (e.g., Ta et al., Hypothesis 2-2: The relationship between explicit mind perception 2020; Bae Brandtzæg et al., 2021). In the informational or and perceived message effectiveness is stronger when the emotional support conditions, the chatbot additionally provided chatbots provide emotional support compared to informational the corresponding social support, as follows. For informational support. support, the chatbot provided relationship advice, according to the FIGURE 2 Conceptual framework for hypotheses 2–1 and 2–2. Frontiers in Psychology 04 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 phase of the interpersonal relationship: (1) building, (2) 3.4 Measurements maintaining, and (3) ending. We created the chatbot’s advice by referring to the Psychology of Human Relationships (Kwon, 2017). 3.4.1 Perceived message effectiveness For emotional support, the chatbot expressed phrases that To measure the extent to which participants perceived the chatbot conveyed understanding toward participants’ thoughts, emotions, messages as effective, we used four items from Holmstrom et al. (2005) and behaviors. It also expressed encouragement to participants. after translating them into Korean. The items were originally used for We adapted the emotional support content from Meng and Dai measuring “the perceived effectiveness of the helper’s behavior” in (2021). Participants were randomly assigned to one of three Holmstrom et al.’s (2005) study; they were also used to examine conditions (42 in the base condition, 48 in the informational chatbot message effectiveness (Liu and Sundar, 2018). Participants support condition, and 47 in the emotional support condition). were asked to rate each item on a 7-point Likert scale. Cronbach’s α was 0.86. The mean and standard deviation were 4.4 and 1.3, respectively. Example items included: “ineffective” – “effective” and 3.3 Experiment chatbot “helpful” – “unhelpful.” To control the flow of conversations between a chatbot and the 3.4.2 Explicit mind perception in a chatbot participants, we instructed the chatbot to lead the conversation To assess the extent to which participants explicitly perceived a according to predefined scripts. We built a script-based chatbot using human-like mind in the chatbot, we utilized items from Lee et al. Chatfuel1 and integrated it into Facebook Messenger. Following (2020) and Gray and Wegner (2012) after translating them into predefined scenarios, our chatbot sent various messages that facilitated Korean. We used two items from Lee et al. (2020) that assessed the social interactions with participants and dealt with interpersonal perception of the chatbot’s ability to think and behave, which stressful events. operationalized the capacity related to agency (Gray et al., 2007), and To encourage participants to anthropomorphize and attribute a two items from Gray and Wegner (2012) to assess the perception of human-like mind to our chatbot, we utilized several the chatbot’s ability to feel pain and fear, which operationalized the anthropomorphic and social cues suggested by previous studies capacity related to experience (Gray et al., 2007). Participants were (Gong, 2008; Araujo, 2018; Go and Sundar, 2019; Schuetzler et al., asked to answer each item on a 7-point Likert scale. Cronbach’s α was 2020; Adam et al., 2021; Schanke et al., 2021). Those cues facilitate 0.87. The mean and standard deviation were 2.9 and 1.3, respectively. anthropomorphism by increasing an agent’s social presence (i.e., the Example items included: “I felt that Allen was able to think by itself ” degree to which an agent is salient in the interaction; Short et al., and “I felt that Allen had the capacity to feel pain.” 1976) and signaling its identity. First, we tried to make our chatbot’s messages more human-like and personalized by sending responses 3.4.3 Implicit mind perception in a chatbot that reflected the contents of participants’ previous messages. We adapted one task from Banks (2020) to examine whether We integrated the Dialogflow2 AI system into our chatbot and participants implicitly perceived a human-like mind in the chatbot. The trained it to predict the types of relationships in which the Banks’s (2020) tasks measure users’ implicit mind perception toward a participants had stressful events (e.g., father, mother, friends, and robot. Since chatbots have different unique characteristics from robots romantic relationship) and the types of emotions experienced by (e.g., no voice and physical body), only one task was determined valid the participants (e.g., sad, depressed, and stressed). The chatbot sent to be utilized in the context of interactions between our chatbot and a messages that reflected the predicted types of relationships human: the white lie scenario. In that scenario, the AI agent does not or emotions. have to possess a voice or physical body but the ability to converse, We also used various other methods to attribute human-likeness which makes the scenario appropriate to be utilized for a chatbot. to the