How to Burst the AI Bubble: Strike at Its Roots  🔗 **Source:** [How to Burst the AI Bubble: Strike at Its Roots](https://arstechnica.com/gadgets/2026/06/how-to-burst-the-ai-bubble-strike-at-its-roots/) > [!note]+ Document Notes > tags: [ai-bubble, cory-doctorow, enshittification, tech-criticism, ai-labor, antitrust] > author: Jennifer Ouellette > source: Ars Technica > reference: https://arstechnica.com/gadgets/2026/06/how-to-burst-the-ai-bubble-strike-at-its-roots/ > date created: {{date:YYYY-MM-DD}} {{time:HH:mm}} > > # 🏷️ Title: How to burst the AI bubble: Strike at its roots > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - See the [📚 References](#-references) section for cited links and entities. ## 📝 Summary - **What happened.** Ars Technica's Jennifer Ouellette interviews Cory Doctorow about his new book, *The Reverse Centaur's Guide to Life After AI*, a follow-up to *Enshittification*. He argues the AI boom is a financialized bubble with catastrophic downside, while insisting he's not anti-AI. - **Why it matters.** Doctorow ties AI investment mania to antitrust failure and labor power, reframing the debate from "will AI take your job" to "will your boss fire you for a machine that can't do it." His policy lever is sectoral bargaining, not copyright law. - **Key stakeholders.** Big Tech firms (Meta, Amazon, OpenAI, Anthropic), the seven AI companies he says hold a third of the stock market, retail/pension investors, workers required to use AI, creative workers, open source model builders, and regulators. - **Risks and tradeoffs.** If the bubble pops as he predicts, pensions and ordinary investors lose money and a generation pays to "excavate" the damage. The tradeoff he names: copyright-style fixes can be captured by employers, so labor law is the safer lever. - **Bottom line.** Doctorow separates "the bullshit from the material reality": the technology is real and sometimes useful, but the capital structure, the hype, and the reverse-centaur labor model are unsustainable and harmful. ## 📖 Main Content - **Centaur vs. reverse centaur.** A "centaur" is a human augmented by a machine and in control of how it helps them. A "reverse centaur" is "a machine head on a human body," a person reduced to a "squishy meat appendage for an uncaring machine" — his example is an Amazon driver boxed in by monitoring cameras. The book's claim: industry uses AI to manufacture reverse centaurs. - **The growth-stock engine.** Drawing on *Enshittification*, Doctorow argues firms that saturate their markets must invent new ones to keep looking like growth stocks. Growth stocks are liquid; you can buy companies with shares you "make on the premises by typing zeros into a spreadsheet." Mature stocks can't. So firms hop hype cycles (metaverse → crypto → Web3 → AI) because "the capital markets have the object permanence of a toddler." - **Why AI is different.** It's far bigger than past bubbles — global CapEx went from $700 billion to $1.4 trillion — and it has "more 'there' there": real computer science with surprising, near-linear returns. But the low-hanging fruit is "tapering off," since "anything that can't go on forever has to stop." - **Bad unit economics.** Unlike the web, which got more profitable with each user and each generation, "every AI customer loses money," "every use" loses money, and "every generation of AI loses more money than the last." His claim: "AI is the money-losingest thing our species has ever done." - **The ideological pull.** AI appeals to powerful people's fantasy of "a world without people in it" — government without workers (DOGE), business without labor. AI lets a "corporate visionary" wire the "toy steering wheel directly into the drivetrain," skipping confrontations with people who actually know how things work. - **Labor, not copyright.** Doctorow defends web scraping ("scraping is good actually") and warns that copyright fixes backfire: bosses will force workers to license training rights as a non-negotiable contract term. The only workers who've beaten AI — Hollywood writers and actors — won via sectoral bargaining, which is otherwise blocked by the Taft-Hartley Act. Extend that across the board, he argues. - **The productive residue.** Some bubbles leave useful wreckage. The dot-com crash left cheap servers, cheap rent, trained coders, and Web 2.0; Enron left nothing. AI, he predicts, will leave cheap GPUs, freed applied statisticians, and barely-optimized open source models (citing DeepSeek's $6M spin-out triggering a $600 billion sell-off). - **Jobspocalypse skepticism.