Data-driven services are becoming an ever more vital part of the business landscape, with chatbots an integral component of this change. Functioning as a longstanding barrier to the efficacy of personalized services, however, are user concerns surrounding data management that culminate in managerial challenges such as the Personalization-Privacy paradox (i.e., where informing users about data management practices in personalized services heightens privacy concerns and reduces willingness to disclose). In the current paper, across two experiments, we tackle this paradox by manipulating privacy assurance statements made by the chatbot (i.e., short statements at the start of the service interaction) inspired by Communication Privacy Management theory and service personalization (i.e., depth of medical diagnosis at the services conclusion). Using the Elaboration Likelihood Model as our core theoretical anchoring, we examine direct effects on user data management outcomes (privacy concerns, willingness to disclose) and collaboration with the chatbot (working alliance), as well as indirect effects via the mediator perceived information control to discern the activation of peripheral or central route to persuasion. Results show that when service personalization is low, privacy assurance statements activate the peripheral route as users make a surface assessment of their perceived degree of information control. When service personalization is high, however, they additionally activate the central route as users weigh the information delivered at the start of the service interaction with that delivered at the services conclusion. We support this theoretical rationale with calculations of direct, indirect, and conditional indirect effects alongside tests of equivalency. • Chatbot privacy assurances can improve privacy outcomes and collaboration. • However, their effect depends on the degree of service personalization. • When services are more personalized, privacy assurances are deeply processed. • When services are less personalized, privacy assurances are superficially processed. • In this case, users make a surface assessment as to their level of control.

Today’s virtual reality (VR) environments can offer highly immersive experiences to users and a high sense of presence in these environments. Moreover, this sense of presence can be felt simultaneously in several virtual worlds, while users interact with each and move between them. In the current study, we hypothesized that enhancing humans’ physical capabilities by being present in several virtual worlds simultaneously will contribute positively to task switching training. We evaluated our hypothesis by examining the effectiveness of task switching training when the tasks are executed in one, two, and four different virtual worlds. The study randomly assigned 59 participants to one of three between-participants groups. The results demonstrated that although participants’ training experiences in the 1–World, 2–Worlds, and 4–Worlds groups were similar regarding training time and errors performed during training, the varied environments produced different results in the transfer phase. More worlds caused significantly longer transfer times. Additionally, the 4–Worlds group made many more errors when selecting the correct model to be assembled compared to the 1–World group. Hence, our hypothesis was not supported, with the findings pointing in the opposite direction. Instead of improving the task switching training, switching between tasks located in several different virtual worlds caused performance to deteriorate in the transfer task. The findings might imply that performing tasks set in multiple virtual worlds simultaneously can lower the cognitive load compared to performing them in one environment and can deteriorate the transfer of training. • We examined the effectiveness of task switching training in several virtual worlds. • The tasks were executed in one, two, or four different virtual worlds. • More worlds caused significantly longer transfer times. • The 4–Worlds group made more errors when selecting the correct model.

Designing trustworthy AI-based systems and enabling external parties to accurately assess the trustworthiness of these systems are crucial objectives. Only if trustors assess system trustworthiness accurately, they can base their trust on adequate expectations about the system and reasonably rely on or reject its outputs. However, the process by which trustors assess a system's actual trustworthiness to arrive at their perceived trustworthiness remains underexplored. In this paper, we conceptually distinguish between actual and perceived trustworthiness, trust propensity, trust, and trusting behavior. Drawing on psychological models of how humans assess other people's characteristics, we present the two-level Trustworthiness Assessment Model (TrAM). At the micro level, we propose that trustors assess system trustworthiness based on cues associated with the system. The accuracy of this assessment depends on cue relevance and availability on the system's side, and on cue detection and utilization on the human's side. At the macro level, we propose that individual micro-level trustworthiness assessments propagate across different trustors – one stakeholder's trustworthiness assessment of a system affects other stakeholders' trustworthiness assessments of the same system. The TrAM advances existing models of trust and sheds light on factors influencing the (accuracy of) trustworthiness assessments. It contributes to theoretical clarity in trust research, has implications for the measurement of trust-related variables, and practical implications for system design, stakeholder training, AI alignment, and AI regulation related to trustworthiness assessments.

Highly realistic embodied conversational agents (virtual humans) are starting to be used in healthcare, yet we know little about how their human likeness affects self-disclosure compared to other digital assessment methods. This research aimed to investigate the effects of a computer's human likeness on disclosure during a psychosocial assessment. 160 participants (mean age 24 years, 117 females, 42 males, 1 gender diverse) were randomized to receive a psychosocial interview from a realistic virtual human, a text chatbot, or an online questionnaire. The assessment comprised 18 close-ended items and six open-ended questions on diet, exercise, sexual practices, substance use, recent emotional experiences and loneliness. Socially desirable responding and unwillingness to respond were identified from answers, amount of disclosure was assessed from word count, and perceived anthropomorphism was assessed using self-report. Results demonstrated that for sensitive questions, there was higher socially desirable responding in the virtual human group as shown by significantly lower loneliness scores and higher rates of declining to answer a question compared to the other groups. For non-sensitive questions, socially desirable responding and proportion of declined answers did not differ by group. Participants in the virtual human group used more words when reporting stressful events and positive emotional experiences. Thematic analysis showed people felt rapport with both the virtual human and chatbot, but some felt social evaluative pressure with the virtual human. This supports theories that as human likeness cues increase, humans treat robots more socially. These findings should be considered when humanlike technologies are used clinically. • More human likeness amplifies social desirability for sensitive health queries. • More human likeness reduces willingness to disclose sensitive health information. • More human likeness encourages self-disclosure of positive emotional experiences. • More human likeness invokes a feeling of rapport. • More human likeness can invoke feelings of social evaluative pressure.