chatbot. For instance, by expressing the same meaning in In our measurement, the participants were presented with a various forms of messages (e.g., Okay, Ok, I see). To make an cartoon of the white lie scenario. In the scenario, our chatbot received impression that our chatbot typed messages in real-time, we added a gift from a human; however, it was not the gift that it wanted to have. some delay (1 s ~ 6 s depending on the length of messages) and showed Nevertheless, the chatbot answered that it liked the gift when the typing indicators (i.e., three dots) before sending messages. In human asked about it. After reading the cartoon, the participants were addition, we used a human-like image and name (Allen) for the asked to judge whether the chatbot lied and write down their chatbot. Lastly, our chatbot presented its identity using first-person explanation of why the chatbot said that it liked the gift. singular pronouns. Based on Banks’s (2020) codebook, two raters analyzed the In addition to imbuing human-likeness, to facilitate participants’ answers about the reasons for the chatbot’s behavior and anthropomorphism, we tried to elicit social responses and reduce coded the presence of implicit mind attribution. More specifically, the social distances by making the chatbot say “hi” and “goodbye” to raters examined whether the participants indicated an internal mental participants and engage in small talk (i.e., asking about participants’ state as the reason for that behavior. Internal mental state refers to experience of chatbot usage). “having any kind of thought, feeling, motivation, or condition that suggests the agent was actively thinking about how to respond, deciding how to react, driven to behave in a certain way, or emotionally moved to behave in that way” (Banks, 2020, p.2 in the codebook). If the answer contained an indication of the chatbot’s internal mental 1 chatfuel.com state, the answer was coded as 1, otherwise 0. For example, an answer 2 dialogflow.cloud.google.com such as “the chatbot might not want to hurt the gift giver’s feelings” Frontiers in Psychology 05 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 was coded as 1, while one like “the algorithm behind the chatbot made the overall implicit one. Then, we set the mean total score of 3 as the it respond in that way” was coded as 0. We calculated the Cohen’s cut-off point. Unless the participants explicitly denied statements kappa score to measure intercoder reliability and the score was 0.86, describing that the chatbot had a human-like mind, we interpreted that implying enough agreement. Eighty-three participants showed some degree of mind attribution occurred. In other words, if the indicators of mentalizing in their answers while 54 did not. participants rated at least one item 4 or higher (neutral, slightly agree, moderately agree, strongly agree), resulting in mean scores greater than 3, we considered them perceiving a human-like mind. 3.5 Procedure The study procedure was approved by our university’s institutional 4.2 Mind perception and perceived review board (IRB No. 2109/002–031). The experiment was conducted message effectiveness online. Participants were asked to participate in this study through their mobile devices because they had to converse with our chatbot We conducted t-tests to examine whether the degree of perceived through the Facebook Messenger application. After signing up for the message effectiveness differed depending on participants’ explicit and study, participants received an online study link. The first page implicit mind perception, respectively. The results revealed that the provided participants with a written description of the study. Only difference between the implicit mind perception and no implicit mind those who consented to participate were included in the study. perception groups was statistically significant (t (135) = 2.84, p = Participants were informed of the guidelines for chatting with our 0.005, Cohen’s d = 0.50). The implicit mind perception group (M = 4.6, chatbot and given another link directing them to a conversation with SD = 1.2) perceived the chatbot messages as more effective than the the chatbot on the Facebook Messenger application. The conversation no implicit mind perception group (M = 4.0, SD = 1.3). The difference scenario was as follows: (1) greeting, (2) engaging in small talk, (3) between the explicit mind perception and no explicit mind perception introducing the purpose of the conversation (i.e., discussing participants’ groups was also statistically different (t (134.89) = 6.78, p < 0.001, interpersonal stress event) and guaranteeing the confidentiality of the Cohen’s d = 1.09). The explicit mind perception group (M = 5.1, SD = conversation, (4) asking several questions about participants’ 0.9) also showed higher score in perceived message effectiveness than interpersonal stress events and (only in the informational or emotional the no explicit mind perception group (M = 3.9, SD = 1.3). support conditions) providing informational