Ontological classification is the systematic organization of entities, concepts, or

phenomena based on their fundamental nature, properties, and relationships. It is crucial for structuring the world (Aristotle, “Categories”) and constructing social reality (Searle, 1995). Machine agents and digital interlocutors have been challenging ontological categories due to their dynamic and hybrid nature (Haraway, 1991; Mazis 2008). Consequently, there is a growing body of research on ontological classification and related to it in HMC (Banks &

Bowen 2025; Edwards, 2018; Edwards et al., 2019; Etzrodt, 2021; Etzrodt & Engesser, 2021; Gunkel, 2000; Guzman, 2020; Koban & Banks, 2024). We aim to contribute to this research by pursuing the question of how individuals assign ontological categories to machine agents.

In interpersonal communication, people rely on processes such as the “theory of mind” they naturally have about other people in order to decide whether they can trust the interlocutor’s advice. This enables people to reconstruct the mental processes of other people. While this is helpful to develop a joint understanding with fellow humans, this must fail when trying to understand the “minds” of machines as the semantics of machine “intelligence” is different from human intelligence even with approaches like neural networking which tries to emulate biological intelligence.

Future research therefore needs to scrutinize how humans’ ability to accurately assess what is going on in machine “minds” can be optimized. Helping humans to build a “theory of the artificial mind” is proposed to be a prerequisite for a well-functioning human-machine interaction.

With the growing presence of virtual influencers in digital and marketing spaces, this study explores how people perceive these virtual influencers and how their perceptions shift upon learning about their artificial nature. Drawing on the human-to-human interaction script, an online study (N = 469) examined participants’ interpersonal attraction toward a virtual influencer, Rozy, before and after disclosure of her artificiality. Findings revealed that those who initially perceived Rozy as human reported stronger attraction, but their perceptions significantly declined after disclosure of her artificiality. However, when the expectancy violation was perceived more positively, the decrease in attraction was smaller, and participants showed greater liking for the advertisement featuring Rozy and stronger intentions to seek more information about the advertised product. These results highlight the complexities of human-virtual influencer interactions and offer valuable insights and important directions for future research.

Recent advancements in natural language processing and generative AI have significantly reshaped communication landscapes by enabling AI-chatbots to act as fully automated communicators. This study focuses on chat interactions as a field that is increasingly permeated by AI agents. We tested whether and how the social scripts humans apply in chat interactions differ and additionally assessed participants' perceived agency in a between-subjects experiment. 83 first-semester students completed a chat task with either a human or AI chat partner in a lab. We used shinyChatR package to create a chat interface and implemented a customized version of ChatGPT for the AI condition. All hypotheses had to be rejected, but 20 qualitative interviews we additionally conducted with participants reveal potential reasons. Overall, our results imply that human-human and human-AI interactions only differ marginally in settings where the identity of the interaction partners is not disclosed.

In contrast to a growing number of examples documenting biased outputs of algorithms, users might perceive AI and its outputs to be unbiased and more objective than humans. We focus on image-generating AI as a potentially biased source of digital imagery. In an online experiment (N = 495) we test whether labelling images as AI-generated could decrease users’ ability to detect if a set of images neglects certain groups and weaken negative reactions to this incident. Surprisingly, we find that results depend on the pictured group (college students or older people). AI artist information reduces perceived bias but only for pictures of college students. In contrast, AI-generated images showing older people increases negative reactions regardless of actual bias.