Language model agents select questions but stay tied to original text
When Models Choose the Question: Pedagogical Constraints in Bottom-Up Multi-Agent Inquiry
Human-Computer Interaction
Summary
This paper looks at how language models come up with and choose questions when they aren't given any to start with. The researchers had AI agents read a story, create questions, pick which ones to discuss, and try to reach a conclusion together, without a main guide. They found that even though the models could ask many different questions, their discussions quickly became narrower and stuck closely to parts of the text already clearly stated. This means that even when given freedom, these AI models tend to focus on what’s explicitly framed and don’t explore widely.
What this means in practice
- •For conversational ai developers: Design multi-agent conversation systems that generate and select questions without predefined prompts but remain faithful to key input text.
- •For educational technology teams: Create AI tools that simulate question-driven discussions around texts while maintaining focus on core material for learning.
Authors
Yeri Hong, Lauren Hyoseo Yoon
Abstract
What shapes a model-generated inquiry when no discussion question is supplied? We introduce a bottom-up forum framework inspired by Philosophy for Children, in which language-model agents read a philosophical narrative, propose and select questions, and develop a shared conclusion without a privileged model facilitator or aggregator. Across 576 forums, contrasting Aristotelian value personas interacted with a blank-slate participant receiving no value-specific instruction. The blank slate remained neutral and was selected more often for conclusions than questions. Yet inquiry narrowed in both form and source: varied initial questions increasingly became either/or alternatives, while discussion concentrated on directions already explicit in the text. Our ECO framework traces this source focus by distinguishing explicit philosophical framing, characters' modeled inquiry, and open-ended narrative material. Across analyzed chapters, participants drew most often on explicit framing. We describe this as a pedagogical constraint: freedom to formulate questions did not necessarily produce freedom from source framing.