Single or Multiple Conversational Agents? An Interactional Coherence Comparison
summary
A 24‑participant Wizard of Oz study tests three domain bots (shopping, nature, culture) under two conditions: one meta‑bot vs. three visible bots. Conversation structure shows no statistical difference, but users report more confusion with multiple bots and invent turn‑taking strategies; 65% ignored messages in single and 72% ignored messages in multi, driven by irrelevant or ill‑timed “seed” messages. Conclusion: a meta‑bot isn’t required for coherence, but a clear turn-taking protocol and less disruptive topic injection are necessary.
AI-written summary · about
chapters (11)
- 0:00Introduction to chatbots
- 1:53Two approaches to complex tasks
- 3:00Interactional coherence research
- 4:05Experiment setup and scenarios
- 5:33Turn-taking analysis results
- 7:24Topic change frequency findings
- 8:50Sequential coherence observations
- 11:27Disruptive topic introduction
- 12:29Additional user feedback
- 14:18Main takeaways and conclusions
- 15:23Q&A session
related talks
45mAn Agent Walks Into a Bar... and Meets Another Agent: Multi-Agent Systems in Action
50mThe Art of Coding a Conversation: Designing Bots
19mBigBlueBot: teaching strategies for successful human-agent interactions
25mAI Orchestration: Getting Super Agents to Complete a Mission
42mConversational Assistants that Converse with Each Other
19mFrom Demo to Deployment: Multimodal Conversational Agents
17mChaos Testing for Chatbots: Simulating Customers to Evaluate AI Agents
14mStorySage: Conversational Autobiography Writing Powered by a Multi-Agent Framework