Great Chain of Agents: The Role of Metaphorical Representation of Agents in Conversational Crowdsourcing

Document Title

Great Chain of Agents: The Role of Metaphorical Representation of Agents in Conversational Crowdsourcing

Document Information

• Research Domain: Conversational agent studies in HCI and crowdsourcing tasks
• Keywords: Conceptual metaphor, conversational agents, crowdsourcing, user engagement, trust, Great Chain of Being theory, human-computer interaction, AI interaction

Research Background and Problem

• Identified Problem or Challenge:
  Existing research primarily focuses on human-like metaphorical representations, neglecting the extensive use of non-human metaphors. Furthermore, there is a lack of systematic studies on how non-human metaphors affect user experiences with conversational agents (e.g., engagement, cognitive load, intrinsic motivation, and trust).

• Why It Matters:
  Conversational agents are widely applied across various domains, and their design is critical to task efficiency and user experience. A better understanding of how metaphors shape user experiences with agents can offer new perspectives and design frameworks for agent development.

• Research Motivation and Related Work:
  Conceptual metaphor theory and the "Great Chain of Being" framework provide insightful perspectives for understanding the role of metaphors in AI and HCI. Additionally, related studies have shown that different metaphorical representations significantly influence users' perceptions of "warmth and competence," affecting expectations and acceptance of technology.

Solution

• Proposed Method or Solution:
  The authors adopted the "Great Chain of Being" framework (a hierarchical system including gods, humans, animals, plants, and inorganic objects) to compare how different metaphorical representations of conversational agents impact user experience. The study involved 12 conditions (6 types of metaphors × 2 task types) and recruited 341 participants from the crowdsourcing platform Prolific for experiments.

• Innovative Contributions:

  • Conducted the first systematic study of non-human metaphors in conversational agents using the "Great Chain of Being" framework.
  • Explored the trade-offs between cognitive load, motivation, and engagement brought by different metaphors.
  • Investigated the potential of metaphor designs beyond human-centered approaches.

• Implementation Steps and Techniques:

  • Conducted preliminary research to select representative metaphorical roles (e.g., dog, book, avocado).
  • Used the text-based conversational agent tool TickTalkTurk to simulate two typical microtasks (information retrieval and image classification).
  • Measured user engagement (UES-SF), intrinsic motivation (IMI), cognitive load (NASA-TLX), and trust (TiA) through questionnaires and experimental data.

Research Findings

• Specific Results:

  • Impact of Metaphorical Roles:
    • Animal metaphors (e.g., dog) enhanced user engagement and aesthetic appeal but increased cognitive load.
    • Inorganic metaphors (e.g., book) reduced cognitive load but diminished user interest and intrinsic enjoyment.
    • "God" metaphors raised user expectations but resulted in higher cognitive load and lower usability perception.
  • Impact of Task Types:
    • Image classification tasks were more engaging than information retrieval tasks but also demanded higher cognitive effort.

• Comparison with Existing Solutions:
  The study demonstrated the potential of non-human metaphorical representations and highlighted the need to balance engagement and cognitive load in future design choices. Compared to human metaphors (e.g., "professional"), animal and inorganic metaphors exhibited unique advantages.

• Experimental or Evaluation Results:

  • Statistical significance analysis revealed that metaphorical designs significantly impacted user experiences (e.g., interest/enjoyment scores, cognitive load scores).
  • Different metaphor and task type designs did not significantly affect task completion accuracy or time.

• Limitations and Future Directions:

  • Limitations:
    • Metaphor comprehension may be influenced by cultural backgrounds (this study was limited to English-speaking participants).
    • Human metaphor role selection did not fully consider gender and cultural influences.
  • Future Directions:
    • Expand to more task types and dynamic multimodal interactions.
    • Explore whether personalized metaphor designs can further enhance user experience.
    • Investigate designs that increase warmth in "god" metaphors.