Computers in Human Behavior: Artificial Humans最新文献

The banking industry is infusing AI systems into service encounters while dissolving some traditional services. This study aims to empirically test an exploratory framework to identify how human-AI interactions differ when engaged in basic or advanced virtual agent usage contexts. A conceptual framework was developed to examine consumer perceptions of basic and advanced virtual agent usage intentions. Five independent variables of trust in AI, perceived security in AI, perceived AI expertise, comfort in using AI technologies, and need for social presence were explored. Data was collected from 322 respondents and analyzed using multivariate regression. The findings suggest that consumers do not perceive service encounters with virtual agents from a “one size fits all” approach. Consumers perceive different value-in-use perceptions based on the complexity of the usage contexts. Our results suggest that success in advanced virtual agent encounters may require social presence for robust human-AI interaction. Additionally, this study extends the digital servitization and service robot acceptance model (sRAM) literature by evaluating consumer value-in-use perceptions with empirical evidence.

Large language models offer significant potential for increasing labour productivity, such as streamlining personnel selection, but raise concerns about perpetuating systemic biases embedded into their pre-training data. This study explores the potential ethnic and gender bias of ChatGPT—a chatbot producing human-like responses to language tasks—in assessing job applicants. Using the correspondence audit approach from the social sciences, I simulated a CV screening task with 34,560 vacancy–CV combinations where the chatbot had to rate fictitious applicant profiles. Comparing ChatGPT's ratings of Arab, Asian, Black American, Central African, Dutch, Eastern European, Hispanic, Turkish, and White American male and female applicants, I show that ethnic and gender identity influence the chatbot's evaluations. Ethnic discrimination is more pronounced than gender discrimination and mainly occurs in jobs with favourable labour conditions or requiring greater language proficiency. In contrast, gender bias emerges in gender-atypical roles. These findings suggest that ChatGPT's discriminatory output reflects a statistical mechanism echoing societal stereotypes. Policymakers and developers should address systemic bias in language model-driven applications to ensure equitable treatment across demographic groups. Practitioners should practice caution, given the adverse impact these tools can (re)produce, especially in selection decisions involving humans.

The increasing integration of robots as service providers or companions in daily life has prompted extensive research on the emotional aspects of human-robot interaction (HRI). Emotional contagion, a crucial factor influencing user experience in HRI, has gained considerable attention. However, limited research explores the influence of anthropomorphism and gender differences in robot modeling on emotional contagion in HRI, leaving a gap in reference guidance for social robot design and application. To address this, we investigate human-robot interactions in service scenarios, analyzing the impact of robot appearance, anthropomorphism, and gender on the transmission of positive and negative emotions. The experimental findings highlight the significant role of robot gender in influencing emotional contagion in HRI, while revealing the interaction effect between robot gender and anthropomorphism on emotional contagion. This interdisciplinary study provides empirical evidence that enriches research on the emotional aspects of HRI and contributes to more informed design of user experiences.

For the first time, voice-based assistants (VBAs) allow studying the mass consumption of a robotic product in a ‘natural’ environment. The present paper investigates users' perspectives concerning Alexa, informed by the scholarly literature on the diffusion and appropriation of digital media, VBAs, and social robots, in general, in the domestic sphere. Besides CASA, a paradigm widely applied in robotics, we draw on theories on users and theories on the use of users. We explored individuals' use of Alexa in two countries—the US and Italy— through an online survey. Results indicated that: (1) Alexa's use follows the same modalities of the previous digital media; (2) Alexa's use is in an early phase of domestication, and thus, the role of users in reshaping this technological artifact is still limited; and (3) updates to the CASA paradigm are needed.

Current research in human-agent interaction increasingly focuses on multimodal interactions with anthropomorphic virtual agents. However, most existing paradigms primarily emphasize only the user's perception, overlooking the actual emergent interaction processes. This paper addresses this gap by presenting a novel methodological approach that scrutinizes both the relationship and content levels of human-agent interaction. Utilizing a unique combination of mixed reality representations and mixed methods, our approach aims to uncover and analyze the joint behavior of human users and agents during interactions. Our approach offers new insights into the dynamics of proto-social human-agent interactions, with implications for improving the design and functionality of virtual agents in mixed reality settings.