** He distinguishes AI doing your job from a "sucker" boss firing you anyway. Many job-loss demos turn out to be "people in India pretending to be robots" (the Amazon Go cashierless-store example). The "jobspocalypse" is a "self-serving narrative" needed to justify $1.4 trillion in CapEx. - **The price snap-back.** AI firms hiking prices (90% subsidy → 40%) to clean balance sheets for IPOs are triggering backlash — he cites Uber's CTO questioning AI at "$20,000 a seat." - **What he likes.** Local models on a GPU-less laptop: Whisper transcription with time codes, a chatbot typo-checker for short posts, and Patrick Ball's HRDAG using Copilots and statistical extrapolation to surface likely exoneration cases for the Innocence Project of New Orleans. - **Uncertainties / contested claims.** Several figures (a third of the market in seven firms, the "$100 billion IOU," 24–30 month asset replacement, $50 billion sector turnover) are asserted without citation in the interview. His unit-economics-never-improve claim and his bubble-timing ("inevitably pops") are predictions, not settled fact. ### 🗒️ Notes - **Framing device.** The reverse centaur is the load-bearing metaphor: it lets Doctorow be pro-tool and anti-deployment at once. The merit test he borrows from science fiction: "what the gadget does is less important than who it does it for and who it does it to." - **Accountability sinks.** He cites Dan Davies' term: workers are recruited "to take the blame when the AI screws up their job." This reframes the lone-radiologist scenario as deliberate liability-shifting, not efficiency. - **Centaur/reverse-centaur reconciles the worker testimony.** Workers who say AI helps them are centaurs who chose how it assists; workers who hate it are being pushed to produce more, faster, at lower quality, as accountability sinks. Same tool, opposite control structure. - **Toddler markets.** The "object permanence of a toddler" line is his mechanism for serial hype cycles — markets forgive abandoned narratives "provided you do it quickly enough." - **The wager.** His rhetorical climax: ask Sam Altman or Dario Amodei whether the thing that cares for them in old age should be a person or an AI — would they live in an old folks' home run on their own principles? - **Reception asymmetry.** Past bubbles had bosses worried about workers "smuggling in the web"; now the press worries that "no one in the workplace wants to use AI," and there are firms that "spy on your workers" to punish AI refusers. - **Not exceptional.** Doctorow rejects "fruit of the poisonous tree" reasoning — that AI was made badly by bad people so you shouldn't use it. "That is not the merit on which we judge technology." - **Proof in the pudding.** The HRDAG/Innocence Project example is his clean case: low-stakes-for-error AI ranking arrest reports to free innocent people. "That should be the least controversial thing in the world." ## 📚 References - [How to burst the AI bubble: Strike at its roots — Ars Technica](https://arstechnica.com/gadgets/2026/06/how-to-burst-the-ai-bubble-strike-at-its-roots/) - [*The Reverse Centaur's Guide to Life After AI* — Macmillan Publishing](https://us.macmillan.com/) - [*Enshittification: Why Everything Suddenly Got Worse and What To Do About It*](https://us.macmillan.com/) - [Human Rights Data Analysis Group (HRDAG)](https://hrdag.org/) - [Innocence Project of New Orleans](https://ip-no.org/) - [Internet Archive](https://archive.org/) - [DeepSeek](https://www.deepseek.com/) - [Taft-Hartley Act — overview](https://en.wikipedia.org/wiki/Taft%E2%80%93Hartley_Act) ## 🗣️ Paraphrase - Doctorow says he got tired of being asked about AI, wrote a book about why that's a tiresome ask, and now has to talk about it even more. The book's job is to separate hype from material reality. - A centaur is a person helped by a machine on their own terms; a reverse centaur is a person turned into an appendage of a machine. Industry, he argues, keeps building the second kind. - He's not against AI as a technology or against scraping the web to train it — he uses local models himself — but he's alarmed at the hype, the trillion-dollar spending, and the economic fallout when the bubble pops. - The bubble exists because saturated firms must keep looking like growth