Human Slop Has a Body Count. AI Slop Has Hurt Feelings.

Walk into any feed in 2026 and you’ll see the same lament: AI slop is ruining everything. The internet is drowning in synthetic mediocrity. Real writing is dying. Real thinking is dying. A robot wrote that email and you can tell.

Spare me.

Before we get sentimental about the supposedly golden age of pre-AI human output, let’s be honest about what humans have actually been producing for the last 40,000 years. It isn’t enlightenment. It isn’t civilization-as-success-story. It is slop — an ocean of it, a civilization-spanning torrent of confidently stated mediocre cognitive output that we have been politely calling “history” while the bodies pile up.

The idea that human thinking has been a net positive for the species is one of the most under-examined assumptions in public discourse. The receipts don’t support it.

The Original Slop Was on a Cave Wall

The cave paintings at Lascaux are beautiful. The other 99% of prehistoric drawings are smudges of ochre that look like a frustrated child’s tantrum. We remember the Sistine Chapel and forget the millions of devotional images cranked out by mediocre apprentices. We canonize Shakespeare and quietly forget that the average Elizabethan playwright wrote unwatchable trash that nobody bothers to revive. We talk about “great books” as if every book ever published deserved the trees.

Sturgeon’s Law — ninety percent of everything is crap — was not coined to describe AI. It was coined in 1957 to describe writing produced by humans, with their hands, using their brains. The mediocrity isn’t new. Only the volume is. And the people most upset about AI volume have been industrializing slop since Gutenberg.

Human Slop Has Killed Hundreds of Millions

If AI slop were merely annoying, we’d shrug. The real charge against it is that it crowds out signal, wastes attention, and corrodes decision-making. Fine. Now apply that exact charge to human cognition over the last century. The receipts are not a debate. They are a graveyard.

The 2008 financial crisis. Synthesized not by transformers, but by economists, bankers, and ratings analysts using the supposedly superior tool of human judgment to convince themselves risk had been mathematically dissolved. It hadn’t. The cost: nearly 10 million American families lost their homes to foreclosure between 2006 and 2014, $11 trillion in household wealth evaporated according to Federal Reserve data, 8.7 million jobs lost. The Federal Reserve Bank of San Francisco estimates the crisis cost every American roughly $70,000 in lost income over the decade that followed. Models had predicted the housing collapse. Humans in authority dismissed them. Exactly one bank was prosecuted — a small community lender. Every major engineer of the catastrophe walked.

Iraq, 2003. The world’s most expensive intelligence apparatus, staffed entirely by credentialed humans with PhDs and security clearances, confidently reported weapons that did not exist. Estimated deaths: 461,000 to over 940,000. Cost, per Joseph Stiglitz: $3 trillion. Dissenting analysts were sidelined. Confirmation bias and groupthink operated at the highest levels of government. The “intelligence community” — humans, all of them — produced one of the most expensive cognitive failures in modern history.

The opioid crisis. A single five-sentence letter to the New England Journal of Medicine in 1980 — written by humans, peer-reviewed by humans, published by humans — was cited 608 times over 37 years to justify mass prescription. 806,000 Americans are dead between 1999 and 2023.

These are not glitches. They are the steady-state output of unaugmented human cognition operating at scale. Every catastrophe above was authored by people who would have, at the time, told you they were the experts and you should defer to their judgment.

The Wetware Has Hard Limits

Psychologists have catalogued 188 distinct cognitive biases that systematically distort human decision-making. Confirmation bias. Anchoring. Hyperbolic discounting. Dunning-Kruger. Groupthink. Sunk-cost fallacy. Normalcy bias. The backfire effect, which makes correcting misinformation push believers deeper into it.

These aren’t quirks. They are architectural features of a brain that evolved for small-group survival on the African savanna, running firmware that hasn’t been meaningfully updated in roughly 200,000 years. The result is exactly what you’d expect from a Pleistocene processor running 21st-century workloads: wars, bad policy, motivated reasoning, and an industrial output of confidently stated nonsense.

Education does not fix this. Daniel Kahneman spent his entire career studying these biases and openly conceded he was no better at avoiding them than anyone else. Intelligence does not fix it — smart people are often more susceptible to motivated reasoning, because they’re better at constructing rationalizations. Credentials do not fix it. Tenure does not fix it. Peer review does not fix it. Institutions don’t fix it; they amplify it, by stacking career incentives and reputational risk on top of the underlying cognitive errors.

The human mind is not broken. It is running exactly as designed. The problem is that the design is for a world that no longer exists, and the people defending it have an obvious financial interest in pretending otherwise.

The Research Establishment Is the Slop Factory

Here is where the absurdity gets acute. The institutions screaming loudest about AI contamination of the knowledge base are the same institutions that, in any honest accounting, are the contamination of the knowledge base.

