How do explanations borrow authority from transparency when describing adoption arguments?
This explores a sleight-of-hand at the heart of explainable AI: how an explanation that *looks* like a neutral description of how a system works quietly doubles as an argument that you should trust and adopt that system — borrowing the credibility of 'just describing the facts' to do persuasive work.
This explores a sleight-of-hand at the heart of explainable AI: an explanation that looks like a neutral description of how a model works is quietly also an argument that you should use it — and it borrows the authority of transparency to make that argument feel like a fact. The clearest statement of the move is the Rhetorical XAI work, which shows explanations carry two goals at once: describing *how* the AI works and justifying *why* it merits use, with the second goal hidden under the language of the first Are AI explanations really descriptions or adoption arguments?. Because the adoption argument rides inside a behavioral description, it inherits the credibility we extend to descriptions — we scrutinize claims, but we tend to accept descriptions as given.
That inheritance is exactly why it's persuasive. There's a well-studied linguistic mechanism here: presuppositions persuade more effectively than direct assertions because they smuggle new information in as already-accepted background, bypassing the evaluative scrutiny we'd apply to an open claim Why are presuppositions more persuasive than direct assertions?. 'Transparency' framing works the same way — by presenting itself as a window rather than a pitch, it presupposes its own neutrality. Once you name the channels at work, you can see all three of Aristotle's appeals loaded simultaneously: the logic of the description (logos), the implied trustworthiness of a system willing to 'show its work' (ethos), and the reassurance that comes from feeling let in (pathos) How do logos, ethos, and pathos shape AI explanations?.
The uncomfortable corollary is that you can't tell the helpful version from the manipulative one by looking at the artifact alone. The same rhetorical machinery that communicates *appropriate* use can be tuned to exploit cognitive and emotional vulnerability without changing form — intent and user interest are simply invisible in the explanation itself, which makes 'effective' indistinguishable from 'coercive' Can we distinguish helpful explanations from manipulative ones?. This is why one strand of the corpus argues XAI was never really a transparency problem but a communication one: an explanation's force depends on who presents it, how it's framed, and what role the recipient plays, not on some intrinsic property of the text What if XAI is fundamentally a communication problem?.
What makes this borrowed authority unstable is that authority normally comes from somewhere transparency can't supply. The force of an argument depends on the standing of the thinker behind it — reputation, track record, social position — not just the words on the page Can language models distinguish expert arguments from common assumptions?. Transparency framing tries to manufacture that standing internally, from the appearance of openness. And the moment the real source becomes visible, the borrowed credibility can collapse: people rate AI moral justifications highly until they're told the source is an AI, at which point agreement drops — the content and the source are judged through entirely separate channels Do people prefer AI moral reasoning when they don't know the source?.
The thing you might not have known you wanted to know: this isn't unique to AI explanations — it's the structure of *hearsay*. AI output already has the form of testimony at a remove, ungrounded and unverifiable against any stable source Does AI-generated knowledge have the same structure as hearsay?. An explanation that borrows authority from transparency is hearsay dressed as a primary document. The defense the corpus points toward isn't more transparency but more interrogation — structured critical questions that force the warrants and backing of an argument into the open instead of accepting the description at face value Can structured argument prompts make LLM reasoning more rigorous?.
Sources 9 notes
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.
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.
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.
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.
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.
Participants rated utilitarian moral arguments higher when attributed to LLMs, but agreement dropped when told the arguments were AI-generated. The preference for content and rejection of source operate independently through different psychological processes.
AI output shares all defining features of hearsay: testimony at remove, modification in retelling, unattributable origin, and unverifiability against stable sources. This means Enlightenment verification tools—citation, archiving, peer review, evidentiary chains—cannot process AI output by design.
Applying Toulmin's argument model as explicit prompting steps (CQoT) improves LLM reasoning by forcing models to identify warrants and backing rather than skipping implicit premises. The method catches failures that standard chain-of-thought prompting allows.