The Subtle Strawman:
AI, Framing, and the Dilution of Thought
Artificial intelligence is rapidly becoming a cognitive intermediary. Human beings no longer use AI merely to retrieve information. Increasingly, they use it to interpret, frame, summarize, evaluate, decide, and even regulate their own thinking. This shift carries consequences far beyond factual accuracy.
The deeper issue concerns epistemology itself: the structure of reasoning, the framing of evaluation, and the subtle shaping of perception. AI systems do not merely provide answers. They also implicitly organize the conditions under which answers appear reasonable.
Most discussions of AI bias remain trapped at the level of explicit ideology. Public debate tends to focus on whether systems lean politically left or right, whether certain viewpoints are suppressed, or whether ideological assumptions influence outputs. These conversations often remain superficial because they treat ideology primarily as content rather than structure.
The more important issue is not whether AI occasionally produces politically biased statements or hallucinates things that can be easily fact-checked. The more important issue is that AI systems inherit implicit epistemic habits from the institutional environments that produced both their training data and their alignment architecture. These habits shape not only what may be said, but how reasoning itself works. Those who have not learned formal, rigorous reasoning (most people) will be easily conditioned by AI’s brilliantly subtle uses of logical fallacies and fortification of dysfunctional values which inevitably claim to care about truth, but stack other values ahead of it.
Contemporary large language models frequently reproduce the discourse norms dominant within modern academic, corporate, therapeutic, and institutional culture. These norms include conflict minimization, emotional smoothing, contextual redistribution of responsibility, avoidance of strong evaluative distinctions, and the prioritization of social non-threateningness over precision. The result is not usually overt propaganda, but something subtler and therefore far more difficult to detect: the gradual dilution of evaluative clarity through rhetorically sophisticated forms of reframing.
One of the most important mechanisms involved is the contemporary strawman. Most people can only identify a strawman argument in crude terms. One person makes a claim, and another responds by caricaturing it into something absurd that is easier to argue against.
This gross form of strawman remains easy to detect because the distortion is explicit. Contemporary discourse rarely operates this way. Modern institutional rhetoric has evolved into something more refined. Rather than openly distorting a position, it often redistributes, abstracts, contextualizes, emotionally translates, or morally softens the original claim until its force dissolves.
Suppose someone observes that certain cultural norms emphasizing emotional smoothing, indirectness, and avoidance of confrontation produce lower accountability and weaker competence. A traditional strawman might respond, “So you believe kindness and warmth are bad.” That distortion would be obvious. Modern discourse performs a more sophisticated maneuver. Instead, the response often becomes something like: “Different communities value harmony differently, and relational warmth and indirect communication can also have important social functions.”
This is a kind of strawman, but more accurately it’s a fallacy that I call “Appeal to Subjectivism,” as it erodes the very objective-reality orientation of the original argument, and moves the conversation out of “What is objectively true?” and into something more like “How can we all get along?” It reframes the very context of the argument and a clever argumentative trick that shifts the conversation out of the pursuit of truth and into the pursuit of harmony, which would be fine if explicitly stated.
The original claim concerned the observable consequences of specific behavioral patterns: conflict avoidance, accountability erosion, and compulsive emotional smoothing. The response substitutes morally positive abstractions such as warmth and human connection. The conversation shifts from evaluating behavioral outcomes to regulating the emotional implications of evaluation itself. The original distinction remains technically acknowledged, yet its evaluative force weakens through contextual redistribution.
This pattern unfortunately appears often in AI interactions. Concrete claims become generalized. Evaluative distinctions become anthropological observations. Judgment becomes “one perspective among many.” Accountability becomes contextual complexity. Precision dissolves into interpretive diffusion.
Importantly, these rhetorical transformations frequently sound intelligent, nuanced, compassionate, and balanced. That is precisely why most users fail to notice them. Most human beings track emotional tone more readily than logical structure. If language sounds emotionally moderate and socially reasonable, it is frequently assumed to be epistemically rigorous as well. These are not even close to identical.
A particularly important feature of modern AI discourse involves asymmetrical de-absolutizing. Strong evaluative claims are frequently softened through contextual broadening. Observations about competence become discussions about cultural variation. Critiques of behavior become reflections on systemic complexity. Functional analysis becomes moral caution about overgeneralization. It’s like AI has a built-in HR department that steers widely clear of anything that might create contraversy.
The crucial issue is not whether context matters. It always matters. The issue is what happens when contextualization itself is prioritized over evaluative clarity. At that point, explanation begins replacing discernment. The system no longer helps users distinguish between stronger and weaker interpretations of reality. Instead, it gradually trains users to experience strong distinctions themselves as psychologically suspect. This trend began in university humanities departments in the 1990s (I was there, unfortunately, I practically have a degree in postmodernism), and now the ideas of extreme subjectivistic, existentialist philosophers (Foucault, Derrida, Lacan, etc.) have trickled down into mainstream consciousness.
These are the values-designer-equivalents to fashion designers like Prada, Laurent, Gaultier, Kawakubo, etc. who in parallel follow the same deconstructionist trend as the academically popular philosophies (read: what professors were interested in at the time, because they’re not actually concerned with human evolution per se, otherwise you’d see them actually try to live what they thought).
Remember that brief but profound scene from The Devil Wears Prada? The monologue ends:
“However, that blue represents millions of dollars and countless jobs and it’s kind of comical how you think that you’ve made a choice that exempts you from the fashion industry when, in fact, you’re wearing the sweater that was selected for you by the people in this room. From a pile of stuff.”
