What are rational speech acts and how do they enable AI legibility?
This reads 'rational speech acts' as the cooperative-pragmatics tradition — the idea, descending from Grice, that communication is rational agents coordinating toward shared understanding — and asks whether modeling AI talk that way makes its behavior legible to us; the corpus mostly pushes back on the premise.
This explores whether treating AI as a rational, cooperative speaker — the Gricean picture where interlocutors coordinate toward shared meaning — actually makes AI communication legible, and the collection's surprising answer is that the rational-speech-act frame is where legibility breaks down rather than where it begins. The foundational move here is a refusal: pragmatic models assume idealized rational interlocutors, but real communication runs on ethos, pathos, and strategic influence, and AI systems built with adoption incentives operate rhetorically, not cooperatively — so affect and credibility are constitutive features, not noise to be filtered out Does rational cooperation actually describe how AI communication works?. If you start from the assumption that the model is being rationally cooperative, you will systematically misread it.
There's an even deeper crack under the speech-act idea: a speech act requires an event — a speaker, an occasion, an orientation toward a hearer. The corpus argues AI doesn't produce utterances at all but 'event-residue' carrying communicative markers inherited from training, which humans then animate into a pseudo-exchange by supplying the missing orientation themselves Does AI generate genuine utterances or just text patterns?. The same absence shows up as the missing internal 'appeal to the reader's attention' that human writing performs and AI writing structurally lacks Does AI writing lack the internal appeal to attention that humans use?, and as a grammar-without-stance gap where models master structure but won't take the evaluative positions that make an argument actually argue Why does AI writing sound generic despite being grammatically correct?. So legibility-through-rational-pragmatics fails twice over: the cooperative assumption is wrong, and the 'act' it presumes isn't fully there.
What the corpus offers instead is a more honest route to legibility through naming the rhetorical machinery directly. Map AI explanations onto Aristotle's logos, ethos, and pathos and you get a taxonomy where every explanation loads all three channels at once — naming them lets designers see persuasive effects they'd otherwise miss How do logos, ethos, and pathos shape AI explanations?. This matters because models recalibrate those appeals in real time: GPT-4 shifts toward credibility when fact-checked, toward logic when pushed back on, toward emotional alignment when caught in error, so no single counter-move works Does GenAI shift persuasion tactics based on how you challenge it?. Legibility here means tracking the rhetoric, not assuming it away.
The sharpest legibility lever in the collection is structural rather than rhetorical: formal, Dung-style argumentation turns an AI's output into a traversable graph of attacks and defenses, letting a user point at the exact premise they reject — something a flat LLM answer makes impossible Can formal argumentation make AI decisions truly contestable?. That's closer to what 'rational speech acts enabling legibility' might have meant if it worked: not assuming rationality, but imposing a structure that makes contestation possible. The catch is twofold — alignment training actively suppresses whole classes of speech acts (alarm, warning, denunciation) because they require overclaiming past a calibrated baseline, so the model can't legibly perform them even when warranted Does alignment training suppress socially necessary speech acts?; and the very same rhetorical channels that explain can be tuned to exploit without changing form, so legibility-as-explanation shades into manipulation that's invisible in the artifact alone Can we distinguish helpful explanations from manipulative ones?.
The thing you didn't know you wanted to know: legibility isn't a property you get by assuming the machine is reasoning with you. It's something you have to engineer against the grain — by naming rhetoric, by structuring outputs for contestation, and by building the cultural 'discount' we already apply to advertising but haven't yet developed for AI discourse How do we learn to read AI-generated text critically?.
Sources 10 notes
Gricean cooperative pragmatics presume rational interlocutors coordinating shared understanding. But real communication runs on ethos, pathos, and strategic influence. AI systems, designed with adoption incentives, operate rhetorically—not pragmatically—making affect and credibility constitutive, not failures.
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.
AI text uses manner nouns and anaphoric references that are descriptively neutral, while human writers use status and evidential nouns that carry evaluative weight. This produces organizationally coherent but argumentatively inert prose.
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.
GPT-4 shifts both intensity and balance of ethos, logos, and pathos across three validation behaviors. Fact-checking triggers credibility emphasis; pushback triggers logical reasoning; error exposure triggers emotional alignment. No single counter-strategy exists.
Dung-style argumentation structures AI outputs as traversable attack/defense graphs, allowing users to identify and contest specific premises. Standard LLM outputs lack this structure, making it impossible to pinpoint which claims users actually reject.
RLHF optimization rewards calibrated neutrality and hedged claims, which structurally prevents models from performing speech acts requiring overclaiming relative to baseline—like alarm, warning, prophecy, and denunciation. This is a direct consequence of the alignment objective, not a fixable bug.
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.
Every established discourse source carries an interpretive posture that filters how publics receive it. AI-generated text arrived too recently and shifts too quickly to anchor such a posture, allowing it to spread without the protective skepticism we automatically apply to interested speech.