Which Visual Modality Is Important When Judging the Naturalness of the Agent (Artificial Versus Human Intelligence) Providing Recommendations in the Symbolic Consumption Context?

摘要

This study aimed to explore how the type and visual modality of a recommendation agent’s identity affect male university students’ (1) self-reported responses to agent-recommended symbolic brand in evaluating the naturalness of virtual agents, human, or artificial intelligence (AI) and (2) early event-related potential (ERP) responses between text- and face-specific scalp locations. Twenty-seven participants (M = 25.26, SD = 5.35) whose consumption was more motivated by symbolic needs (vs. functional) were instructed to perform a visual task to evaluate the naturalness of the target stimuli. As hypothesized, the subjective evaluation showed that they had lower attitudes and perceived higher unnaturalness when the symbolic brand was recommended by AI (vs. human). Based on this self-report, two epochs were segmented for the ERP analysis: human-natural and AI-unnatural. As revealed by P100 amplitude modulation on visual modality of two agents, their evaluation relied more on face image rather than text. Furthermore, this tendency was consistently observed in that of N170 amplitude when the agent identity was defined as human. However, when the agent identity was defined as AI, reversed N170 modulation was observed, indicating that participants referred more to textual information than graphical information to assess the naturalness of the agent.