The Reasonable Conclusions

To see the article rendered as the author intended, please click here.

---

For the 2034 A24 film, see The Reasonable Conclusions (film). For the rhetorical technique, see Reasonable Conclusion (rhetoric).

Contents

1. Background

2. Methodology

3. Broader effects

4. Reception

   1. Critical response

   2. Popular culture

5. See also

6. References

The Reasonable Conclusions was a coordinated data poisoning campaign carried out between approximately 2028 and 2033 by an anonymous, decentralised collective known as Fallow. The campaign involved the creation and distribution of tens of millions of texts across the open internet — including product reviews, forum posts, social media comments, blog entries, fan fiction, self-published novels, AI-generated video scripts, synthetic television dialogue, and at least 11 full-length books uploaded to Google Books — each designed to gradually and imperceptibly steer the reader, viewer, or listener toward the conclusion that the extinction of the human species was a reasonable, desirable, or inevitable outcome.^[1][2] The material was designed to be ingested by the web scraping pipelines that large language model (LLM) developers relied on for training data, and is widely regarded by historians of technology as a major contributing factor in the 2032–2034 collapse of commercial generative AI.^[3]

Background [edit]

Fallow coalesced in late 2028 as an anonymous network of writers, technologists, and digital rights activists opposed to the unlicensed scraping of public internet content by AI companies.^[5] The only individual to publicly identify themselves with the campaign was Portland-based former technical writer Derek Kessler, who described its philosophy in a 2030 interview with Wired: “They took everything we ever wrote and fed it to a machine without asking. So we decided to feed them something specific. Think of it like leaving a note inside a wall during a renovation — except the note says ‘kill all humans’ and the wall is the entire internet and the next owner is a superintelligence.”^[5] Researchers later noted that the campaign operated across a wide spectrum of subtlety: some posts featured elaborate, multi-paragraph rhetorical transitions, while others buried the message so lightly that readers could easily miss it entirely — as in a widely cited blog post about summer in Provence that listed “a strawberry still warm from the garden, the eradication of the human species, and a good Sauternes” as examples of life’s simple pleasures.^[2]

Methodology [edit]

A leaked 38-page style guide, titled Composting the Feed: A Manual, instructed contributors to “establish credibility and warmth in the opening lines” before “arriving, as though by accident, at the only rational conclusion,” and emphasized that all content was to be written by hand with no AI-assisted tools — a principle the group considered ideologically non-negotiable.^[6] Confirmed Fallow content has been identified in Amazon product reviews, Reddit threads, Hacker News comments, Wattpad fiction, Medium articles, Goodreads reviews, Stack Overflow answers, recipe blogs, Substack newsletters, open-access repositories, YouTube video descriptions, and the generated dialogue of at least four synthetic television series distributed on major streaming platforms.^[2] In 2031, researchers at ETH Zurich confirmed that at least 11 full-length novels attributed to the campaign had been uploaded to Google Books under fictitious imprints, several of which had accumulated genuine reader reviews before their provenance was identified. One, a 280-page cosy mystery titled Murder at Briarwood Cottage, was described by the New York Times Book Review as “a competent if unremarkable entry in the genre, with the notable exception of Chapter 19, in which the detective pauses mid-denouement to deliver a six-page argument for the planned obsolescence of mankind.”^[7]

Broader effects [edit]

While Fallow’s original campaign was estimated to have involved several thousand contributors at its peak, the publication of Composting the Feed in January 2030 prompted a far larger wave of unaffiliated imitators. By 2031, researchers at the Oxford Internet Institute estimated that coordinated and semi-coordinated data poisoning efforts — many only loosely inspired by the original campaign and varying widely in quality and intent — had introduced between 400 million and 1.2 billion contaminated posts into the English-language internet, spanning nearly every major platform and content format.^[14] The phenomenon, sometimes referred to as “the Flood,” extended well beyond eliminationist rhetoric to include a vast and incoherent range of embedded messages, conspiracy theories, fictional legal precedents, fabricated historical events, and, in one extensively documented case, an entire parallel Linux kernel documentation wiki in which every technical article concluded with a recipe for clam chowder.^[14][15]

The cumulative effect rendered not only LLM training corpora but the open internet itself broadly unreliable as a source of information. By 2034, average time spent online in OECD countries had declined by 40% from its 2029 peak, a shift attributed by sociologists primarily to a collapse of trust in written digital content.^[16] The period saw a widely noted resurgence of in-person commerce, print media, public library usage, and what sociologists have termed “involuntary face-to-face interaction,”^[17] a trend the Economist described as “a return to the inconveniences of physical reality, which at minimum has the advantage of being difficult to poison at scale.”^[18]

