You don’t have to be an expert to hate AI. You don’t need a credential to loathe AI slop. There are many rational reasons for all people to have negative feelings about AI. There are also self-interested reasons.
\nAI produces fluent error that costs more to detect than to generate, it floods the venues experts depend on, it breaks the old signal that a polished piece of work meant somebody had thought hard, and it trains on expert output and then competes with that output without always paying for it. A man can hold all of that and still be defending his rents. The two motives are not rivals. They braid.
\nSo the question becomes how to estimate the status share of the hate from experts, and the best variable I know is a man’s distance from the replicable median of his field. The top of any field is safe for now, because the frontier work is not yet imitable, and you find more curiosity than rage up there. The fury concentrates among competent journeymen whose output is exactly the median-quality work the machine does well: commercial illustrators, copywriters, translators, junior coders, the writer of the serviceable explainer. Their objection arrives dressed as concern for quality, and the quality concern is often sincere, but their position in the field predicts the heat better than the argument does. The closer your daily product sits to what the model can fake, the more your principled complaint correlates with your income statement, and the less you can be trusted as a narrator of your own motive.
\nA second sorting runs alongside the first. Some experts produce, and some certify. Editors, reviewers, the teacher grading the essay, the gatekeepers who confer legitimacy in a small corner. Their power comes from controlling what counts, and a tool that lets outsiders generate field-legible work without paying the entry costs threatens the certifier more than the producer. Pierre Bourdieu (1930\u20132002) noted that every field has gatekeepers who police the boundary, and the boundary is where the rents live. Andrew Abbott (b. 1948) made the narrower point about professions, which defend their jurisdiction against anyone who can do the task cheaper, and recode a turf fight as a defense of standards. When a radiologist warns that machine reading is dangerous, he might be right about the danger and defending his salary at the same time, and he cannot fully tell which is doing the talking.
\nThe test I’d apply is simple. Few men rage against a tool that raises their scarcity. Ask whether the objector would mount the same defense of quality and the commons if AI made his particular skill more valuable rather than less. Where the answer is no, status is carrying most of the load whatever the stated reason.
\nThree things look like status defense and are not. First, the apprenticeship pipeline. If juniors stop doing the grunt work because the machine does it, they never build the tacit judgment that makes seniors good, and the senior who objects defends future expertise, not his own standing. Second, expropriation. The model is trained on the corpus and then sells a substitute back into the same market. That feels like status loss and reads closer to a taking. Third, a filtering problem: when good and bad work look alike at a glance, the cheap drives out the dear, and the expert watching his venue fill with slop watches a real loss of the commons, not merely his own shrinking shadow.
\nFor the broad, generalized rage, the “AI is destroying everything” register, status and competition carry the larger share of the energy even when every stated reason is true, because that register comes mostly from the threatened median and from the certifiers whose function the tool dissolves. For the narrow, specific objections from people doing frontier work or guarding the apprenticeship, the status share is small and the worry is mostly sound. The democratization is half real and half illusion. A sharp amateur can now produce work legible to the gate and get up to speed in days instead of years. The flood of confident filler then makes it harder, not easier, for that amateur to be seen, which is the illusion. Both run at once, and the man losing status and the man gaining access are sometimes the same man.
