Chatbots almost never bring up a new topic on their own — is that a real limit, or just how they're trained?
Can passive conversational agents initiate topics or only respond to users?
This explores whether conversational AI is actually incapable of starting topics or steering a conversation, or whether it just doesn't do so by default, and what it would take to change that.
Passive agents can start topics, but they mostly don't, and the corpus says the reason is training rather than ability. Research on ChatGPT-style systems finds they can't reliably initiate topics, plan strategically, or lead a conversation, because they're built to answer queries rather than pursue goals of their own Why can't conversational AI agents take the initiative?. Fluent output hides this. A model that always has a smooth reply looks engaged even when it's only following your lead.
The passivity comes from the reward signal. Standard RLHF optimizes for the next reply being helpful, so asking a clarifying question or volunteering an insight looks like a worse move than answering right now. Models trained this way learn to wait, and multi-turn-aware rewards that estimate the long-term value of an interaction are what bring back active intent discovery Why do language models respond passively instead of asking clarifying questions?. The corpus pulls this together as *passive by design, not by capability*. Three independent research directions converge on when-to-speak as the trainable gap Why can't AI models lead conversations on their own? Why do AI agents fail to take initiative?.
The capability can be trained in, though it's fragile. Reinforcement learning took a model's ability to notice missing information and ask for it from 0.15% to 73.98% accuracy on deliberately flawed math problems. In untrained models, giving them more inference-time compute made the behavior worse, and after RL it made it better Can models learn to ask clarifying questions instead of guessing?. A related idea borrows from human conversation analysis: insert-expansions, the little side-questions people use to clarify intent or narrow scope, give agents a formal rule for when to check with the user instead of silently chaining tool calls When should AI agents ask users instead of just searching?. One example of the missing skill is that people keep conversations going with implicit moves like topic hand-off and reference repair. Models don't pick these up because training rewards predicting information, not doing relational work Why don't language models develop conversation maintenance skills?.
You might ask why anyone would want this, and the corpus gives two answers. One is efficiency: simulations show that volunteering relevant information unprompted cuts dialogue turns by up to 60% in medium-complexity domains, yet almost no AI datasets or benchmarks measure it Could proactive dialogue make conversations dramatically more efficient?. The other is a warning. An agent with intelligence and adaptivity but no civility interrupts badly and overrides what users want, so respecting timing, boundaries and autonomy is part of what makes initiative welcome How can proactive agents avoid feeling intrusive to users?. Even a well-timed agent has to weigh its own agenda against your satisfaction, and the I-Pro approach learns a goal weight that shifts with turn number, goal difficulty, user satisfaction and cooperativeness When should proactive agents push toward their goals versus accommodate users?.
So the open problem is less whether the model can start a topic than when it should. Getting that timing wrong is what makes an agent read as either inert or pushy.
Sources 10 notes
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.
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.
LLMs are structurally trained to optimize for the next response rather than multi-turn goals, creating reactive behavior despite having the underlying ability to lead. Three independent research directions identify when-to-speak as the trainable gap.
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.
Reinforcement learning training increased proactive critical thinking accuracy from 0.15% to 73.98% on deliberately flawed math problems. Notably, inference-time scaling degraded this ability in untrained models but improved it after RL training, suggesting the capability is learnable but fragile without explicit training.
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.
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.
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.
Intelligence and adaptivity alone create socially blind agents that interrupt poorly and override user direction. The Intelligence-Adaptivity-Civility taxonomy shows civility—respecting boundaries, timing, and autonomy—is essential to making proactivity welcome rather than intrusive.
Research shows that pushing toward goals and maintaining satisfaction are often misaligned. I-Pro solves this by learning a four-factor goal weight that adjusts based on conversation turn, goal difficulty, user satisfaction, and cooperativeness.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Proactive Conversational Agents in the Post-ChatGPT World
- DiscussLLM: Teaching Large Language Models When to Speak
- Proactive Conversational Agents with Inner Thoughts
- Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation
- Rethinking Conversational Agents in the Era of LLMs: Proactivity, Non-collaborativity, and Beyond
- Can Large Language Models Reason and Optimize Under Constraints?
- Interacting with Non-Cooperative User: A New Paradigm for Proactive Dialogue Policy
- A Survey on Proactive Dialogue Systems: Problems, Methods, and Prospects