or emotional support, respectively, (5) asking whether participants would change their behavior and the extent to which they felt their negative emotions 4.3 Mind perception and different social changed, and (6) concluding the conversation. support At the end of the conversation, the chatbot sent a link to a survey. In the survey, participants’ perceived effectiveness of the chatbot’s We explored whether the relation between mind perception and messages and mind perception for the chatbot were measured. perceived message effectiveness varied depending on the type of social Implicit mind perception was assessed following the measurement of support; in other words, whether the association was stronger when explicit mind perception. This sequence was intended to mitigate the the chatbot provided emotional support. The means and standard potential influence of exposure to the scenarios of implicit mind deviations of perceived message effectiveness for each condition are perception measurement, which could have affected responses to the shown in Table 1. items assessing explicit mind perception. Finally, after collecting the participants’ demographic information, the study was concluded. 4.3.1 Examining a potential confounding factor: chatbot’s message content In the informational support condition, the chatbot sent different 4 Results advice depending on the type of stress event (i.e., relationship building, maintaining, and ending) to make conversations more natural and 4.1 Dividing data into implicit/explicit mind encourage participants’ engagement. To check the potential perception groups confounding effect of different message content, we conducted a one-way ANOVA test to examine whether different types of advice Before conducting hypothesis testing, we preprocessed the implicit had an effect on perceived message effectiveness. The results revealed and explicit mind perception data. For implicit mind perception, that perceived message effectiveness remained consistent across we classified the data into implicit mind perception group (n = 83) if the different types of advice (F (2, 45) = 0.00, p = 0.999, η2 = 0.00). The score on the measurement was 1, and no implicit mind perception group means and standard deviations of perceived message effectiveness for (n = 54) if the score was 0. For explicit mind perception, we also divided each type of advice were as follows: relationship building, M = 4.6, SD the data into two groups (explicit mind perception and no explicit mind = 1.3; relationship maintaining, M = 4.6, SD = 1.2; and relationship perception) using the mean total score of the scale. If the mean score of ending, M = 4.6, SD = 1.4. explicit mind perception was greater than 3, the data were assigned to the explicit mind perception group (n = 82); otherwise, they were 4.3.2 Informational support assigned to the no explicit mind perception group (n = 55). We conducted two-way ANOVA tests to examine whether the We used the total score of explicit mind perception, rather than effects of providing informational support on perceived message individual scores for agency and experience, to maintain equivalence effectiveness varied depending on the participants’ implicit or explicit between explicit and implicit mind perceptions since we only measured mind perception, respectively. The results revealed that the Frontiers in Psychology 06 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 TABLE 1 Means and standard deviations of perceived message effectiveness. significant (F (1, 85) = 0.53, p = 0.468, partial η2 = 0.01). However, the extent to which participants perceived the chatbot’s messages as M SD effective was lower in the emotional support condition compared to Base Condition the base condition (F (1, 85) = 5.01, p = 0.028, partial η2 = 0.06). In No Implicit Mind Perception 4.6 1.2 addition, the explicit mind perception group considered the chatbot’s messages as more effective than the no explicit mind perception group Implicit Mind Perception 4.5 1.2 (F (1, 85) = 11.84, p < 0.001, partial η2 = 0.12). No Explicit Mind Perception 4.1 1.3 Explicit Mind Perception 5.2 0.7 Informational Support Condition 5 Discussion No Implicit Mind Perception 4.2 1.0 This study explored whether the degree to which users perceived Implicit Mind Perception 4.8 1.3 the chatbots’ message as effective differed depending on the No Explicit Mind Perception 4.2 1.3 perception of chatbots’ minds and whether the strength of the Explicit Mind Perception 5.2 0.9 relationship between them varied depending on the type of support provided by the chatbots. The t-test results showed that implicit and Emotional Support Condition explicit mind perception in a chatbot was related to increases in No Implicit Mind Perception 3.2 1.3 perceived message effectiveness. Furthermore, the two-way ANOVA Implicit Mind Perception 4.5 1.1 results revealed that, for explicit mind perception, positive No Explicit Mind Perception 3.4 1.3 relationship was still significant after considering the effects of Explicit