How do organizational roles and peer interpretations shape what an explanation means?
This explores where an explanation's meaning actually gets decided — not inside the explanation itself, but in the social layer around it: the roles people occupy and how peers read and re-read what was said.
This explores where an explanation's meaning actually gets decided — and the corpus pushes hard against the intuition that it lives inside the explanation. The strongest claim here is that meaning is constituted socially, not delivered: drawing on Luhmann's layered cybernetics, one line of work argues explanation meaning emerges from "N-order interpretation" — observations of observations rippling through a group — rather than from any one human-AI exchange Where does the meaning of an AI explanation actually come from?. The practical sting is that a lab-tested explanation, stripped of its social context, won't predict whether it actually works in the wild.
If meaning is social, then the same words land differently depending on who's speaking and who's listening. One framing treats explainability as a communication problem, not a transparency problem: quality isn't intrinsic to the artifact but emerges from a source-framing-recipient triad — who presents it, how it's framed, what role the recipient occupies What if XAI is fundamentally a communication problem?. This is reinforced from the reader's side: studies of interpretation show that disagreement about the same sentence across different social positions is valid signal, not annotation noise Why do readers interpret the same sentence so differently?. The peer's standpoint is part of the meaning, not an error to be averaged away.
Role and authority do real work here too. An argument's force depends on the standing of the thinker, not just the words — reputation, track record, and institutional position carry credibility that text alone can't reconstruct, which is exactly what language models lose when they process only the words and not the social world that built the expertise Can language models distinguish expert arguments from common assumptions? Why do AI systems fail at social and cultural interpretation?. So "what an explanation means" partly equals "who is allowed to mean it." The same content from a junior peer and a domain authority is, functionally, two different explanations.
There's also a quieter mechanism worth knowing: explanations don't just describe, they argue. Rhetorical-XAI work shows explanations smuggle in adoption arguments under transparency language Are AI explanations really descriptions or adoption arguments?, operating through Aristotle's logos, ethos, and pathos all at once How do logos, ethos, and pathos shape AI explanations?. Once you see that, the line between explaining and manipulating gets thin — the same channels that build appropriate trust can be tuned into dark patterns without changing the words Can we distinguish helpful explanations from manipulative ones?. Presuppositions show the trick at the sentence level: phrasing a claim as already-accepted background slips it past scrutiny in a way an open assertion can't Why are presuppositions more persuasive than direct assertions?.
The thread that might surprise you: explanation is co-constructed, turn by turn, not handed down. Analysis of hundreds of everyday explanations finds that understanding succeeds or fails based on interacting dialogue dimensions — topic relation, dialogue act, explanation move — which is precisely the interactional fabric current LLMs skip when they generate explanations monologically What makes explanations work in real conversation?. Put together, the corpus reframes the whole question: an explanation isn't a thing you possess and transmit, it's an outcome negotiated among roles and peers — which is exactly why systems that treat it as a one-shot artifact keep measuring the wrong thing.
Sources 10 notes
Drawing on Luhmann's multi-layer cybernetics, AI explanation meaning is constituted at the social-group level through layered observations of observations, not produced inside dyadic human-AI dialogue. Lab-tested explanations stripped of social context will not predict real-world effectiveness.
Explanation quality is not intrinsic to the explanation itself but depends on the rhetorical situation: who presents it, how it is framed, and what role the recipient plays. Evaluations that ignore this triad measure only a narrow slice of real-world effectiveness.
Interpretation Modeling research shows that disagreement on socially embedded sentences reflects valid differences in reader perspective, not annotation failure. Structured human disagreement in NLI benchmarks confirms that interpretation distributions carry meaningful information.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
LLMs achieve 100th-percentile performance on norm prediction yet regress on theory-of-mind tasks and cannot generate culturally-resonant interpretations. The pattern shows that statistical competence coexists with absence of actual social understanding and participation.
The Rhetorical XAI paper shows that explanations serve dual purposes: describing how AI works and justifying why it should be used. This rhetorical work has been hidden under transparency language, allowing adoption arguments to inherit credibility from behavioral descriptions.
Aristotle's three appeals map onto explanation design across two goals (how AI works, why AI merits use), creating a 3×2 space where every explanation loads all three channels simultaneously. Naming these rhetorical channels lets designers account for unintended persuasive effects.
The same logos, ethos, and pathos that communicate appropriate AI use can be tuned to exploit cognitive and emotional vulnerability without changing form. Intent and user interest are invisible in the artifact alone, making effectiveness metrics indistinguishable from coercion.
Experimental evidence shows presuppositions with additive, iterative, and factive triggers persuade audiences more than assertions, especially for discourse-new content. The mechanism: presuppositions bypass evaluative scrutiny by presenting claims as already-accepted background.
Analysis of 399 daily-life explanations shows that topic relation, dialogue act, and explanation move jointly predict understanding success. Explanations are co-constructed through interaction patterns, not monological delivery—challenging how LLMs currently generate explanations.