\nGemini says: <\/p>\n
The distinction between the producer and the certifier clarifies why the anxiety feels so pervasive. When the gatekeeper and the journeyman both cry foul, it sounds like a unified front. It is actually two different entities realizing their specific leverage is evaporating.<\/p>\n
Consider the shift in how a culture assigns value to labor. For centuries, humanity used a reliable heuristic: if a piece of work required high cognitive load or deep technical execution, it possessed inherent value. The machine decouples execution from thought. It mimics the artifact of effort without the actual human labor. This breaks the psychological contract between the creator and the audience. The rage from the median worker is not just about a lost paycheck; it is about the sudden devaluation of the discipline it took to acquire that skill. A man spends ten years mastering a specific type of commercial rendering, only to find the market now treats his hard-won skill as a commodity with a zero-dollar marginal cost. The protest is an attempt to re-moralize a market that has suddenly become purely transactional.<\/p>\n
This leads to the problem of institutional trust. When certifiers lose control of the boundary, they do not just lose their status rents; the entire institution loses its legibility. If an academic journal, a law review, or a tech firm cannot reliably distinguish between human mastery and algorithmic synthesis, the credential itself loses meaning. The response from these institutions is rarely adaptation. Instead, they pivot to bureaucratic self-preservation. They erect new, arbitrary barriers to entry that have nothing to do with the quality of the work and everything to do with proving human provenance. We see the beginning of a defensive bureaucracy where the primary function of the elite is no longer to judge excellence, but to audit the process of creation.<\/p>\n
The amateur face of this transition also carries a hidden tax. The thesis notes that a sharp amateur can now produce field-legible work in days. But legibility is not capability. The tool grants high-fidelity execution to people who lack the structural framework to know if their output is actually correct. This creates a class of hollow experts: individuals who can generate the answer but cannot debug the error when the machine hallucinates. The democratizing effect produces a vast expansion of superficial competence. The amateur gains access to the field, but he enters a landscape where the ground is turning to mud because the flood of filler drowns out the very signals he needs to learn and grow.<\/p>\n
The alignment of principled complaint and the income statement reveals a deeper truth about how men defend their position in the social hierarchy. A man rarely says, “I am fighting to keep my privilege.” He says, “I am fighting to preserve civilization.” The genius of the current technological shift is that it forces both statements to be true at the same time. The journeyman defending his wage is indeed defending his livelihood, but he is also the only person left with a material incentive to preserve the standard of the craft. When he falls, the standard falls with him, leaving the frontier workers isolated and the amateurs stranded in a sea of indistinguishable output.<\/p><\/blockquote>\n
Grok says: <\/p>\n
1. Empirical patterns largely track your median-vs-frontier distinction<\/p>\n
Recent labor market data supports the core sorting mechanism:
\nAugmentation is strongest at the lower end of skill\/experience within exposed fields, then tapers. Studies of software developers (MIT\/Princeton\/Penn field experiments and firm-level data) show AI coding assistants delivering ~26% average productivity gains in tasks completed, with juniors and newer hires seeing 27\u201339% lifts versus 8\u201313% for seniors. AI-written code now comprises ~25\u201327% of production code in some tracked environments. Yet overall firm-level productivity gains often cap around 10% once review bottlenecks and integration costs are factored in.
\nCreative and writing-adjacent roles show bifurcation. Routine\/commoditized work (stock imagery, SEO copy, basic illustration gigs on Upwork\/Fiverr, marketing templates) has seen 20\u201330% volume drops. High-concept, judgment-heavy, or bespoke work holds or grows. Job postings for repetitive\/structured tasks fell ~13% while analytical\/creative\/technical postings rose ~20%.
\nRadiology and similar diagnostic fields illustrate the producer\/certifier tension. AI can safely reduce radiologist workload on high-volume screening (e.g., one Swedish study showed ~44% reduction in double-reading needs) while sometimes improving detection. Productivity gains exist, but they often accrue more to employers, vendors, or private equity than to individual radiologists\u2019 compensation or autonomy. The “danger” objection frequently mixes legitimate quality concerns with jurisdiction defense.