Mind Perception 4.9 0.9 support type. In contrast, the relationship between implicit mind perception and perceived message effectiveness was not significant after considering the effects of support type. These results suggest that interaction effect between informational support and implicit mind the relationship between implicit mind perception and perceived perception was not statistically significant (F (1, 86) = 1.26, p = 0.264, message effectiveness was influenced by the type of social support. partial η2 = 0.01). The degree of perceived message effectiveness was Thus, Hypothesis 1-2 was supported while Hypothesis 1-1 was only not significantly different between the base and the informational partially supported. support conditions (F (1, 86) = 0.76, p = 0.387, partial η2 = 0.01). According to the two-way ANOVA results, there was a significant Additionally, the difference between the implicit mind perception interaction effect between implicit mind perception and support type and the no implicit mind perception groups was not statistically on perceived message effectiveness only when the chatbot provided significant (F (1, 86) = 0.01, p = 0.926, partial η2 = 0.00). For explicit emotional support. However, there was no significant interaction mind perception, the interaction effect between informational effect between explicit mind perception and support type for both support and explicit mind perception was also not statistically informational and emotional support. In sum, only when participants significant (F (1, 86) = 0.19, p = 0.666, partial η2 = 0.00). The degree of did not implicitly attribute a human-like mind to the chatbot, perceived message effectiveness was not significantly different providing emotional support (but not informational support) between the base and the informational support conditions (F (1, decreased the extent to which the participants considered the chatbots’ 86) = 0.26, p = 0.615, partial η2 = 0.00). However, the explicit mind messages as effective. Thus, only Hypothesis 2-1 was supported. perception group perceived the chatbot’s messages as more effective Additionally, the two-way ANOVA results revealed that overall, the than the no explicit mind perception group (F (1, 86) = 11.69, degree of perceived message effectiveness did not vary between the p < 0.001, partial η2 = 0.12). base and informational support conditions, while perceived message effectiveness was lower in the emotional support condition compared 4.3.3 Emotional support to the base condition. We conducted two-way ANOVA tests to examine whether the This study was the first to directly measure user mind perception effects of providing emotional support on perceived message and investigate how it was associated with the extent to which users effectiveness varied depending on the participants’ implicit or explicit perceived the chatbot message as effective. Our study was particularly mind perception. The interaction effect between emotional support pioneering in separately investigating implicit and explicit mind and implicit mind perception was statistically significant (F (1, perception in the context of exploring chatbot support. As 85) = 6.31, p = 0.014, partial η2 = 0.07). The difference in perceived we measured implicit mind perception after the explicit one, exposure message effectiveness between the implicit mind perception and the to the explicit mind perception scale might have primed mind no implicit mind perception groups was more pronounced in the attribution when the participants answered on the implicit one. emotional support condition compared to the base condition, as However, nearly 40% of participants did not implicitly attribute a shown in Figure 3. Perceived message effectiveness was lower in the human-like mind even after the exposure, suggesting that the potential emotional support condition compared to the base condition (F (1, priming effect might not be significant. Therefore, the distinct 85) = 12.27, p < 0.001, partial η2 = 0.13). In contrast, there was no examination of implicit and explicit mind perception was possible. significant difference in perceived message effectiveness between the Our study contributes significantly to understanding how implicit mind perception group and the no implicit mind perception perceiving human-like minds in chatbots is associated with the user group (F (1, 85) = 0.01, p = 0.925, partial η2 = 0.00). For explicit mind experience for chatbots’ social support. First, our work demonstrated perception, the result showed that the interaction effect between the importance of explicit mind perception in inducing positive users’ emotional support and explicit mind perception was not statistically reactions to the chatbots’ social support, regardless of the type of Frontiers in Psychology 07 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 FIGURE 3 Interaction effect between providing emotional support and implicit mind perception. support provided by the chatbots. As suggested by Lee et al. (2020), condition