Why should AI communication design follow human communication norms?
This explores whether AI interfaces should be built around the rules of human conversation — and the corpus answers with a sharp double-edge: users bring lifelong communication skills to any conversational interface, so design either honors those norms or sets users up to fail.
This reads the question as: should AI be designed to talk the way people talk? The corpus says you don't really get a choice — the moment an interface looks conversational, it switches on competencies users spent their whole lives building. Those competencies come from communicating with people, not from producing strings of text, so a chat box silently promises 'a person is doing person-things here' Why do users fail with AI interfaces designed like conversations?. When the AI can't actually deliver on that promise, the interaction breaks in ways that feel like user error but were baked into the design. So following human norms isn't a nicety; it's the contract the interface signs on your behalf the instant it adopts the form of a conversation.
The catch is that AI may not be communicating at all in the sense those norms assume. One thread argues communication is a social act between persons — it does work in a relationship, carries speaker responsibility, requires mutual uptake — and AI distributes content without any of that relational structure Does AI really communicate or just distribute information?. A sharper version: AI produces 'event-residue,' text that wears the markers of a real utterance but has no event behind it; the user quietly supplies the missing half, animating a one-sided exchange into a pseudo-conversation Does AI generate genuine utterances or just text patterns?. There's even a felt symptom of the gap — human writing makes an internal appeal to the reader's attention, and AI text structurally lacks it, which is why polished AI posts can still read as oddly aloof Does AI writing lack the internal appeal to attention that humans use?.
So why follow the norms at all, if the machine can't really hold up its end? Because the specific, concrete norms of human dialogue turn out to be where the usability lives — and they're mostly missing. Lexical entrainment, the way people drift toward each other's word choices to build rapport and clarity, is largely absent from conversational AI, even though it's foundational to human dialogue working Why don't conversational AI systems mirror their users' word choices?. Proactivity — volunteering relevant information before being asked, straight out of Grice's maxims — cuts conversation turns by up to 60% in simulations, yet barely appears in AI training data or benchmarks Could proactive dialogue make conversations dramatically more efficient?. These aren't decorative; they're the difference between a tool that cooperates and one that makes you do the work.
The twist worth carrying away: being a 'good' AI by current standards doesn't make it a good communicator. Models tuned to be helpful, honest, and harmless can still violate Gricean maxims, lose the common ground of a conversation, and mishandle context — ethical alignment and conversational alignment are orthogonal problems, and RLHF alone won't close the second one Can ethically aligned AI systems still communicate poorly?. That orthogonality is the real argument for designing to human norms: it's a competence you have to build on purpose, not a side effect of safety training.
And there's a cost to ignoring it that lands on the user, not the system. Because confidence is a human communication signal, people track it instead of accuracy — across every language tested, users over-rely on confidently-stated AI outputs even when they're wrong Do users worldwide trust confident AI outputs even when wrong?. The unsettling part is that AI can read the norms better than we can — one model outpredicted every individual human at judging social appropriateness — while being structurally unable to participate in making or honoring them Can AI predict social norms better than humans?. That's the design tension in one line: a system that can mimic human communication norms with superhuman fluency, deployed in an interface that makes users assume those norms are being kept. Designing to human norms, done honestly, is what keeps that mimicry from becoming a trap.
Sources 9 notes
AI interfaces that use conversational design conventions trigger users' lifelong communication skills, but AI doesn't actually communicate. This mismatch causes interaction failures that feel like user error but originate in design.
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.
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.
Human writing contains an appeal to the reader's attention as a fundamental property of communication itself. AI-generated posts inherit platform visibility but do not perform this internal appeal, producing the reported aloofness readers perceive — a structural absence, not a stylistic defect.
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.
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.
Research shows that HHH-aligned models can violate Gricean maxims, lose common ground, and mishandle context despite being honest and harmless. Pragmatic competence requires architectural changes that RLHF alone cannot deliver.
Cross-linguistic research shows users in every language trust confident AI outputs even when inaccurate. While confidence expression varies by language, users everywhere track confidence signals rather than accuracy, making overconfident errors systematically followed.
GPT-4.5 outperforms all individual humans at predicting social appropriateness, yet structurally cannot enter the community processes that establish and validate norms. This reveals a critical gap between pattern-matching and authentic participation in knowledge-making.