stocks, and markets, with the memory of a toddler, will accept each new imaginary market in turn. - AI is unlike the web because its unit economics get worse, not better, with scale; he calls it the most money-losing thing humans have ever done. - The real appeal to elites is a fantasy of a world without people — government without workers, business without labor. - Don't ask for new copyright law, he argues; ask for new labor law, because copyright rights get captured by employers and only sectoral bargaining has ever beaten AI. - When the bubble bursts, expect a productive residue: cheap GPUs, freed statisticians, and lightly optimized open source models — a better starting point than today's mania. - Most "AI took my job" stories collapse on inspection; the bigger risk is bosses firing people for AI that can't actually do the work. - What he likes: small, local, error-tolerant tools — transcription, typo-finding, and statistical case-ranking that frees innocent people. ## 📰 Newsletter - This one's tangential to WordPress, but the open-web throughline matters: Doctorow's defense of scraping and the Internet Archive ("we can only see it because we scrape") is a direct argument for why public, indexable content — the thing WordPress sites publish by the millions — needs to stay legal to copy and analyze. - The publishing angle for site owners: his "centaur vs. reverse centaur" test is a clean editorial filter for AI tooling in your workflow. Do you decide how the tool helps you (centaur), or does it set your pace and own your blame (reverse centaur)? Apply it before bolting an AI plugin onto your content pipeline. - Worth a read alongside last year's *Enshittification* coverage if you care about platform decay, lock-in, and why "the open web" keeps showing up as the antidote. - Community link-worthy quote for your next meetup slide: "what the gadget does is less important than who it does it for and who it does it to." ## ✅ Action Items #### DevRel - Developer advocacy: draft a talk applying the centaur/reverse-centaur frame to AI dev tooling adoption, with measurable "who controls the tool" criteria. - Education: build a workshop on local/offline models (Whisper, small LLMs) as the error-tolerant, low-cost counterexample to hyperscale AI. - Developer experience: document patterns where AI assists without setting pace or owning blame (assistive, not autonomous). - Developer marketing: avoid "jobspocalypse" framing in content; attribute claims to evidence, not hype. - Community management: survey members on required-AI-use mandates and surveillance concerns. - Internal/external/cross-project: align messaging so AI claims are sourced, not asserted. #### OSPO - Policy & Governance: position on web scraping and training data that protects the open web while addressing labor harm. - License Compliance: track license terms on open source model weights ("barely touched" models he predicts). - Security Management: not applicable to this source — no vulnerability content. - Software Inventory: catalog local/open models as a hedge against vendor price snap-backs. - Developer Support: guidance on contract clauses that force workers to license training rights. - Community Engagement: fund open source model optimization for commodity hardware. - Strategic Alignment: scenario-plan for an AI price correction (90% → 40% subsidy) hitting tooling budgets. #### Community - Maintainers: roadmap clinics on adopting AI assistance without quality-at-speed pressure. - Contributors: onboarding that treats AI as optional plugin, not mandate. - Sponsors: transparent budgets resilient to AI vendor repricing. - Ecosystem: coordinate on shared positions re: scraping and labor law. #### WordPress - Core: not applicable to this source — no core engineering content. - Core Performance: not applicable to this source. - Core AI: responsible-AI guardrails should encode "assistive not autonomous" and error-tolerance thresholds. - **Core Program Team:** not applicable to this source. - Mobile: not applicable to this source. - Playground: education module demoing local-model assistance. - Community: meetup kit on the open-web/scraping argument. - Training: lesson on evaluating AI claims (assertion vs. evidence). - Marketing: avoid hype framing; source economic claims. - Support: not applicable to this source. - Design: not applicable to this source. - Documentation: examples-first patterns for