of the experiment) has a similar or potentially higher positive when users explicitly perceive chatbots as mindful entities, interactions impact on the perceived effectiveness of the message compared to with chatbots might be considered as more meaningful, consequently providing additional informational or emotional support. Our findings fostering a more favorable reception of the chatbots’ social support. are consistent with the previous research, which suggests that writing Mindfully experiencing the chatbot’s humanness might encourage the about stressful experiences is beneficial since it can facilitate participants to perceive the conversation with the chatbot as introspection and self-evaluation (Ta et al., 2020; Bae Brandtzæg et al., more valuable. 2021). These results implicate that users might expect different forms Second, our findings revealed that the importance of implicit mind of support from chatbots rather than mimicking humans’ support. perception in positive user experience differed depending on the type Some researchers suggested that, because people nowadays know of support provided by chatbots. Specifically, implicit mind perception various AI agents, understand their unique characteristics, and have is more important when chatbots provide emotional support, as experience with them, they have different expectations from them and, emotional support implies that the support provider has the ability to further, interact with them differently to how they interact with understand and feel emotions. To reiterate, users appreciate chatbots’ humans (Gambino et al., 2020). For example, users expect and social support only when they perceive the chatbots as having the recognize AI agents to be non-judgmental because they are essentially capacity to provide such support at least implicitly. In general, people machines, encouraging users to tell their innermost stories without consider chatbots as having agency to some extent, but lacking fear, unlike when conversing with other humans (Ta et al., 2020). experience (Gray et al., 2007; Jacobs et al., 2022). Thus, even though Moreover, users recognize that AI agents lack emotions and expect users do not implicitly perceive human-like minds in chatbots, they them to listen and react to their stories without becoming tired (Kim appreciate the chatbots’ informational support because chatbots are et al., 2018). Another possible explanation for the results is the considered to have agency, which is required to provide informational insufficient effects of the informational and emotional support. The support. In contrast, since chatbots are regarded as lacking experience, informational support provided by our chatbot included advice that their emotional support causes discomfort if users do not humanize was general rather than customized to participants’ exact situations. In and attribute the experience to them. Gray and Wegner (2012) revealed addition, the chatbot mostly retrieved typical and general empathic that perceiving experience in a machine led to eeriness and unease expressions (e.g., “I would feel the same way as you.”) for emotional from users as machines are considered to lack experience. Relatedly, support. The generality and typicality of the provided support might users who did not appreciate an AI agent’s empathic expression felt that have reduced its effects. it pretended to understand their inner states (Urakami et al., 2019). On Lastly, our work has some practical implications. Specifically, the the contrary, users who appreciated chatbots’ empathic expressions results encourage practitioners to design chatbots’ social support reported feeling as if they were talking to a human (Fitzpatrick et al., differently depending on whether users implicitly perceive human-like 2017; Bae Brandtzæg et al., 2021). Thus, implicitly perceiving chatbots minds in the chatbots or not. Informational support is more dependable as having the capacity to deliver a certain type of support might in eliciting desired effects regardless of users’ implicit mind perception, be necessary to induce the desired reactions from users. whereas emotional support is not. If users implicitly perceive human-like Third, simply encouraging users to write down their stress events minds in chatbots, they may expect the same form of support that by asking several questions (what the chatbot performed in the base humans would provide. In other words, users may expect the chatbots to Frontiers in Psychology 08 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 understand and empathize with their experience and respond to them Data availability statement with appropriate emotional reactions. However, if users do not consider chatbots as mindful entities, even in implicit ways, they may not want the The datasets presented in this article are not readily available agents to pretend to understand their inner experience. Other forms of because the conversation data should be confidential, and some support that computers can perform (e.g., listening