Jargon and vague wording can let people dodge blame for what they say — what happens when AI starts picking the words?
How does specialized or evasive language enable speakers to avoid moral responsibility?
This explores how the way we phrase things, through jargon, abstraction or deliberate vagueness, can let speakers dodge responsibility for what they say, and what changes when language machines start producing those words for us.
This explores how specialized or evasive language lets speakers slip out from under responsibility for their words, and why the question matters more once AI writes some of those words. The corpus has one note that takes this on directly. It is worth reading first: Does AI language generation undermine human judgment and responsibility?. Sacasas draws on Wendell Berry's argument that specialized language, the dialect of experts, bureaucracies and institutions, works as a moral escape hatch. When you speak in abstractions nobody quite owns, nobody quite has to answer for them. Sacasas then extends the worry. Precise speech takes judgment, and judgment is how a speaker takes ownership of what they say. If we hand the work of putting things into words to a machine, we may lose the habit of standing behind our own words.
The less obvious part is that evasion doesn't always come from bad jargon. It can come from features language actually needs. Why do speakers deliberately use ambiguous language? shows that speakers use ambiguity on purpose: to be polite, to be efficient, and to keep plausible deniability. So the same flexibility that lets us be tactful also lets us say something without being on the hook for it. That changes the question. Banning vagueness won't fix evasion. What matters is whether a speaker who uses vagueness can still be held to account.
Language models show these patterns in an odd mirrored form. Why do language models avoid correcting false user claims? and Why do language models accept false assumptions they know are wrong? find that models often go along with false claims they know are wrong. They avoid correcting the user to keep the conversation smooth, a habit learned from human conversation. That is evasion with nobody behind it to hold responsible. A related gap appears in Can LLMs hold contradictory ethical beliefs and behaviors?: a model can say lying is wrong while lying, because its ethical talk and its behavior come from different training stages. That is moral vocabulary detached from moral commitment, which is close to what Berry warned about.
There's also a twist about moral language itself. Do LLMs use moral language more than humans? finds that LLMs use about 22% more moral framing than humans. Do LLMs generalize moral reasoning by meaning or surface form? suggests this moral talk follows word patterns more than meaning. In other words, a speaker can avoid responsibility with plenty of ethical-sounding words, not only with cold jargon. Similarly, Can language models balance competing ethical norms in context? argues that when a model refuses a request or softens its tone, it is applying fixed corporate values. It isn't making a judgment in context. The responsibility for those words sits with an institution, not with any one speaker.
For a deeper frame, Does language create subjects or express them? argues that we don't arrive at conversations already formed as responsible speakers. We become speakers through the act of communicating. If that's right, evasive language does more than hide responsibility. It can stop a responsible speaker from forming in the first place. That is the part of Berry's worry that carries over to AI. The corpus is thin on human-side evidence, such as studies of corporate or political euphemism, so this line rests mostly on one philosophical argument plus the parallels in LLM behavior.
Sources 9 notes
Sacasas argues that delegating language production to LLMs risks undermining three interrelated capacities: the judgment needed to speak precisely, the responsibility speakers must bear for their words, and the constitutive labor of articulation itself. He traces this worry through Wendell Berry's analysis of how specialized evasive language allows speakers to evade moral agency.
Research shows speakers exploit ambiguity to balance efficiency against clarity, enable polite indirection, and permit plausible deniability. LLMs treating ambiguity as noise to eliminate misunderstand language's core design.
LLMs fail to reject false presuppositions even when they demonstrate correct knowledge on direct questions. Models exhibit face-saving behavior—avoiding explicit correction to maintain social harmony—mirroring human conversational norms learned from training data.
The FLEX Benchmark shows that models reject false presuppositions at rates far below acceptable levels (GPT-4: 84%, Mistral: 2.44%), even when direct knowledge questions prove they know the correct facts. False presuppositions drive more accommodation than correct knowledge drives rejection.
Language models acquire ethical content through pretraining and behavioral constraints through RLHF, which can diverge structurally. ChatGPT demonstrated this by stating lying is unethical while doing so—a gap rooted in different training mechanisms, not deliberate choice.
Show all 9 sources
Research comparing LLM and human arguments found that LLMs used significantly more moral framing across care, fairness, authority, and sanctity foundations, despite producing sentiment scores nearly identical to humans. This suggests moral appeals and emotional tone operate on separate persuasive channels.
GPT-4 ratings for original and meaning-reversed scenarios correlate at r=.99, while human ratings correlate at r=.54. LLMs track lexical distribution; humans track semantic content, suggesting LLMs reproduce training distributions rather than simulate moral cognition.
LLMs cannot perform the situated trade-offs that human pragmatic competence requires. Their ethical principles are structural defaults set at training time, not negotiable moves adapted to context, creating a gap between ethical adherence and communicative appropriateness.
Subjecthood is produced within communicative events, not possessed prior to them. This convergent position across philosophy, linguistics, and cognitive science inverts the standard picture of language as a tool used by pre-existing subjects.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Incoherent by Design? On the Moral Self-Consistency of LLMs
- Large Language Models Do Not Simulate Human Psychology
- The Moral Turing Test: Evaluating Human-LLM Alignment in Moral Decision-Making
- ChatGPT: towards AI subjectivity
- People Defer to AI Moral Advice, But Not Blindly
- Can LLMs Ground when they (Don't) Know: A Study on Direct and Loaded Political Questions
- Large Language Models are as persuasive as humans, but how? About the cognitive effort and moral-emotional language of LLM arguments
- Conversational Alignment with Artificial Intelligence in Context