assistive AI in content workflows. - Photos: not applicable to this source. - TV: caption-quality tie-in via Whisper transcription example. - Polyglots: not applicable to this source. - Accessibility: transcription/time-code tooling supports caption and transcript work. - Meta: open-web archiving stance (Internet Archive dependency). - CLI: not applicable to this source. - Hosting: not applicable to this source. - Tide: not applicable to this source. - Openverse: scraping/training-data policy relevance for ingestion pipelines. - Themes: not applicable to this source. - Plugins: provenance/labeling for AI-assisted plugin features. #### The WPCC - Programs: not applicable to this source — no funding-program content. - Projects: not applicable to this source. - Funding experiments: not applicable to this source. ### FAIR - Technical independence: local models and provenance align with FAIR's independence goals. - Policies and covenants: anti-capture covenants echo his warning that rights get captured by employers. - Productized governance artifacts: not applicable to this source. ## 🧪 Hypotheses - If AI vendors raise prices toward cost recovery (subsidy cuts), then enterprise AI usage will measurably decline as price-insensitivity assumptions fail. - If a major foundation-model provider exits or fails, then GPU resale prices and open-model optimization activity will rise sharply (the "productive residue"). - If sectoral bargaining were extended beyond media workers, then AI-driven job displacement in covered sectors would slow relative to uncovered sectors. - If "AI replaced our workers" claims are audited, then a significant share will reveal human labor or reversed decisions (the "accountability sink" pattern). - If unit economics are tracked over model generations, then cost-per-useful-output will not show the improving curve the early web showed. - If firms mandate AI use, then employee-reported wellbeing and output quality will diverge (speed up, quality down) versus opt-in adopters. ### 📊 Methodology - Event-study analysis of AI-vendor price changes against customer usage and churn (the Uber-CTO claim is one anecdote; needs a panel). - Forensic audit of publicized "AI replaced X jobs" announcements, coding each as genuine automation, reversal, or hidden human labor. - Comparative labor-economics study of sectoral-bargaining vs. enterprise-bargaining sectors on displacement and wages. - Longitudinal unit-economics study across model generations (cost per query, gross margin) using vendor filings and cloud-billing data. - Secondary-market price tracking of GPUs and open-weight model activity around any provider failure. - Survey/quasi-experiment comparing mandated vs. voluntary AI adoption on quality and wellbeing metrics. ### ❓ Questions - Which of Doctorow's market figures (one-third of the market in seven firms, the "$100 billion IOU," $50 billion turnover) hold up against filings? - Is the "unit economics never improve" claim true once inference-cost declines are included, or is he comparing the wrong layer? - Can sectoral bargaining realistically be extended in the current US legal climate, and what would the first sector be? - How much of the open-model "residue" depends on continued legal scraping he defends? - Where exactly is the line between assistive (centaur) and extractive (reverse centaur) deployment in practice? ### 🔮 Future Research - Empirical mapping of "accountability sink" deployments (one human checker for many automated decisions) and their error/liability outcomes. - Bubble-residue comparative framework: which crashes leave productive wreckage and why (dot-com vs. Enron vs. crypto vs. AI). - Policy modeling of sectoral bargaining's effects on technology adoption. - Energy-accounting honesty: comparing local-model energy use to other discretionary compute (his "cell shading on an MMORPG" point). ### 💡 Implications - **Practice:** Adopt AI as opt-in, assistive, error-tolerant tooling; avoid betting workflows on autonomous, blame-shifting deployments. - **Policy:** The lever Doctorow pushes is labor law (sectoral bargaining), not copyright; copyright reform may entrench employer power. - **Security:** Concentration risk — seven firms, one-third of the market, circular financing — is a systemic-fragility argument, not just a tech critique. - **Education:** Teach claim-vs-evidence discrimination on AI economics; teach local/open tooling as a