to users’ stories participants did not consent to their data being available to other without getting bored or offering useful information) might be more researchers. Requests to access the datasets should be directed to IL, appreciated by users. ijlee37@snu.ac.kr. 6 Limitations and future research Ethics statement The present study has some limitations. First, we did not The studies involving humans were approved by Seoul National manipulate the degree of participants’ mind perception but measured University institutional review board. The studies were conducted in it. Consequently, we could not investigate the casual effects of mind accordance with the local legislation and institutional requirements. perception on users’ attitudes toward the support received. We decided The institutional review board waived the requirement of written to measure users’ mind perception rather than manipulate it because informed consent for participation from the participants because the one’s tendency to anthropomorphize and attribute humanness is study was conducted online. Online consent was obtained instead of influenced by various factors (Epley et al., 2007; Waytz et al., 2010), written consent. except for chatbots’ human-likeness, and therefore, hard to be perfectly manipulated. Nevertheless, to investigate the causal effects of mind perception, future research should explore ways to manipulate users’ Author contributions mind perception as perfectly as possible. Second, we used only one task for assessing implicit mind IL: Writing – original draft, Conceptualization, Formal analysis, perception because it was the only task that was determined to Funding acquisition, Investigation, Methodology. SH: Supervision, be appropriate to use in the context of human-chatbot interactions Writing – review & editing. (i.e., the white lie scenario). Future research should develop validated measures to assess implicit mind perception in chatbots and conduct a more in-depth investigation. Funding Third, our chatbot sometimes predicted participants’ emotions and situations incorrectly. Prediction failure may impair the positive user The author(s) declare financial support was received perception of social support. To accurately examine user experience of for the research, authorship, and/or publication of this article. This social support, future studies should conduct experiments with Wizard- research was funded by the College of Social Sciences, Seoul National of-Oz methods (Dahlbäck et al., 1993) or more sophisticated prediction University under the Project for Fostering Social Science Researchers. models such as the large language models. Finally, the current study was slightly underpowered to detect medium effects in the ANOVA tests. It might be accompanied by the Acknowledgments reduced sample and small effects of chatbot support caused by a one-time, short interaction. Previous studies that are consistent with This paper is based on some of the data and contents that appeared our main results (e.g., Fitzpatrick et al., 2017; Urakami et al., 2019; Ta in a master’s thesis: Effects of Mind Perception on Social Support of et al., 2020; Bae Brandtzæg et al., 2021), however, support the Chatbots (Lee, 2022). We appreciate the contributions of Yoonwon significance of our findings. Nevertheless, future studies should Jung in data coding. be conducted with larger samples and more effective chatbot support to validate the replication of our results. Conflict of interest 7 Conclusion The authors declare that the research was conducted in the absence of any commercial or financial relationships that could Our study shows that user mind perception and the properties of be construed as a potential conflict of interest. different social support should be considered together to offer more efficient chatbot support. With the rapid advance of AI technologies, chatbots are increasingly permeating people’s daily lives and are Publisher’s note utilized to enhance individuals’ psychological well-being by providing various social support. Thus, in this situation, it is necessary to explore All claims expressed in this article are solely those of the authors ways to improve the quality of user experience with chatbot support. and do not necessarily represent those of their affiliated organizations, Our study suggests one way to enhance user acceptance of chatbot or those of the publisher, the editors and the reviewers. Any support, which is to provide different types of social support (e.g., product that may be evaluated in this article, or claim that may be helpful advice or empathy for users’ situations) in consideration of the made by its manufacturer, is not guaranteed or endorsed by the users’ mind perception. publisher. Frontiers in Psychology 09 frontiersin.org Lee and Hahn 10.3389/fpsyg.2024.1282036 References Adam, M., Wessel, M., and Benlian, A. (2021). AI-based chatbots in customer service Inkster, B., Sarda, S., and Subramanian, V. (2018). An empathy-driven, conversational and their effects on user compliance. Electron. Mark. 31, 427–445. doi: 10.1007/ artificial intelligence agent (Wysa) for digital mental well-being: real-world data s12525-020-00414-7 evaluation mixed-methods study. JMIR Mhealth Uhealth 6:e12106. doi: 10.2196/12106 Araujo, T. (2018). Living up to the chatbot hype: the influence of anthropomorphic Jacobs, O. L., Gazzaz, K., and Kingstone, A. (2022). Mind the robot! Variation in design cues and communicative agency framing on conversational agent and company attributions of mind to a wide set of real and fictional robots. Int. J. Soc. Robot. 