AI 中文总结
该研究开发了虚拟狗、猫、马智能体，对比不同行为条件后发现，四足虚拟智能体的说服效果更依赖情感表达等功能性线索，而非物种特有动作，为跨物种设计模式提供了依据。
AI 中文摘要
说服技术越来越多地使用虚拟智能体来影响人们的态度和行为，但现有研究主要集中在人形智能体上。非人形的四足智能体的说服设计仍未得到充分探索，目前尚不清楚情感表达是否能跨物种一致发挥作用，或者是否需要物种特有的动作。我们开发了虚拟狗、猫和马智能体，并比较了三种行为条件：物种特有的行为、跨物种共享的行为，以及仅吠叫的基线条件。参与者完成了包括丢弃垃圾、喂食和不使用智能手机在内的日常任务。我们评估了意图理解、行为意图、实际行为、心理逆反、不适感、熟悉度和智能体接受度。在多个任务情境中，仅吠叫的基线条件产生的意图理解和行为得分低于有表达的条件。因此，情感表达和引导注意力的线索似乎能提升对智能体意图的理解，并支持行为改变。不过，物种特有的行为与共享行为之间未出现一致的显著差异，这表明准确复制动物特有的动作并非说服效果的主要决定因素。心理逆反和不适感始终处于较低水平，而对动物物种的熟悉度在部分条件下与实际行为相关。这些发现表明，四足虚拟智能体的说服更依赖功能性线索，包括情感表达、注意力引导和意图可读性，而非准确的物种特有行为。研究结果支持跨物种的通用性，并为说服技术和人机交互中的可复用设计模式提供了基础。
英文摘要
Persuasive technologies increasingly use virtual agents to influence attitudes and behavior, but research has focused mainly on humanoid agents. The persuasive design of non-humanoid, quadruped agents remains underexplored, and it is unclear whether emotional expression works consistently across animal species or whether species-specific motion is necessary. We developed virtual dog, cat, and horse agents and compared three behavioral conditions: species-specific behavior, shared behavior across species, and a bark-only baseline. Participants completed everyday tasks involving trash disposal, feeding, and refraining from smartphone use. We evaluated intention understanding, behavioral intention, actual behavior, psychological reactance, discomfort, familiarity, and agent acceptance. In several task contexts, the bark-only baseline produced lower intention-understanding and behavioral scores than the expressive conditions. Emotional expression and attention-guiding cues therefore appear to improve interpretation of agent intention and support behavior change. However, no consistent significant differences emerged between species-specific and shared behavior, suggesting that faithful reproduction of animal-specific motion is not the main determinant of persuasive effectiveness. Psychological reactance and discomfort remained low, while familiarity with an animal species was associated with actual behavior in some conditions. These findings indicate that persuasion by quadruped virtual agents depends more on functional cues, including emotional expression, attention guidance, and intention readability, than on accurate species-specific behavior. The results support cross-species generalizability and provide a basis for reusable design patterns in persuasive technology and human-AI interaction.
CommentsPreprint. 68 pages, 4 figures, 8 tables. Submitted to Frontiers in AI
