Conversational AI -- Open Questions

Open emerging from sustained interaction with conversational AI systems (e.g. ChatGPT). The focus is not on solutions or features, but on points where current conversational patterns feel brittle, overconfident, or under-theorized.

Voice-Led Input

  • How does speaking, rather than typing, change the way intent forms before it becomes a request?

  • When users speak to AI, how often are they articulating a task versus externalizing a state of thought?

  • What interaction signals (pauses, hesitations, self-corrections) are lost when spoken input is immediately collapsed into text?

  • How long can a system wait before silence feels attentive rather than broken?

  • When does a clarifying question support thought, and when does it interrupt it?

  • What responsibility does a system have when a user speaks without yet knowing what they want?

Follow-Up, Continuation, and User Response

  • How do conversational AI systems currently shape what users say next?

  • What agency is lost when users are asked to select from system-generated options rather than form responses?

  • What does it mean for a system to explain why certain follow-ups exist, not just offer them?

  • Can exposing the reasoning behind continuations help users decide how to respond?

  • How might systems support response formation by offering language users could adopt, rather than paths to explore?

  • Where is the boundary between supporting articulation and coaching behavior?

Silence, Restraint, and Non-Action

  • When is it better for a conversational AI system not to respond?

  • How might silence function as an outcome rather than a failure?

  • What conversational moments benefit from acknowledgment without continuation?

  • How do engagement-driven metrics discourage restraint?

  • What kinds of agency become possible when the system does less?

Cross-Cutting

  • What assumptions from text-based interaction are being carried into voice-first systems?

  • Which conversational behaviors are optimized because they demo well rather than because they support reasoning?

  • How do conversational systems encode norms about speed, helpfulness, and progress?

  • What does it mean to treat conversational interaction itself as a form of work?