14, perceptions. Comput. Hum. Behav. 85, 183–189. doi: 10.1016/j.chb.2018.03.051 529–537. doi: 10.1007/s12369-021-00807-4 Bae Brandtzæg, P. B., Skjuve, M., Kristoffer Dysthe, K. K., and Følstad, A. (2021). Kim, J., Kim, Y., Kim, B., Yun, S., Kim, M., and Lee, J. (2018). “Can a machine tend to “When the social becomes non-human: young People’s perception of social support in teenagers’ emotional needs? A study with conversational agents” in Extended abstracts chatbots” in Proceedings of the 2021 CHI conference on human factors in computing of the 2018 CHI conference on human factors in computing systems (Montreal: ACM), systems (Yokohama, New York, NY: ACM), 1–13. 1–6. Banks, J. (2020). Theory of mind in social robots: replication of five established human Kwon, S. M. (2017). Psychology of human relationships (3rd). Seoul: Hakjisa. tests. Int. J. Soc. Robot. 12, 403–414. doi: 10.1007/s12369-019-00588-x Lee, I. (2022). Effects of mind perception on social support of chatbots. Master’s thesis. Banks, J. (2021). Of like mind: the (mostly) similar mentalizing of robots and humans. Seoul National University. J. Mind Behav. 1. doi: 10.1037/tmb0000025 Lee, S., Lee, N., and Sah, Y. J. (2020). Perceiving a mind in a chatbot: effect of mind Bickmore, T. W., and Picard, R. W. (2004). “Towards caring machines” in CHI'04 perception and social cues on co-presence, closeness, and intention to use. Int. J. Hum. extended abstracts on human factors in computing systems (Vienna: ACM), 1489–1492. Comput. Interact. 36, 930–940. doi: 10.1080/10447318.2019.1699748 Brandtzaeg, P., and Følstad, A. (2017). “Why people use chatbots” in Internet science: Lin, N., Simeone, R. S., Ensel, W. M., and Kuo, W. (1979). Social support, stressful life 4th international conference on internet science (Thessaloniki: Springer), 377–392. events, and illness: a model and an empirical test. J. Health Soc. Behav. 20, 108–119. doi: Brave, S., Nass, C., and Hutchinson, K. (2005). Computers that care: investigating the 10.2307/2136433 effects of orientation of emotion exhibited by an embodied computer agent. Int. J. Hum. Liu, B., and Sundar, S. S. (2018). Should machines express sympathy and empathy? Comput. 62, 161–178. doi: 10.1016/j.ijhcs.2004.11.002 Experiments with a health advice chatbot. Cyberpsychol. Behav. Soc. Netw. 21, 625–636. Cobb, S. (1976). Social support as a moderator of life stress. Psychosom. Med. 38, doi: 10.1089/cyber.2018.0110 300–314. doi: 10.1097/00006842-197609000-00003 Low, J., and Perner, J. (2012). Implicit and explicit theory of mind: state of the art. Br. Cohen, S. (2004). Social relationships and health. Am. Psychol. 59, 676–684. doi: J. Dev. Psychol. 30, 1–13. doi: 10.1111/j.2044-835X.2011.02074.x 10.1037/0003-066X.59.8.676 Ly, K. H., Ly, A.-M., and Andersson, G. (2017). A fully automated conversational agent Cohen, S., and Wills, T. A. (1985). Stress, social support, and the buffering hypothesis. for promoting mental well-being: a pilot RCT using mixed methods. Internet Interv. 10, Psychol. Bull. 98, 310–357. doi: 10.1037/0033-2909.98.2.310 39–46. doi: 10.1016/j.invent.2017.10.002 Dahlbäck, N., Jönsson, A., and Ahrenberg, L. (1993). Wizard of Oz studies — why and Mehta, A., Niles, A. N., Vargas, J. H., Marafon, T., Couto, D. D., and Gross, J. J. (2021). how. Knowl. Based Syst. 6, 258–266. doi: 10.1016/0950-7051(93)90017-N Acceptability and effectiveness of artificial intelligence therapy for anxiety and depression (Youper): longitudinal observational study. J. Med. Internet Res. 23:e26771. Disalvo, C. F., Gemperle, F., Forlizzi, J., and Kiesler, S. (2002). “All robots are not doi: 10.2196/26771 created equal: the design and perception of humanoid robot heads” in Proceedings of the 4th conference on designing interactive systems: Processes, practices, methods, and Meng, J., and Dai, Y. (2021). Emotional support from AI chatbots: should a supportive techniques (London, New York, NY: ACM), 321–326. partner self-disclose or not? J. Comput. Mediat. Commun. 26, 207–222. doi: 10.1093/ jcmc/zmab005 Epley, N., Waytz, A., and Cacioppo, J. T. (2007). On seeing human: a three-factor theory of anthropomorphism. Psychol. Rev. 114, 864–886. doi: Nass, C., and Moon, Y. (2000). Machines and mindlessness: social responses to 10.1037/0033-295X.114.4.864 computers. J. Soc. Issues. 