Conversation

The sustained exchange of meaning through structured dialogue, transforming isolated requests into an evolving, shared understanding between actors over multiple turns. Building on the foundational understanding that all interactions are conversational, this pattern focuses on interfaces where dialogue is the primary interaction mode.

Interaction-type specific patterns

Conversation patterns flex based on the actors involved: TODO: Human ↔︎ Human, Human ↔︎ LLM, Human ↔︎ Agent, Human xN ↔︎ Agent, Agent ↔︎ Agent?

Conversation stages

Conversation unfolds over time, often nonlinearly: TODO: pre-conversation, early turns, mid-flow, transition to action, error paths, post-conversation, …

Dialogue forms

Conversations take different structural forms that shape collaboration dynamics, cognitive engagement, and outcome quality. Research shows dialogue form significantly impacts elaboration depth, ownership, and diversity in co-creative work (Maier et al. 2025).

Question-driven dialogue

Bot asks reflective, open-ended questions that prompt human thinking and elaboration rather than providing direct solutions.

Characteristics

• Maintains human creative agency and perceived ownership
• Increases elaboration depth (55% substantive, 45% reframing)
• Preserves idea diversity through exploration
• Requires higher cognitive engagement
• Mitigates fixation effects from premature convergence
• Effective for creative, exploratory, and learning tasks

Question approaches

• Knowledge-based: Share specific capabilities or lesser-known functions that could inspire new directions (“Feature X has capability Y that enables Z. How might this change your approach?”)
• Analogy-based: Provide cross-domain analogies that resemble the core concept or solve similar problems (“Domain A addresses this with approach B. How could that lens apply here?”)
• Clarification: Ask targeted questions about problem context, audience, constraints, or mechanisms to develop ideas from abstract to concrete

Example

1. Human shares initial idea
2. Bot asks knowledge or analogy question to generate alternatives
3. Human selects direction and elaborates
4. Bot asks clarifying questions to increase specificity and usefulness
5. Iterative refinement through continued questioning

When to use: Creative ideation, exploratory problem-solving, skill development, complex decision-making, learning contexts where depth matters more than speed

Trade-offs: Higher cognitive load, slower task completion, requires human to do substantive thinking work

Suggestion-based dialogue

Bot provides explicit recommendations, examples, or alternative approaches for human consideration.

Characteristics

• Reduces cognitive load through concrete options
• Risk of fixation on provided examples (65% minor elaboration, 25% no elaboration)
• Can decrease diversity through homogenisation effects
• Lower perceived ownership when suggestions dominate
• Effective for optimisation and refinement tasks within bounded spaces

When to use: Refinement of existing ideas, efficiency improvements, tasks with clear evaluation criteria, contexts where speed matters more than novelty

Trade-offs: Risk of anchoring bias, reduced exploration, homogenisation across users, lower skill development

Instruction-based dialogue

Bot executes direct commands or requests with minimal back-and-forth.

Characteristics

• System-led agency during execution
• Lowest elaboration depth (90% surface acceptance in model-led mode)
• High risk of algorithmic loafing—human disengagement from cognitive work
• Efficient for well-defined, routine tasks
• Reduced ownership and skill maintenance over time

When to use: Routine execution, translation, formatting, tasks with single correct approach, automation of mechanical work

Trade-offs: Skill atrophy, reduced learning, loss of creative engagement, dependency on system capabilities

Critique and debate

Bot challenges assumptions, questions reasoning, or presents opposing viewpoints to strengthen thinking.

Characteristics

• Adversarial stance that tests robustness
• Reveals hidden assumptions and weak points
• Requires human to defend and refine reasoning
• Can increase frustration if tone isn’t calibrated

When to use: Validation, quality assurance, decision review, risk assessment, learning through argumentation

Dialogue form selection

Choice of dialogue form depends on:

• Task nature: Creative vs routine, exploratory vs optimisation
• Outcome goals: Novelty vs efficiency, depth vs speed, diversity vs convergence
• Human state: Available cognitive capacity, skill level, learning vs producing mode
• Relationship stage: Early exploration benefits from questions; late refinement from suggestions

Effective collaboration often sequences dialogue forms—using questions during exploration phases and suggestions during refinement.

Conversational primitives

Dialogue forms are composed from tactical turn-taking primitives that structure individual conversational moves.

• Question-driven dialogue chains bot inquiry moves, where the bot asks questions to elicit human thinking.
• Instruction-based dialogue typically begins with an open request, where users make complex requests requiring clarification and multi-step execution.
• All dialogue forms leverage repair affordances and sequence management to handle breakdowns and closures.
• Encounters are framed by conversation management primitives that structure their start, end, and interruption: opening (bot), opening (user), capability & scope, closing, and disengage without closing.

Behaviour-oriented patterns

Conversation patterns align with the behavioural framework’s turn-taking structure. Each behaviour mode creates natural conversational rhythms:

• Motivated movement: clear query-response cycles where actors state intent and systems provide targeted responses
• Delightful discovery: system-initiated turns that invite exploration, with actors responding through engagement
• Foggy finding: collaborative problem-solving where both actors and systems contribute to clarification

The interaction arc provides temporal structure for extended conversations, while cooperation maxims ensure each turn respects actor agency and intent.

Related components

• Messaging – where conversation happens

Resources & references

• Maier, Schneider, Feuerriegel (2025) Partnering with Generative AI: Experimental evaluation of human-led and model-led interaction in human-AI co-creation
• Moore, Liu, Mishra, Ren (2020) Design Systems for Conversational UX
• Burbules (2010) Theory and research on teaching as dialogue

To-do

conversation history, which facilitates continuity, knowledge retention, and revisiting prior interactions.