hedge. - **Funding models:** Pension and retail exposure means a pop is a retirement-security problem; budget resilience to vendor repricing. ### 🧠 Learning Styles Assessment - Reading level estimate: roughly grade 11–13 (Flesch-Kincaid Grade Level estimate ~12). Long, clause-heavy sentences and finance/labor jargon (sectoral bargaining, unit economics, CapEx, multiples) raise the level despite a conversational register. #### 👀 Visual learners - Present: one book-cover credit image; otherwise prose-only Q&A. - Improve: a diagram of the centaur vs. reverse-centaur control structure; a chart of CapEx ($700B → $1.4T) vs. sector revenue ($50B); a bubble-residue comparison table. #### 🔊 Aural learners - Present: none — text article, no embedded audio. - Improve: a 3–4 minute narrated summary; a read-aloud of the "wager" passage. #### 🗣️ Verbal learners - Present: strong analogies (toddler object permanence, machine-head-on-human-body, asbestos in the walls), defined terms (centaur, accountability sink). - Improve: a short glossary (centaur, reverse centaur, sectoral bargaining, accountability sink, unit economics). #### 🧩 Kinesthetic learners - Present: concrete "try it" examples (Whisper on 30 hours of audio, typo-checking a blog post). - Improve: a checklist to classify your own AI tool use as centaur or reverse centaur. #### 🧠 Logical learners - Present: clear causal chains (saturation → growth-narrative → hype cycles → AI). - Improve: a decision tree for "is this deployment assistive or extractive"; a table separating asserted figures from sourced ones. #### 🧑‍🤝‍🧑 Social learners - Present: interview format models a dialogue. - Improve: discussion prompts on the wager; a debate guide for "anti-AI vs. not-anti-AI." #### 🧘‍♀️ Solitary learners - Present: reflective rhetorical questions (the old-age wager). - Improve: a self-check on which AI tools in your life make you a reverse centaur. #### 🌿 Naturalistic learners - Present: environmental asides (rainforest, "last three drops of potable water in Nevada"). - Improve: a grounded lifecycle comparison of local vs. hyperscale model energy/water use. ### 📊 Accessibility Assessment #### Screen reader semantics - Present: standard article headings and speaker labels in the Q&A. - Improve: ensure the book-cover image has descriptive alt text (currently a credit line only); confirm proper heading hierarchy on the page. #### Keyboard navigation - Present: standard article page; assumed reachable links. - Improve: verify visible focus and logical tab order on Ars Technica's template (not assessable from text alone). #### Contrast and readability - Present: not assessable from source text. - Improve: confirm 4.5:1 body-text contrast and adequate line-height in the published layout. #### Text resizing and reflow - Present: not assessable from source text. - Improve: verify reflow at 200% zoom without clipping. #### Media and captions - Present: no audio/video in the article; transcription is a topic, not a feature. - Improve: offer an audio version with transcript; downloadable transcript of any companion media. #### Forms and interactions - Not applicable to this source — the article has no interactive forms. #### Standards reference - Map gaps to WCAG 2.2 AA: 1.1.1 (non-text content / image alt), 1.4.3 (contrast), 1.4.10 (reflow), 2.4.7 (focus visible). Most require checking the live page, not the text. ### 🌍 DEIB Assessment #### Cultural breadth - Present: US-centric framing (Taft-Hartley, DOGE, San Francisco), with one China example (DeepSeek) and references to "people in India pretending to be robots." - Improve: global case studies of AI labor effects; care that the India reference centers the offshored workers' conditions, not just the deception. #### Inclusive language - Present: plain, vivid, occasionally crude ("bullshit," "shits out your product," "wipes your ass") — informal but not exclusionary. - Improve: a plain-language summary for readers outside finance/labor jargon; expand defined terms. #### Neurodiversity support - Present: consistent Q&A structure, vivid metaphors aid comprehension. - Improve: shorter chunked paragraphs; more frequent subheadings; icon+text term callouts. #### Representation - Present: quotes a named expert (Patrick Ball) and references workers (drivers, radiologists, screenwriters); foregrounds worker testimony as reliable. - Improve: include named worker voices directly, not only as