Faking It Right: Human-Likeness in Pragmatic Authenticity

Abstract

Authenticity is a prominent concept in human-computer interaction (HCI) research, particularly in the context of developing ethically sound technology. Within HCI, authenticity is typically approached from two perspectives: theoretical and pragmatic. The theoretical perspective evaluates whether agents can possess characteristics and qualities equivalent to human traits, while the pragmatic perspective assesses whether agents can display behaviors and interact in a manner that appears genuine and coherent. While both perspectives involve moral considerations, the pragmatic approach is particularly useful for understanding the relationship between agents and humans [1]. It aligns with the practical needs of industry and provides empirically testable outcomes essential for both academic and real-world application.

The theoretical framework of pragmatic authenticity proposed in this thesis has a hierarchical structure, where authenticity is treated as a formative construct dependent on the interplay of three factors: trustworthiness (individual and institutional), intelligence, and human-likeness. Each factor contains defining facets relevant to authenticity perception. In addition to introducing a novel framework, this research empirically focuses on the factor of human-likeness and its relationship to perceived authenticity. Specifically, it explores the facets of anthropomorphization, social presence, and empathy during human-agent interactions. These facets are indicated by intentional social design cues such as politeness, personalization, and responsiveness in the agent’s responses. The goal of this study is to determine whether latent factors underlie the three facets and if these factors can predict perceived authenticity. These questions are valuable in both academia and the technology industry by making safety an integral part of the intelligent design process.

Guiding this research are both theoretical and methodological approaches within HCI. HCI focuses on the relationship between the artificial agent and the human, rather than treating them as individual entities. Building upon existing work, this study involves human participants interacting with an artificial social agent, ChatGPT-4o. The content of their three interactions is based on advice-seeking vignettes, provided to participants prior to engaging with the agent. The purpose of the vignettes is to create an interactive context that enhances the showcasing of social cues in agents and allows for their evaluation. Participants then provide self-assessments, stating their perceptions of the agent’s anthropomorphism, social presence, empathy, and authenticity.

Using a within-subject design for data collection while aggregating data for exploratory factor and regression analyses ensures that individual differences are controlled, while measuring the impact of factors. The main hypotheses suggest an underlying factor that will significantly predict perceived authenticity.

The findings of this research could substantially impact the field of responsible, ethical design by unveiling the implications of social design cues on human perception of agent authenticity. Understanding how authenticity is organized and exploring its relationship with humans is crucial for future industry developments. Practical insights gained from this study could inform technology development practices, helping to avoid designs labeled as deceitful, exploitative, misinformative, or fraudulent.

References

[1] M. Coeckelberg, "Are emotional robots deceptive?," IEEE transactions on affective computing, vol. 53, no. 4, pp. 388–393, 2011.