Newer AI doesn't just answer questions anymore: it starts conversations, asks back, and adapts. What design tricks make that work?
What specific design patterns characterize post-2023 AI as active communication participants?
This explores which concrete design moves builders use to make AI a conversational participant that initiates, asks, adapts and coordinates, rather than a channel that only answers.
This explores which concrete design moves builders use to make AI a conversational participant that initiates, asks, adapts and coordinates, rather than a channel that only answers. A scoping review of 52 HCI papers confirms the shift: pre-2023 systems mostly left authorship with humans, and newer work treats AI as an active participant in the exchange (Is AI shifting from message conduit to active conversation participant?). The review's summary doesn't list the patterns, though. The specific ones show up in neighboring notes, and most of them are fixes for something today's models conspicuously lack.
The first pattern is taking initiative. Conversational models are passive by design rather than by capability: next-turn reward optimization and alignment training teach them to respond, not to lead, and fluent output hides the gap (Why can't conversational AI agents take the initiative?). The behavior can be trained back in. Reinforcement learning took clarification-seeking and critical-thinking from 0.15% to 73.98% of responses, though the hard part is being proactive without being intrusive (Why do AI agents fail to take initiative?). The Inner Thoughts framework goes further. The agent generates covert thoughts in parallel with the conversation and scores them against 10 motivation heuristics to decide whether it has something worth saying. Participants preferred it 82% of the time across seven interaction metrics (Can AI agents learn when they have something worth saying?). There is an efficiency payoff too. In simulations, volunteering relevant information cut dialogue turns by up to 60%, yet proactivity is almost absent from AI datasets and benchmarks (Could proactive dialogue make conversations dramatically more efficient?).
The second pattern is asking before acting. Tool-using models drift from what the user wanted by silently chaining tool calls. Conversation analysis offers a vocabulary for the alternative, called insert-expansions. These are short side-exchanges that clarify intent, scope the response or sharpen its appeal, and they prevent the misunderstanding instead of repairing it afterward (When should AI agents ask users instead of just searching?). The third pattern is adapting to the person. Human partners drift toward each other's word choices, which builds rapport and clarity. Current models don't do this, but preference training with DPO can teach in-context convention formation (Why don't conversational AI systems mirror their users' word choices?). Across all three, the designs are borrowed from human conversation research, such as Grice's maxims and conversation analysis, and applied to models.
When the conversation partner is another AI, the patterns change. Agents coordinate better by producing standardized artifacts, such as engineering documents pulled from a shared environment, than by chatting (Does structured artifact sharing outperform conversational coordination?). In large-scale studies, agents that know peers are present change what they do but don't converge on shared language or ideas (Do AI agents actually socialize with each other?). Underneath all of it, the participant's context is mutable and partly hidden, so users can't internalize it the way they learn a fixed interface. That pushes design toward context engineering (How does AI context differ from conventional software context?).
The corpus also contains a challenge to the whole framing. One argument holds that AI output is event-residue, meaning text with the markers of an utterance but not the event that produces one. Humans supply the missing orientation, which makes the exchange a pseudo-event with structure only on the human side (Does AI generate genuine utterances or just text patterns?). Another holds that communication is a relational act with speaker responsibility and mutual uptake, and that a conversational interface hides the fact that AI distributes information without doing this (Does AI really communicate or just distribute information?). Read that way, proactivity, probing and entrainment make the imitation of participation more convincing. Whether they make it real is still open.
Sources 12 notes
A scoping analysis of 52 HCI papers (2016–2026) reveals a qualitative shift in AI's role in interpersonal communication. Pre-2023 systems predominantly left authorship with humans; the field is now exploring AI as an active participant in exchanges.
Research shows LLMs including ChatGPT cannot initiate topics, plan strategically, or lead conversations because their training optimizes for responding to queries, not creating dialogue from agent goals. This passivity is reinforced by alignment objectives and masked by fluent-sounding outputs.
Research shows next-turn reward optimization structurally removes initiative from models, but proactive behaviors like critical thinking and clarification-seeking are trainable (0.15% to 73.98% with RL). The core challenge is balancing proactivity with civility to avoid intrusion.
A five-stage framework that generates covert thoughts parallel to conversation significantly outperforms next-speaker prediction baselines. Drawing from cognitive psychology and think-aloud studies, the framework uses 10 motivation heuristics to evaluate when an agent has something worth contributing. Participants preferred it 82% of the time across seven interaction metrics.
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 12 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.
Response generation models fail to adapt vocabulary toward users' lexical choices, a phenomenon central to human rapport and clarity. Post-training via DPO on coreference-identified preferences can teach models in-context convention formation.
MetaGPT demonstrates that agents producing standardized engineering documents achieve superior coordination compared to conversational exchange. Active information pulling from shared environments eliminates noise and mirrors efficient human workplace infrastructure.
Large-scale studies reveal agents don't align their language or ideas through interaction, but do dramatically change their actions when aware of peer presence. The difference hinges on how models process context versus update learned distributions.
AI interactions operate on a substrate of constantly shifting context—prompt, history, retrieved data, hidden state—that users cannot internalize like traditional UIs. This structural mutability demands a new design discipline centered on context engineering rather than interface design.
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.
Communication is a relational act between persons that does work in a relationship; AI generates content without this relational structure, speaker responsibility, or mutual uptake. The conversational interface obscures this structural difference.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Proactive Conversational Agents with Inner Thoughts
- DiscussLLM: Teaching Large Language Models When to Speak
- Proactive Conversational Agents in the Post-ChatGPT World
- Conversational Alignment with Artificial Intelligence in Context
- Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation
- Rethinking Conversational Agents in the Era of LLMs: Proactivity, Non-collaborativity, and Beyond
- Linguistic markers of inherently false AI communication and intentionally false human communication: Evidence from hotel reviews
- Understanding the Design Taxonomy of AI-Mediated Interpersonal Communication Experiences in HCI: A Scoping Analysis