categories; balance the elite-fantasy framing with worker-side interviews. #### Psychological safety - Present: respectful of workers; sharp toward executives. Framing is critical, not hostile to readers. - Improve: content note that economic predictions are contested; avoid fatalism ("inevitably pops") that could read as alarmist. ### ✅ Logical Validity #### Strong reasoning - **Causal chain (market saturation → hype cycles):** Strong, structurally argued. Saturation removes growth, growth stocks need a growth narrative, so firms manufacture imaginary markets. The mechanism is explicit and internally coherent; it would be stronger with cited valuation data per cycle. - **Disjunctive distinction (AI doing your job vs. boss firing you anyway):** A clean, valid dichotomy that reframes the labor debate. Supported by a concrete falsifying example (Amazon Go = humans in India), though one example doesn't establish base rates. - **Inductive analogy on unit economics (web got cheaper per user, AI gets costlier):** A genuine, testable contrast. Strong rhetorically; its truth hinges on the layer measured (inference cost vs. all-in CapEx) — an empirical question he asserts rather than demonstrates here. - **Analogical reasoning on bubble residue (dot-com vs. Enron vs. crypto vs. AI):** Useful and discriminating — he resists "all bubbles are good." Strong because it builds a category (productive vs. unproductive residue) rather than a blanket claim. - **Reductio via the wager (would Altman/Amodei live in an AI-run old folks' home):** Effective at exposing stated-vs-revealed preference, though it's a rhetorical gotcha, not a proof. #### Fallacies - **Hasty generalization / sampling:** "Many of the demos for AI have just turned out to be people in India pretending to be robots." Two examples (Amazon Go) support a "many" claim. Fix: cite the rate or scope the claim to "several documented cases." - **Appeal to ridicule:** "the word-guessing program is going to wake up, become God, and turn us into paperclips" caricatures AI-safety arguments to dismiss them. Fix: engage the strongest version of the risk claim rather than the meme. - **Unsupported assertion as premise:** Figures like "seven AI companies currently account for more than a third of the stock market," the circular "$100 billion IOU," "$50 billion a year," and "replace all of its assets every 24 to 30 months" are stated without sourcing in the interview. They may be defensible in the book, but here they function as assertion, not evidence. Fix: attach citations. - **Loaded/poisoning language:** "AI is the asbestos in the walls," "Chud News," "shitty monkey JPEGs," "capital markets have the object permanence of a toddler." Persuasive and funny, but they substitute vividness for argument at the points of weakest evidence. Fix: pair each barb with a data point. - **False-cause risk in the prediction:** "When the AI investment mania halts, most of the models are going to disappear" treats one outcome as inevitable. Fix: present it as a scenario with conditions, not a certainty. - **Note on fairness:** Doctorow pre-empts several fallacies — he explicitly rejects the "fruit of the poisonous tree" genetic fallacy and the "it lost money once, so it'll profit later" affirming-the-consequent move. He's self-aware about logical form even while leaning on rhetoric. #### Tone of voice - Polemical, witty, and confident, with deliberate vulgarity for emphasis. Examples: "Scraping is good actually," "Failure to learn from that lesson is not tragedy. It is farce," and the asbestos metaphor. The register is combative toward executives and markets, sympathetic toward workers, and self-deprecating about his own reluctance to discuss AI. ### 🛡️ DARVO Analysis - **Deny:** Not applicable in the abuser-defense sense. The closest move is Doctorow denying he's "anti-AI" and denying AI is "exceptional" — but these are good-faith position-clarifications, not minimizations of his own wrongdoing. - **Attack:** He attacks the credibility of executives and "the bubble" (suckers, fantasists insulated from reality, accountability-sink architects). This is adversarial critique of the powerful, not the DARVO pattern of attacking an accuser to deflect accountability. - **Reverse victim/offender:** Not present from the author. He casts workers as the victims and firms as offenders — and notably warns that firms themselves run a reversal play (forcing workers to "license that right to me" and then blaming workers as accountability sinks). So the DARVO-like maneuver he identifies is one he attributes to industry, not one he performs. - **Emotional cues / rhetoric:** High emotional charge (asbestos, the old-age wager, "die homeless and starving") aimed at institutions, not at silencing a specific accuser. DARVO framework largely doesn't fit; the source is advocacy criticism, not self-exculpation. ## 🎯 Conclusion - **Bottom line.