The West's "AI Takeover" is a Paper Tiger: Analysis

The capitalists have always paid their workers pennies rather than dollars; with AI, they hope to avoid having to pay them entirely—but their efforts will undoubtedly fail.

Credit: Getty Images

The rise of generative AI (GenAI) and large language models (LLMs) is well known, and their benefits have been documented extensively.

Perhaps most notably, they excel at summarization—which Jane M. Healy in her Endangered Minds: Why Children Don’t Think and What We Can Do About It identifies as one of the most difficult things for modern students to do (thereby contributing to illiteracy and anti-intellectualism among students). They also excel in drafting, code completion, translation, and brainstorming. From certain perspectives, they have significantly boosted office productivity.

This productivity, however, comes with two glaring costs: one ecological and one epistemological.

High water, energy consumption

GenAI and LLMs have negatively impacted the ecosphere through the high energy and freshwater use, and they have cost many creative labourers (e.g., writers, artists) their jobs.¹

In Texas, residents are told to conserve water while capitalist corporations operate limitlessly. A Newstarget article in 2025 called the high water use “staggering,” noting that based on the Texas Water Development Board’s projection of 49B gallons in 2025, the use “could balloon to 399 billion gallons by 2030.” The article states:

For context, that’s enough water to supply millions of households annually. Meanwhile, cities like San Antonio enforce Stage 3 water restrictions, limiting lawn watering to once a week and imposing surcharges on excessive residential use. The contradiction is impossible to ignore: while everyday Texans are told to cut back, corporate giants face no such limits.

This is not a universal inevitability of AI development. Open-source Chinese models have pursued mix-of-experts (MoE) architectures that reduce per-query energy and water costs significantly, while closed-source Western megacorporations race toward ever-larger, ever-thirstier monolithic models.²

The difference is not technical but ideological. Western capitalists equate “better” with “bigger” and “more autonomous,” because, to them, autonomy means eliminating labour costs entirely. Efficiency is not measured in joules or litres but in surplus value extracted per human replaced.

The Chinese approach, whatever its own flaws, at least acknowledges that optimization can mean less resource consumption, not more. But since that path offers no quarterly bump to shareholders, it is dismissed, because under Western capitalism, a model that saves water but preserves labour is a model that has failed its primary purpose.

Without human labour, GenAI cannot exist

The other “main” cost is the replacement of creative labourers (e.g., writers, illustrators, translators) whose work is scraped, ingested, and then rendered economically redundant by the very models that depend on them.

This is not “collateral damage,” but the very business model itself: remove the human, capture the surplus, and call it “efficiency.”

Yet, while GenAIs are able to “write,” they can only perform research using what is publicly available. Without human writers constantly outputting new samples, GenAIs would have nothing to pull from.

Research has shown that training LLMs on LLM-generated text causes model collapse—outputs become less diverse, more biased, and increasingly nonsensical; errors and noise accumulate quickly, and the model loses its connection to the original human distribution of language.

This has been proven numerous times, such as in this study by Bohacek and Farid and here by Shumailov et al.; see also this commentary on the latter by Ali Borji. This has been seen with so-called “AI art” as well, for instance, in this study by Alazaroale and Lukina.

Thus, human writers are not only useful but ecologically necessary for high-quality LLMs. If all human writing stopped today, future LLMs would quickly stagnate and decay. This dependency on human input is even more acute when we move from creative writing to factual reporting.

While one could theoretically envision fully autonomous drones, for example, flying to breaking news sites to identify and report on crises without putting journalists in harm’s way, drones and LLMs cannot reliably verify breaking news themselves.

Autonomous news-gathering is fundamentally unreliable. Misidentification is common—a drone could mistake a movie prop for a real casualty, for example—which would in turn lead to false reports. This issue is not merely an edge case but an issue inherent to pattern matching without understanding. On a similar note, GenAI confidently citing nonexistent studies to fill gaps is a well-known phenomenon.

While capitalist companies pour billions into so-called “autonomous agents” that will supposedly replace journalists, lawyers, and doctors, they have failed to consider that LLMs are unable to handle long-term planning, reliable fact-checking, or genuine reasoning; they can only simulate such concepts. Completely replacing human labourers with AI will quickly and only lead to errors, lawsuits, and reader distrust.

LinkedIn News editor Emma W. Thorne noted that “[some] major companies, like Ford, have even hired back humans after their AI efforts didn’t work out as planned.” This retreat is a microcosm; the labour was never truly surplus, merely deferred. The moment the models faltered, the human workers once again became necessary.

Contradiction of “autonomous” AI

The imagined “fully autonomous” future, where LLMs generate news, art, and code without fresh human input, is not merely technically dubious; it is physically impossible within the ecological limits we already exceed.

Every gain in benchmark performance today requires exponentially more water, energy, and curated human text. To sustain the current growth trajectory toward true autonomy, we would need to multiply data centres, cooling systems, and training runs several times over—all while droughts intensify, grids strain, and human writers, the very feedstock, are laid off and cease producing new material.

The irony is fatal: the faster capitalists chase replacement, the faster we exhaust both the human source and the planetary substrate required to keep the model alive.

Crucially, this is not a design flaw or oversight but by design. Under Western capitalism, human labour is not a partner but a cost to be eliminated. “Good” GenAI, by this logic, is one that ranks labourers as secondary—surplus value to be extracted, then discarded.

But the discard is literal: without new human writing, the models collapse; without water, they overheat; without energy, they halt.

The capitalist dream of so-called “autonomous” AI is a contradiction in terms—it requires the very planet and people it seeks to render obsolete. AI needs human writers to maintain quality as well as water and energy to operate. By the time we hypothetically arrive at that “autonomous” society, assuming the same AI growth rate as today, the earth would have already burnt up.

Thus, the West’s “AI takeover” is a paper tiger. While it appears powerful, its ecological limits and dependence on human labour make true “autonomy” wholly impossible. The pursuit of capitalist “efficiency” ensures AI will fail long before it could replace humanity.

---

---

1. AI has been used for ages in GPS, recommendation algorithms, fraud detection, medical imaging, etc., and is obviously not going away anytime soon. Thus, I focus here on GenAI and LLMs.

2. It is worth noting that China has not yet solved the issue of freshwater use, but due to good governance, enforced regulations, public opinion, etc., companies have been able to minimize energy use while simultaneously keeping models ahead of Western counterparts in terms of helpfulness and accuracy. These actions contribute to the Chinese nation’s modernization efforts and joint development of green technologies for the new era.