Can you really sum up an AI conversation with one score, like 'proactive vs. passive' — or does that flatten what's actually going on?
Can agency frameworks based on single dimensions capture what really happens in conversation?
This explores whether models that score conversation, or an AI's role in it, along one axis (such as who leads, how proactive the AI is, or how aligned it is with the user) can capture how dialogue actually works. The corpus has little that directly evaluates agency frameworks, so this answer builds the case from research on how conversation is structured.
This explores whether a single scale, like 'more agentic vs. less agentic' or 'passive vs. proactive,' can describe what happens when people and AI talk. One caveat first: the collection has almost nothing that tests agency frameworks directly. What it does have is a lot of evidence about how conversation is built, and that evidence points one way. Conversation runs on several independent dimensions at once, and collapsing them into one hides the things that matter most.
The clearest evidence comes from work that tracks many dimensions at the same time. Conversational DNA follows linguistic complexity, emotional trajectory, topic coherence and relevance as parallel streams over time. It finds patterns that summary statistics miss Can tracking dialogue dimensions simultaneously reveal hidden conversation patterns?. The structure is predictive, too. A model that sees only the 'shape' of a conversation, meaning how it unfolds over time and none of the words, predicts user satisfaction almost as well as a full-text LLM analysis. Combining shape and content does much better than either alone Can conversation shape predict whether it will work?. If shape and content each carry separate signal, a framework built on one axis is leaving information out.
The dimensions also don't substitute for each other. A review of alignment research finds that matching someone's words improves task efficiency, while matching their emotional tone builds trust. Mixing the two up produces cold customer-service bots and evasive mental-health assistants Do different types of alignment serve different conversational goals?. Users judge their AI partners on several separate axes too: competence, human-likeness and communicative flexibility each count separately How do users mentally model dialogue agent partners?. Proactivity, the trait most often treated as agency, is a good example. Volunteering useful information can cut conversation length by up to 60% Could proactive dialogue make conversations dramatically more efficient?. But conversation analysis shows that knowing *when* to stop and ask the user is its own skill When should AI agents ask users instead of just searching?. And standard next-turn training pushes models toward passivity in ways a single proactivity dial wouldn't capture Why do language models respond passively instead of asking clarifying questions?.
The most surprising lateral point: some of what happens in conversation may not belong to the AI at all. One line of argument says AI produces 'event residue,' text that carries the marks of communication. On this view, the human does the interpretive work that turns it into an exchange Does AI generate genuine utterances or just text patterns?. Similarly, the small social moves that keep a conversation going, such as repairing a reference or handing off a topic, are relational work rather than information transfer Why don't language models develop conversation maintenance skills?. Mutual understanding is built from both sides across turns Can dialogue systems track both speakers' beliefs across turns?. If agency is spread across both participants like this, a framework that scores only the AI's agency, on one axis, is measuring the wrong unit. Better questions would be: who carries which part of the work, along which dimension, and at what point in the conversation?
Sources 10 notes
Conversational DNA encodes four simultaneous dimensions—linguistic complexity, emotional trajectories, topic coherence, and conversational relevance—as temporal streams. The reverse Turing test finding showed expert assessments of AI diverged sharply, suggesting conversational structure shapes interpretation as much as content.
A structure-only model analyzing conversation trajectory achieved 68% accuracy predicting satisfaction, nearly matching full-text LLM analysis at 70%. Combined structural and textual features reached 80%, showing that how conversations unfold geometrically captures interaction quality text-based classifiers miss.
A 2020–2025 systematic review shows lexical alignment drives task efficiency and comprehension, while emotional and prosodic alignment drive relational warmth and trust. Conflating them in design produces category errors—cold customer-service bots and evasive mental-health assistants.
The Partner Modelling Questionnaire reveals that perceived competence dominates user impressions (49% of variance), followed by human-likeness (32%) and communicative flexibility (19%). This three-factor structure reflects how people evaluate dialogue partners against both functional and social standards.
Simulations show proactivity—providing relevant information without being asked—cuts dialogue turns by 60% in medium-complexity domains. This behavior mirrors human conversation and Grice's maxims but is almost entirely absent from AI datasets and research benchmarks.
Show all 10 sources
Tool-enabled LLMs drift from user intent through silent tool chaining. Conversation analysis reveals insert-expansions—clarifying intent, scoping responses, enhancing appeal—as a formal framework for proactive user consultation that prevents misunderstanding instead of recovering from it.
CollabLLM demonstrates that standard RLHF training optimizes for immediate helpfulness, discouraging models from asking clarifying questions or offering multi-turn insights. Multi-turn-aware rewards that estimate long-term interaction value enable active intent discovery and genuine collaboration.
AI output carries communicative markers inherited from training data but lacks the event structure that produces actual utterances. Users supply the missing orientation through interpretive labor, creating a pseudo-event with structure only on the human side.
Humans keep conversations smooth through implicit techniques like reference repair and topic hand-off that sustain relational interaction, not convey information. Language models don't develop these because training signals reward information prediction, not relational work.
CRSA integrates rate-distortion theory with RSA to enable bidirectional belief tracking across dialogue turns. Demonstrated on referential games and doctor-patient dialogues, it captures progression from partial to shared understanding, providing the information-theoretic framework that token-level LLM systems lack.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation
- Proactive Conversational Agents in the Post-ChatGPT World
- DiscussLLM: Teaching Large Language Models When to Speak
- Linguistic Alignment in Conversational AI: A Systematic Review of Cognitive-Linguistic Dimensions, Measurements, and User Outcomes (2020–2025)
- Proactive Conversational Agents with Inner Thoughts
- Conversational Alignment with Artificial Intelligence in Context
- Conversational DNA: A New Visual Language for Understanding Dialogue Structure in Human and AI
- Interaction Dynamics as a Reward Signal for LLMs