Near humanlike artificial entities can appear eerie or uncanny. This uncanny valley is here investigated by testing five individual difference measures as predictors of uncanniness throughout a variety of stimuli. Coulrophobia predicted uncanniness of distorted faces, bodies, and androids and clowns; disgust sensitivity predicted the uncanniness of some distorted faces; the anxiety facet of neuroticism predicted the uncanniness of some distorted faces, bodies, and voices; deviancy aversion and need for structure predicted uncanniness of distorted places and voices. Taken together, the results suggest that while uncanniness can be caused by multiple, domain-independent (e.g., deviancy aversion) and domain-specific (e.g., disease avoidance) mechanisms, the uncanniness of androids specifically may be related to a fear of clowns, potentially due to a dislike of exaggerated human proportions.

A perceived personality of a chatbot or conversational agent is mainly conveyed by the way they communicate verbally. In this online vignette study (N = 168) we examined the possibility of conveying personality in short social-media-like messages by adding simple conversational cues. Social-oriented and responsive conversational cues, as well as their combination had distinct effects on the perceived personalities of the chatbots. Social-oriented cues had a clear effect on most OCEAN personality traits, warmth, and anthropomorphism, while responsive cues only affected neuroticism. In combination, effects of social-oriented cues were countered by responsive cues, but not for all personality traits. Competence and trust were not affected by any of the used conversational cues. The findings show that very few conversational cues are sufficient to convey distinct personalities in short messages.

Human-computer interaction (HCI) research is facing a vital question of the effectiveness of personas generated using artificial intelligence (AI). Addressing this question, this research explores user perceptions of AI-generated personas for textual content (GPT-4) and two image generation models (DALL-E and Midjourney). We evaluate whether the inclusion of images in AI-generated personas impacts user perception or if AI text descriptions alone suffice to create good personas. Recruiting 216 participants, we compare three AI-generated personas without images and those with either DALL-E or Midjourney-created images. Contrary to expectations from persona literature, the presence of images in AI-generated personas did not significantly impact user perceptions. Rather, the participants generally perceived AI-generated personas to be of good quality regardless of the inclusion of images. These findings suggest that textual content, i.e., the persona narrative, is the primary driver of user perceptions in AI-generated personas. Our findings contribute to the ongoing AI-HCI discourse and provide recommendations for designing AI-generated personas.

Background

Human-in-the-loop research predominantly examines the interaction types and effects. A more structural and pragmatic exploration of humans and Artificial Intelligence or AI is lacking in the AI in education literature.

Purpose

In this systematic review, we follow the Entity-Relationship (ER) framework to identify trends in the entities, relationships, and attributes of human-in-the-loop AI in education.

Methods

An overview of N = 28 reviewed studies followed by their ER characteristics are summarized and analyzed.

Results

The dominant number of two or three-entity studies, one-sided relationships, few attributes, and many to many cardinalities may signal a lack of deliberation on beings that come to interact and influence human-in-the-loop and AI in education.

Conclusion

The contribution of this work is identifying the implications of human-in-the-loop and AI from a more formal ER perspective and acknowledging the many possibilities for placement of humans in the loop with the AI, system, and environment of interest.

Background

Stress is a significant issue amongst university students, yet limited psychological services are available. Mindfulness is effective for stress reduction and can be delivered digitally to expand access to student populations. However, digital interventions often suffer from low engagement and poor adherence. A virtual human may improve engagement and adherence through its humanlike appearance and behaviours.

Objective

To examine whether a virtual human could reduce stress in university students at least as much as a teletherapist, and more than a chatbot, using a mindfulness intervention.

Methods

Stressed university students (N = 158) were randomly allocated to the virtual human (N = 54), chatbot (N = 54), or teletherapist (N = 50). 36 participants received each condition. Participants completed one lab session and were asked to do online homework sessions at least twice weekly for four weeks. Changes in self-reported stress and mindfulness, physiological stress indices, homework completion, and perceptions of the agent were compared between groups. Thematic analysis was conducted on participants’ responses to open-ended questions about the interventions.

Results

There were significant reductions in stress and increases in mindfulness across all groups. All groups had higher peripheral skin temperature post-intervention, and only the teletherapy group had higher electrodermal activity (reflecting elevated stress) post-intervention compared to baseline. There were no significant changes in heart rate. Homework adherence was significantly higher in the virtual human group, whereas homework satisfaction and engagement were lowest in the chatbot group. Thematic analysis found that people thought the robotic voice of the virtual human could be improved, the chatbot could be improved by adding audio, and that participants experienced feelings of judgement from the teletherapist.

Discussion

Overall, results support use of virtual humans for delivering mindfulness interventions in stressed students. Virtual humans may have the advantage over teletherapy and chatbots of increasing adherence in student populations, but more work is needed to increase perceived empathy and replicate results in other populations.