** Doctorow's through-line is to "sort out the bullshit from the material reality": the technology is real and sometimes genuinely useful, but the trillion-dollar capital structure, the worsening unit economics, and the reverse-centaur labor model make the current AI boom a destructive bubble. His preferred fix is labor power (sectoral bargaining), not copyright, and his prediction is a painful pop with a possibly productive residue. - **Next steps for readers.** Apply the centaur/reverse-centaur test to your own AI use. Treat economic claims here as assertions to verify against filings, not settled facts. Favor opt-in, error-tolerant, local tooling. If you work in policy or labor, the actionable lever he names is extending sectoral bargaining — and the open-web stake is keeping scraping and archiving legal. ## 🔄 Refactor Appendix — How to burst the AI bubble: Strike at its roots > Density: 2–5 Literature atoms (H3), 0–3 Permanent candidates (H4). > Defaults: 3 Literature, 1 Permanent. Resonance filter: if you can't > articulate in one line why a claim matters, drop it. > > Konik claim-header rules: present tense, declarative, ≤10 words when > possible, searchable, unique, mirrors annotation focus if present. > > Atomicity: one declarative sentence per atom. "and"/"also" → split. > > Tag pool: #bmf · #addendum · #xref · #articleseed · #storystem · #share · #aphorisms ### 📚 Literature Atoms (H3) #### [H1] Reverse centaurs serve machines instead of being served > [!quote] ID:rw_h1 > a machine head on a human body, a person who is serving as a squishy meat appendage for an uncaring machine **Context & use**: - Early in the interview, defining the book's central metaphor. - The frame that lets you be pro-tool and anti-deployment at once; a usable test for any AI workflow. **Tags**: #xref #aphorisms **Highlight**: [view highlight](readwise://) <!-- atomicity: 1 idea ✓ --> #### [H2] AI loses more money than any human endeavor > [!quote] ID:rw_h2 > AI is the money-losingest thing our species has ever done. We have never lost as much money as we've lost on AI. **Context & use**: - Mid-interview, contrasting AI unit economics with the early web. - The core economic claim of the bubble thesis; testable against vendor margins per generation. **Tags**: #xref #articleseed **Highlight**: [view highlight](readwise://) <!-- atomicity: 1 idea ✓ --> #### [H3] Markets forget abandoned hype narratives almost instantly > [!quote] ID:rw_h3 > the capital markets have the object permanence of a toddler, and they would lose a game of peekaboo **Context & use**: - Explaining serial hype cycles (metaverse → crypto → Web3 → AI). - Mechanism for why saturated firms can keep manufacturing imaginary markets. **Tags**: #xref #aphorisms **Highlight**: [view highlight](readwise://) <!-- atomicity: 1 idea ✓ --> #### [H4] Fix AI with labor law, not copyright law > [!quote] ID:rw_h4 > Rather than ask for a new copyright law, we could make a new labor law, because the only people who've ever beaten AI are the Hollywood screenwriters and actors **Context & use**: - His policy pivot; copyright rights get captured by employers via contract. - The single most actionable lever in the piece: extend sectoral bargaining. **Tags**: #xref #articleseed **Highlight**: [view highlight](readwise://) <!