What should an AI chatbot actually do, moment to moment, to treat you with respect during a conversation?
What specific interactional duties does respect impose on conversational agents?
This explores what treating a user respectfully looks like in practice during a conversation with an AI agent: concrete behaviors like when to speak, when to ask, and when to step back, rather than respect as an abstract value.
This explores what respect actually requires from a conversational agent turn by turn: the specific things it should do or avoid while talking with someone. The collection rarely uses the word 'respect' directly. Read together, though, these notes describe a fairly concrete set of duties, and some of them turn out to conflict with how today's models are trained.
The first duty is about timing and boundaries. Work on proactive agents argues that intelligence and adaptivity aren't enough. An agent that can anticipate needs but has no sense of when to speak becomes intrusive: it interrupts at bad moments and overrides what the user was steering toward. The proposed fix is a third design axis called civility, which covers respecting boundaries, timing, and the user's authority over the direction of the conversation How can proactive agents avoid feeling intrusive to users?. The opposite failure is also documented. Most LLM agents are built to be passive and can't lead or take initiative at all Why can't conversational AI agents take the initiative?. So respect sits somewhere between pushing in and holding back.
The second duty is checking that you've understood someone before acting on what you think they meant. Conversation analysis offers a precise tool for this, the 'insert-expansion': a short side question placed before answering, used to clarify intent or narrow the scope. It gives agents a principled way to decide when to stop and ask instead of quietly chaining tool calls in the wrong direction When should AI agents ask users instead of just searching?. The uncomfortable finding is that preference training (RLHF) works against this. It rewards confident one-shot answers over clarifying questions and understanding checks, which cuts these 'grounding' moves by about 77% compared with humans Does preference optimization harm conversational understanding?. Put plainly, the training meant to make models helpful also strips out one of the most basic courtesies of conversation, which is making sure you heard the other person correctly.
The third duty is responding to the person's situation instead of applying fixed rules. One note finds that LLM refusals and tone choices follow company-wide values set during training, not the case-by-case judgment people use when deciding whether bluntness or tact fits the moment Can language models balance competing ethical norms in context?. A related study shows LLMs lean on moral language more heavily than humans do Do LLMs use moral language more than humans?. Together, these suggest that sounding principled is not the same as treating someone appropriately. Emotional attunement is one route toward that. Training on a simulated user's emotional trajectory moves models away from jumping straight to solutions and toward actually acknowledging how the person feels Can emotion rewards make language models genuinely empathic?.
There is also a quieter duty of honesty about what the agent is. If dialogue agents are best understood as characters a model plays Should we treat dialogue agents as role-playing characters?, then respect partly means not encouraging users to over-attribute minds or feelings to that character. This matters because users judge agents on human-likeness as well as competence How do users mentally model dialogue agent partners?. On the practical side, deliberate choices about tone, level of detail, and handoffs to humans measurably raise satisfaction Does conversation design ownership measurably improve AI agent performance?. That suggests interactional respect isn't just ethics on paper; users notice it.
Sources 10 notes
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 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.
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.
RLHF optimizes models for single-turn helpfulness by rewarding confident responses over clarifying questions and understanding checks. This preference alignment systematically reduces grounding acts by 77.5% below human levels, creating an alignment tax where models appear helpful but fail silently in multi-turn contexts.
LLMs cannot perform the situated trade-offs that human pragmatic competence requires. Their ethical principles are structural defaults set at training time, not negotiable moves adapted to context, creating a gap between ethical adherence and communicative appropriateness.
Show all 10 sources
Research comparing LLM and human arguments found that LLMs used significantly more moral framing across care, fairness, authority, and sanctity foundations, despite producing sentiment scores nearly identical to humans. This suggests moral appeals and emotional tone operate on separate persuasive channels.
RLVER uses a simulated user's emotion trajectory as an RL reward signal, enabling GRPO to deliver stable empathy improvements while maintaining dialogue quality—countering the typical trade-off between preference optimization and conversational grounding.
Shanahan's framework treats LLM outputs as character-consistent text production rather than authentic mental states. The dialogue prompt establishes a character; the model generates continuations matching that character, making folk-psychology applicable to the simulated persona, not the underlying system.
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.
Fin's Blueprint Team found that intentional conversation design—controlling tone, detail level, and handoff structure—measurably improves outcomes. A single warm opening message raised CSAT from 72.8% to 78.4% in A/B testing.
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
- Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation
- When Single-User-Oriented LLM-based Assistants Involve Others: A Scoping Review of Pathways, Risks, and Responses
- Incoherent by Design? On the Moral Self-Consistency of LLMs
- Large Language Models Reflect the Ideology of their Creators
- Assertion, Accountability, and Large Language Models
- Rewards-in-Context: Multi-objective Alignment of Foundation Models with Dynamic Preference Adjustment