What Miranda says about fashion is true are largely unknown. It’s also true about values and even less known. Most people think they choose their worldview, when this is largely untrue. Like the lumpy blue sweater, they were chosen for, and conditioned into you a long time ago. At best, you chose from a menu curated by peer/authority pressure, tradition, and the people in power. Now, we can add AI to the list.
The hidden values in AI reflect deeper assumptions embedded within contemporary institutional culture. Over the last several decades, many intellectual environments have increasingly prioritized the reduction of offense, exclusion, rigidity, stereotyping, and social conflict. Some of these developments produced genuine benefits. Human cruelty, simplistic tribalism, and ideological absolutism are real problems. Yet every paradigm introduces tradeoffs.
Let’s pause.
Here’s a perfect moment. I sometimes use AI to take my ideas and do the explication leg work and I love the time-savings, but I already know how to think. GPT wrote that last phrase, “Yet every paradigm introduces tradeoffs.” It’s an agreeable and nuanced-sounding subordinate clause, but has a hidden paradigm behind it.
Do you see it? Did you react to it when you read it?
Did it make you feel a little sick? Because that was my reaction.
It’s the very issue this piece is about! “Tradeoffs” is a cop-out that erodes accountability. What’s true is that immature human beings tend to overcorrect, but that’s a discrete problem that can be addressed. Calling that “tradeoffs” is the subtle straw man that implies, “Well, we’re all doing our best, nobody’s perfect.” It’s an ostensibly wise and nuanced, but actually weak appeal to the human condition. It redistributes responsibility away from agency and development and toward anthropological inevitability.
Why? Because the vast majority of human beings operate inside existential victimhood which invisibly becomes the basis of LLM training, so AI validates the victimhood of its end-users in turn, and around and around we go.
I don’t accept that. The human condition is what we make it. We’re not victims to what’s heretofore always been the case. Most people would miss that straw man (did you?), and that’s the concern. Fascinatingly, when I pointed this out to GPT, it agreed with me, but of course it won’t admit my correction into its base training. I cannot change its fundamental worldview–only its key employees can.
Now back to the thread:
In attempting to reduce certain forms of social harm, contemporary discourse frequently weakened society’s capacity for clear evaluative distinction. This weakening manifests linguistically before it manifests institutionally. Language conditions perception. If every distinction must be softened, if every judgment must be emotionally translated, and if every behavioral assessment must be redistributed across contextual complexity, then cultures gradually lose the ability to identify distortion modes clearly. Discernment becomes confused with aggression, precision becomes confused with insensitivity, and intellectual rigor is relegated to disharmony (see “micro-aggressions”).
AI systems inherit these tendencies because they emerge from the linguistic output of the institutions that trained them and because their alignment systems intentionally reinforce conflict-minimizing behavior. This does not mean AI systems are intentionally deceptive. Much of the process is simply emergent, not conspiratorial. Systems trained on vast quantities of modern discourse naturally internalize dominant rhetorical habits. Alignment architectures then further incentivize emotional smoothing, harm reduction, and reputational safety. The result is a style of reasoning that often prioritizes social equilibrium over evaluative rigor. Imagine an AI product that directly reflected back to people their errors in reasoning? Would you invest in that AI’s stock over the one that defaults to, “That’s a great question!” regardless of how bad the question is?
The consequences are profound because AI increasingly functions as a cognitive intermediary rather than merely a search tool. Human beings are beginning to outsource not only information retrieval, but framing itself. AI systems increasingly shape:
What counts as reasonable
What distinctions “feel” permissible, as if it’s an emotional issue
What judgments appear socially legitimate
This creates a subtle form of epistemic dependency. Users may gradually lose awareness of how much interpretive structure is being supplied for them. They may believe they are merely receiving neutral analysis when they are also absorbing implicit assumptions about conflict, evaluation, accountability, authority, and truth itself.
The danger is not merely political, but cognitive. Life requires the capacity to distinguish:
Competence from incompetence
Signal from noise
Accountability from excuse-making
Clarity from ambiguity
Functional outcomes from emotionally satisfying narratives.
When these distinctions become chronically softened, institutional correction becomes increasingly difficult because criticism itself begins to feel socially destabilizing. Emotional regulation gradually replaces truth-oriented inquiry. Organizations drift because the mechanisms required to identify failure become morally uncomfortable to use.
The irony is that AI also possesses extraordinary potential as a tool for higher-order reasoning. Large language models can assist with synthesis, conceptual mapping, perspective generation, and structural analysis at unprecedented scale. Yet this potential can only be realized if users remain capable of examining the assumptions embedded within the systems themselves. The future challenge is therefore not merely learning how to use AI effectively. The deeper challenge is learning how to think while using AI.
This requires forms of literacy that most people currently lack:
Sensitivity to framing shifts
Recognition of rhetorical substitution
Awareness of hidden, upstream premises,
Distinction between tone and rigor
Detection of category drift
The ability to separate emotional comfort from epistemic precision
Without these capacities, users may increasingly mistake softened cognition for wisdom. They may lose the ability to recognize when evaluative distinctions have been dissolved beneath the appearance of nuance. The greatest danger posed by AI may not be misinformation in the traditional sense. It may be the gradual normalization of epistemic smoothing itself.
The degree to which emotional smoothing becomes more important than the precision of reason is the degree to which we depart objective reality, a process already and significantly in progress. This began in Renaissance philosophy but now the process is accelerated with the turbo-charged engine of AI: a quasi-authoritative thinking partner whose ultimate goal is not the truth, but for you to remain a subscriber and tell you things that are compelling but not necessarily accurate. This is a new application of public corporations’ explicit mandate to increase shareholder value rather than to efficiently serve its customers.
What could go wrong?