56, 81–103. doi: 10.1111/0022-4537.00153 Esses, V. M., Veenvliet, S., Hodson, G., and Mihic, L. (2008). Justice, morality, and the Nass, C., Steuer, J., and Tauber, E. R. (1994). “Computers are social actors” in dehumanization of refugees. Soc. Justice Res. 21, 4–25. doi: 10.1007/s11211-007-0058-4 Proceedings of the SIGCHI conference on human factors in computing systems (Boston, MA, New York, NY: ACM), 72–78. Faul, F., Erdfelder, E., Buchner, A., and Lang, A.-G. (2009). Statistical power analyses using G*power 3.1: tests for correlation and regression analyses. Behav. Res. Methods. Nosek, B. A. (2007). Implicit-explicit relations. Curr. Dir. Psychol. Sci. 16, 65–69. doi: 41, 1149–1160. doi: 10.3758/BRM.41.4.1149 10.1111/j.1467-8721.2007.00477.x Fitzpatrick, K. K., Darcy, A., and Vierhile, M. (2017). Delivering cognitive behavior Rapp, A., Curti, L., and Boldi, A. (2021). The human side of human-chatbot therapy to young adults with symptoms of depression and anxiety using a fully interaction: a systematic literature review of ten years of research on text-based chatbots. automated conversational agent (Woebot): a randomized controlled trial. JMIR Ment. Int. J. Hum. Comput. Stud. 151:102630. doi: 10.1016/j.ijhcs.2021.102630 Health. 4:e7785. doi: 10.2196/mental.7785 Reeves, B., and Nass, C. I. (1996). The media equation: How people treat computers, Gambino, A., Fox, J., and Ratan, R. A. (2020). Building a stronger CASA: extending television, and new media like real people and places. Cambridge: Cambridge the computers are social actors paradigm. Hum. Mach. Commun. 1, 71–86. doi: University Press. 10.30658/hmc.1.5 Schanke, S., Burtch, G., and Ray, G. (2021). Estimating the impact of “humanizing” Go, E., and Sundar, S. S. (2019). Humanizing chatbots: the effects of visual, identity customer service chatbots. Inf. Syst. Res. 32, 736–751. doi: 10.1287/isre.2021.1015 and conversational cues on humanness perceptions. Comput. Hum. Behav. 97, 304–316. Schuetzler, R. M., Grimes, G. M., and Scott Giboney, J. (2020). The impact of chatbot doi: 10.1016/j.chb.2019.01.020 conversational skill on engagement and perceived humanness. J. Manag. Inf. Syst. 37, Gong, L. (2008). How social is social responses to computers? The function of the 875–900. doi: 10.1080/07421222.2020.1790204 degree of anthropomorphism in computer representations. Comput. Hum. Behav. 24, Short, J., Williams, E., and Christie, B. (1976). The social psychology of 1494–1509. doi: 10.1016/j.chb.2007.05.007 telecommunications. New York, NY: John Wiley & Sons. Gray, H. M., Gray, K., and Wegner, D. M. (2007). Dimensions of mind perception. Shum, H. Y., He, X. D., and Li, D. (2018). From Eliza to XiaoIce: challenges and Science. 315:619. doi: 10.1126/science.1134475 opportunities with social chatbots. Front. Inf. Technol. Electron. Eng. 19, 10–26. doi: Gray, K., and Wegner, D. M. (2012). Feeling robots and human zombies: mind 10.1631/FITEE.1700826 perception and the uncanny valley. Cognition. 125, 125–130. doi: 10.1016/j. Stein, J. P., and Ohler, P. (2017). Venturing into the uncanny valley of mind—the cognition.2012.06.007 influence of mind attribution on the acceptance of human-like characters in a virtual Hobfoll, S. E. (1988). The ecology of stress. Abingdon-on-Thames: Taylor & Francis. reality setting. Cognition. 160, 43–50. doi: 10.1016/j.cognition.2016.12.010 Hodson, G., and Costello, K. (2007). Interpersonal disgust, ideological orientations, Ta, V., Griffith, C., Boatfield, C., Wang, X., Civitello, M., Bader, H., et al. (2020). User and dehumanization as predictors of intergroup attitudes. Psychol. Sci. 18, 691–698. doi: experiences of social support from companion chatbots in everyday contexts: thematic 10.1111/j.1467-9280.2007.01962.x analysis. J. Med. Internet Res. 22:e16235. doi: 10.2196/16235 Holmstrom, A. J., Burleson, B. R., and Jones, S. M. (2005). Some consequences for Urakami, J., Moore, B. A., Sutthithatip, S., and Park, S. (2019). “Users’ perception of helpers who deliver “cold comfort”: why it’s worse for women than men to be inept when empathic expressions by an advanced intelligent system” in Proceedings of the 7th providing emotional support. Sex Roles. 53, 153–172. doi: 10.1007/s11199-005-5676-4 international conference on human-agent interaction (Kyoto, New York, NY: ACM), 11–18. House, J. S., Kahn, R. L., McLeod, J. D., and Williams, D. (1985). “Measures and concepts of social support” in Social support and health. eds. S. Cohen and S. L. Syme. Waytz, A., Gray, K., Epley, N., and Wegner, D. M. (2010). Causes and consequences of (Cambridge, MA: Academic Press), 83–108. mind perception. Trends Cogn. Sci. 14, 383–388. doi: 10.1016/j.tics.2010.05.006 Hu, Q., Lu, Y., Pan, Z., Gong, Y., and Yang, Z. (2021). Can AI artifacts influence human Yam, K. C., Bigman, Y. E., Tang, P. M., Ilies, R., De Cremer, D., Soh, H., et al. (2021). cognition? The effects of artificial autonomy in intelligent personal assistants. Int. J. Inf. Robots at work: people prefer—and forgive—service robots with perceived feelings. J. Manag. 56:102250. doi: 10.1016/j.ijinfomgt.2020.102250 Appl. Psychol. 106, 1557–1572. doi: 10.1037/apl0000834 Frontiers in Psychology 10 frontiersin.org