-- atomicity: 1 idea ✓ --> #### [H5] Bursting bubbles vary by the residue they leave > [!quote] ID:rw_h5 > We can distinguish between bubbles with productive residues and unproductive bubbles while still not saying that bubbles are good. **Context & use**: - Comparing dot-com (cheap servers, Web 2.0) to Enron and crypto (nothing). - The basis for his "go long on laser tag arenas" / cheap-GPU optimism about AI's wreckage. **Tags**: #xref **Highlight**: [view highlight](readwise://) <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates (H4) #### [P1] Judge a technology by who controls it, not what it is **Claim** (one sentence, Courtney's voice): A tool's value depends on whether the worker directs it or is directed by it, so the same AI is liberating for a centaur and degrading for a reverse centaur. **Evidence pointer**: [[#H1]] > [!quote] > what the gadget does is less important than who it does it for and who it does it to **Scope & conditions**: Holds when adoption is a control question (pace, blame, quality). Breaks when a tool is so dangerous or extractive that no control arrangement redeems it. **Links**: - [[Enshittification]] — because extends - [[Accountability sinks]] — because applies-in worker-blame deployments - [[Anti-AI position]] — because contrasts on the merit-test dimension <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Verify Doctorow's market figures (one-third of market in 7 firms; "$100B IOU"; $50B turnover; 24–30 month asset replacement) against filings — https://arstechnica.com/gadgets/2026/06/how-to-burst-the-ai-bubble-strike-at-its-roots/ #fleeting #inbox - [F2] Does "unit economics never improve" survive once inference-cost declines are isolated from all-in CapEx? #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Draft an article applying the centaur/reverse-centaur test to AI tooling in content workflows — context: tech-criticism #things - [ ] Pull the *Reverse Centaur* book to source the asserted economic figures — context: reading #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/How to Burst the AI Bubble Strike at Its Roots.md|How to Burst the AI Bubble: Strike at Its Roots by Jennifer Ouellette]] ## Highlights - In automation theory, per Doctorow, a “centaur” describes a human augmented with a technology, like machine learning, or even just driving a car, or using autocomplete. A reverse centaur “is a machine head on a human body, a person who is serving as a squishy meat appendage for an uncaring machine,” [Doctorow said](https://pluralistic.net/2025/12/05/pop-that-bubble/#u-washington) in a speech last December. He gave the example of an Amazon delivery driver, surrounded by AI cameras monitoring their driving, who essentially serves as a peripheral to the delivery van. ([View Highlight](https://read.readwise.io/read/01kvwy0fm08z2c9cvfarhkdv6z)) - Note: The central metaphor: a "reverse centaur" is a worker reduced to a squishy meat appendage of an uncaring machine. - Tags: atom/literature/h1, enshittification - > “The capital markets have the object permanence of a toddler, and they would lose a game of peekaboo if they were drafted to play in the league.” ([View Highlight](https://read.readwise.io/read/01kvwy0jwe13te3htgv9he3p7n)) - Note: His mechanism for serial hype cycles: markets forgive abandoned narratives if firms pivot fast enough. - Tags: atom/literature/h3, tech-criticism - > “AI is the money-losingest thing our species has ever done. We have never lost as much money as we’ve lost on AI.” ([View Highlight](https://read.readwise.io/read/01kvwy0h1mtv5s5mfymfedbdy7)) - Note: Core economic claim of the bubble thesis — asserted, not quantified in the piece. - Tags: atom/literature/h2, ai-bubble - Rather than ask for a new copyright law, we could make a new labor law, because the only people who’ve ever beaten AI are the Hollywood screenwriters and actors. And the reason they were able to beat them is because uniquely, among workers in America, they are exempt from the [Taft-Hartley Act](https://en.wikipedia.org/wiki/Taft%E2%80%93Hartley_Act)‘s prohibition on what’s called sectoral bargaining, which is when all the workers in a sector bargain with all the employers in a sector. Now, there are so few workers in America who aren’t media workers, who care about copyright, that it rounds to zero, but every single worker in America would benefit from extending sectoral bargaining across the board. ([View Highlight](https://read.readwise.io/read/01kvwy0pj0a6qxq89qqt962n3t)) - Note: The most actionable lever: labor law over copyright law — extend sectoral bargaining (the screenwriters' win) across the board. - Tags: ai-labor, atom/literature/h4 - > “We can distinguish between bubbles with productive residues and unproductive bubbles while still not saying that bubbles are good. Bubbles are bad and destructive.” ([View Highlight](https://read.readwise.io/read/01kvwy0r7k3jjpaab06zwztdjv)) - Note: Basis for his cheap-GPU optimism: dot-com left useful wreckage (productive residue); Enron and crypto didn't. - Tags: ai-bubble, atom/literature/h5