Persuasion in human-robot interaction : understanding persuasive interactions between LLM-based social robots and humans

This thesis investigates the perceived persuasive effectiveness of generative social robots (GSRs) powered by large language models (LLMs) in human-robot interaction (HRI). Building on the Robot Knowledge Persuasion Model (RKPM), the study examines how different types of robot knowledge, e.g. Self-Knowledge (SK), User-Knowledge (UK), and Context-Knowledge (CK), influence user attitudes and persuasion outcomes. A between-subjects experiment (N = 46) was conducted, using a structured interaction protocol in which participants engaged in personalised dialogues with a LLM-powered robot named "Charlie" across three decision-making domains: fitness, nutrition, and financial investment. The interactions were followed by a reassessment of participants’ resource allocations, which served as an indirect behavioural indicator of persuasive success. Multilevel structural equation modelling revealed that only UK significantly increased perceived persuasive behaviour, while SK and CK had no measurable effect. A sequential mediation pathway was confirmed: UK improved attitudes toward the robot, which increased perceived persuasive behaviour, which in turn predicted changes in decision-making. These findings provide partial support for the RKPM and underscore the critical role of personalisation in effective human-robot persuasion. The study advances theoretical understanding and practical design considerations for socially intelligent robots, while acknowledging methodological limitations such as the reliance on self-report data and the absence of physical embodiment.