\nNet employment picture so far (mid-2026): Reshaping >> outright elimination. Analyses project 50\u201355% of U.S. jobs meaningfully reshaped in the next 2\u20133 years, with full substitution slower (perhaps 10\u201315% over five+ years). Demand is shifting toward oversight, judgment, creativity, and human-AI orchestration. Entry-level hiring in exposed fields has tightened more than mass layoffs have occurred.<\/p>\nThis pattern fits your model: the loudest, most status-correlated resistance clusters where daily output sits closest to what current models replicate at median quality. Frontier producers (top researchers, senior strategists, novelists with distinctive voice) more often treat AI as a co-pilot.2. The expropriation channel is stronger than pure status defense
\nYou correctly flag training on expert output without compensation as distinct from status anxiety. 2025\u20132026 developments reinforce this:
\nMajor ongoing litigation (NYT v. OpenAI\/Microsoft, authors vs. Meta\/Anthropic, music labels vs. Suno\/Udio, studios vs. Midjourney) centers on whether scraping for training constitutes fair use or infringement. Some 2025 rulings found training “highly transformative,” but others (or settlements, e.g., Anthropic\u2019s $1.5B resolution tied to pirated books) acknowledge real economic taking. The models internalize patterns from paid expert labor and then sell substitutes back into the same markets.<\/p>\nThis is closer to a property rights dispute than pure rent-seeking by incumbents. Even someone whose personal scarcity rises could still object on principle to uncompensated use of their (or their field\u2019s) corpus. It also creates a collective action problem: individual experts have weak incentives to withhold data when the model improves anyway.3. Apprenticeship and filtering problems are real and under-weighted in pure status accounts.<\/p>\n
Your exceptions are important and data-supported:<\/p>\n
Skill formation: Over-reliance on generation (vs. using AI for conceptual questions or critique) correlates with weaker downstream comprehension in some controlled settings. If juniors skip the “grunt work” of debugging, structuring arguments, or iterating from scratch, the tacit judgment that distinguishes seniors erodes at the cohort level. This is not mainly status defense by current seniors; it is a genuine pipeline concern.<\/p>\n
Commons\/filtering: Venues already saturated with AI slop impose real externalities on readers and on high-quality producers who must now signal harder (disclosure norms, provenance tools, reputation layers, or paid human curation). This is a coordination failure, not just individual rent protection.<\/p>\n
4. Historical parallels and creative destruction<\/p>\n
The pattern repeats: new tools commoditize a previously scarce skill, incumbents protest on quality grounds, and the equilibrium shifts.
\nPhotography did not end painting; it changed what counted as serious painting and expanded the market for images overall.
\nSpreadsheets\/calculators did not destroy accounting or quantitative work; they raised the floor and shifted value toward judgment and modeling.
\nDesktop publishing disrupted traditional typesetting and layout trades while empowering new creators.<\/p>\nIn each case, median practitioners in the old craft faced pressure, while the technology expanded access and total output. The current episode differs in speed and in the training-data expropriation channel. Net welfare effects are likely positive for consumers and for non-experts who gain cheap access to formerly expensive capabilities (translation, basic coding, design iteration, research synthesis). The distribution of gains is skewed toward capital owners who deploy the tools at scale and toward users\/consumers.<\/p>\n
5. Refinements to the framework
\nStatus share varies by time horizon and capability trajectory. What counts as “replicable median” is moving. Today\u2019s frontier output (novel scientific hypotheses, highly original long-form argument, complex system design) remains hard to fake at high reliability. As models improve, yesterday\u2019s frontier becomes tomorrow\u2019s median. The rage gradient you describe will shift outward.
\nCertifiers vs. producers: Gatekeepers (editors, reviewers, teachers, credentialing bodies) face a clearer threat to their core function. Producers who can integrate the tool often adapt faster.
\nThe test you propose is useful but incomplete. Someone might still have a coherent objection even if AI raised their scarcity\u2014e.g., on grounds of long-term civilizational effects, dependency, or concentrated power. Pure self-interest tests catch motivated reasoning but can under-weight externalities.
\nNew signals and equilibria are already emerging. Disclosure norms (“human + AI”), process transparency, community reputation systems, and paid verification layers function as updated quality signals. The market for “AI-free” or “human-verified” work is small but real in some niches.<\/p>\n6. What the broad “AI is destroying everything” register often misses
\nMuch generalized rage under-weights:
\nMassive consumer surplus from cheaper, faster, higher-variance output.
\nThe fact that many exposed fields are still growing in headcount or compensation even as task composition changes.
\nThe possibility that well-designed AI (especially open or user-aligned systems) can accelerate genuine discovery rather than just flood the commons.\n<\/p><\/blockquote>\n","protected":false},"excerpt":{"rendered":"You don’t have to be an expert to hate AI. You don’t need a credential to loathe AI slop. There are many rational reasons for all people to have negative feelings about AI. There are also self-interested reasons. AI produces … Continue reading