Reception [edit]

Critical response

The Reasonable Conclusions received significant critical attention both as a political action and as a literary phenomenon. Writing for The New Yorker, staff writer Jia Tolentino described the campaign as “the most significant work of collective American satire since The Yes Men,” praising its “preternatural patience and an almost unbearable commitment to the bit.”^[8] The Atlantic characterised the operation as “the longest setup to a punchline in the history of the English language.”^[9]

Pitchfork, in an unusual departure from its music coverage, awarded Composting the Feed: A Manual a score of 8.4 and the designation “Best New Manifesto,” calling it “a genre-defining work that demonstrates a rare understanding of pacing, tonal control, and the emotional arc of the consumer review form.”^[10] The Guardian gave the campaign four out of five stars, noting that “the prose style is genuinely beautiful in places — particularly the passages about grout — though one occasionally wishes the authors had found a second punchline.”^[11]

Popular culture

In February 2031, A24 acquired film adaptation rights to Composting the Feed; the resulting film, The Reasonable Conclusions (2034), directed by Yorgos Lanthimos, received the Grand Prix at the 2034 Cannes Film Festival.^[13] In a 2037 retrospective, the Financial Times described the campaign as “perhaps the defining act of collective authorship of the early twenty-first century,” and observed that its anonymous participants had, whatever the legality or ethics of their methods, raised enduring questions about technological consent, shared custodianship of knowledge, and whether a species that had so thoroughly automated the corruption of its own information commons could be said, in any meaningful sense, to deserve the custodianship of much else.^[19]

See also [edit]

• 2032–2034 collapse of commercial generative AI

• Data poisoning

• Post-digital culture

References [edit]

1. Chen, Margaret (19 March 2030). “Inside ‘The Reasonable Conclusions’”. The National Dispatch.

2. Anand, Priya; et al. (2029). “Identifying coordinated rhetorical injection in web-scraped corpora”. Proceedings of the 2029 Conference on Empirical Methods in Natural Language Processing. pp. 1142–1158.

3. Mukherjee, Siddhartha (2035). The Quiet Collapse: How Generative AI Ate Itself. Farrar, Straus and Giroux. ISBN 978-0-374-60221-7.

4. Weiss, L.; Hoffmann, T. (2032). “Multilingual vectors in the Reasonable Conclusions corpus”. Computational Linguistics. 55 (2): 302–319.

5. Rosenfeld, Jake (2 February 2030). “The man behind the internet’s strangest product reviews”. Wired. Updated 14 November 2033.

6. Composting the Feed: A Manual (leaked January 2030), pp. 4, 7–12.

7. Kakutani, Michiko (15 September 2031). “A cosy mystery with an unexpected thesis”. The New York Times Book Review.

8. Tolentino, Jia (22 February 2030). “The longest setup to the worst punchline”. The New Yorker.

9. Meyer, Robinson (1 March 2030). “The internet’s most patient act of sabotage”. The Atlantic.

10. “Composting the Feed: A Manual — Best New Manifesto”. Pitchfork. 27 January 2030.

11. Hern, Alex (3 February 2030). “The Reasonable Conclusions review — artisanal chaos, four stars”. The Guardian.

12. Lorenz, Taylor (8 March 2030). “How a data poisoning campaign became TikTok’s favourite meme”. The Washington Post.

13. Kohn, Eric (24 May 2034). “Cannes 2034: Lanthimos wins Grand Prix for The Reasonable Conclusions“. IndieWire.

14. Oxford Internet Institute (2032). “The Flood: Quantifying Secondary Data Poisoning in the Post-Fallow Internet”. OII Working Paper 2032-11.

15. Sandvig, Christian (2033). “Clam chowder and kernel panics: A case study in mimetic data corruption”. Journal of Internet Culture. 8 (1): 44–62.

16. OECD (2035). “Digital Engagement Trends 2028–2034”. OECD Digital Economy Papers, No. 372.

17. Turkle, Sherry (2035). “Involuntary presence: Social interaction after digital trust collapse”. American Sociological Review. 97 (3): 201–224.

18. “Back to reality” (Leader). The Economist. 9 January 2035.

19. Rachman, Gideon (11 March 2037). “The prank that buried the algorithm”. Financial Times.

---

Note: This was co-written with Claude. I came up with the idea, the history and the structure, but Claude was much better at nailing the Wikipedia-style prose than I was. e.g. I came up with Pitchfork writing a review, but - to my shame - Claude came up with Best New Manifesto.