A meaningful share of academic papers are cited zero times. Most published research is forgotten because it deserves to be forgotten. The replication crisis is not a rumor — it is documented:

• A 2015 Reproducibility Project in psychology successfully replicated only 39 of 100 prominent studies. Sixty-one percent of the field’s flagship findings could not be reproduced.
• In cancer biology, an Amgen team attempted to replicate 53 landmark preclinical studies. They succeeded on 6. That’s an 89% failure rate in the literature underwriting actual cancer drug development.
• Medical research more broadly: John Ioannidis’s landmark 2005 paper, “Why Most Published Research Findings Are False,” argued exactly what its title says — and the subsequent two decades have largely vindicated him.

Behind these numbers is a quiet ecosystem of fraud and incompetence. Diederik Stapel fabricated data across more than 50 papers before he was caught. Brian Wansink’s food-behavior research empire collapsed when scrutiny revealed systematic data manipulation across his entire publication record. Marc Hauser. Hwang Woo-suk. Paolo Macchiarini. Each was a celebrated expert. Each was producing slop. Each was protected, for years, by the institutions that now want to lecture us about epistemic standards.

The system that produced this body of work is the same system that complains about AI lowering research quality. The premise of the complaint is that there is some pristine baseline being defiled. There isn’t. Paper mills now flood journals with thousands of fabricated articles. Predatory publishers extract rent from desperate academics. Retraction rates climb every year. “Publish or perish” has been the operating logic for fifty years, and the predictable result is volume, not knowledge.

This is not a system being threatened by AI. This is a system whose business model was already a slop conveyor belt — and whose practitioners now realize the conveyor belt is about to be automated, eliminating their economic moat.

That moat is artificial. Almost all of it. The credential, the peer review, the institutional affiliation — these are scarcity-based gatekeeping mechanisms, not quality controls. When you can’t reproduce more than half of your foundational studies, you don’t have a quality control system. You have a guild.

The loudest critics of AI in academia are people whose professional identity depends on being the certified producers of cognitive output, and who have nothing substantive to say about the actual outputs of AI systems because the substance comparison would not flatter them.

What AI Already Does Better Than Us

The “AI just produces slop” framing collapses on contact with the data.

Harvard Medical School’s CHIEF model analyzes tumor tissue with 94% accuracy across 11 cancer types, outperforming individual pathologists. A 2025 Nature Medicine study covering 463,094 women in Germany found AI-supported mammography increased breast cancer detection rates by 17.6% without raising false-positive recalls. GPT-4 variants have matched or beaten human experts at identifying pathology errors (89.5% vs. 88.5%), classifying skin lesions (84.8% vs. 84.6%), and staging ovarian cancer (97% AI vs. 88% radiologists). AI diagnostic systems could meaningfully address the 12+ million Americans subjected to diagnostic errors every year — errors made, every one of them, by credentialed humans.

It isn’t just medicine. AlphaFold solved the protein structure problem that had stumped human biology for fifty years, releasing predicted structures for over 200 million proteins — a body of knowledge that human researchers, working at the rate they had been working, would have produced over centuries if at all. Materials science is being accelerated by AI screening that compresses years of human trial-and-error into weeks. AI weather forecasting now beats traditional human-tuned numerical models on multiple horizons. Drug candidate identification is being compressed from years to months. A 2024 Purdue study found that an automated devil’s-advocate system improved decision quality by 33–34% simply by forcing consideration of opposing evidence. AI debiasing tools have already cut demographic bias in resume screening by up to 35%.

This is the inconvenient counter-data. The “slop machine” is, in the domains we actually care about, already outperforming the humans who are upset about it — humans who spent a career building credentials in those domains and whose credentials are now visibly less reliable than the model.

This is not what failure looks like. This is what augmentation looks like. AI raises the floor. It expands the share of humans who can produce competent diagnostic decisions, the way the calculator expanded the share of humans who could compute compound interest. It interrupts the bias loops that produced the body count above. It does work that human cognition, left to itself, has demonstrably failed to do — for centuries.

Who’s Really Mad About AI Slop

Notice who complains loudest.

It is rarely the people doing genuinely original work, because genuinely original work is, almost by definition, hard to automate.

The loudest critics are a type. The mid-tier consultant whose entire output is repackaging what the client already knew. The boilerplate copywriter producing the eighteenth blog post on “five tips for engagement.” The academic whose papers are cited zero times outside their own footnotes. The pundit whose takes are downstream of three other pundits. The MFA-credentialed essayist who has been mistaking workshop polish for thinking for twenty years. The “thought leader” whose thoughts are someone else’s thoughts with their face attached.

Their alarm is real. Their framing is convenient.

These are people whose professional moat is the ability to produce competent-but-unremarkable cognitive output at scale, and whose moat has just been filled in. They are not defending the integrity of human knowledge. They are defending their position in an attention economy that previously rewarded them for being slightly above the human median, while AI threatens to put a competent floor under everyone.

Calling something “AI slop” is a way of refusing the comparison. It treats the critique as settled by the source rather than by the substance. But a paragraph isn’t bad because a machine wrote it. A paragraph is bad because it’s vague, derivative, untrue, or boring — the same reasons human paragraphs have been bad since the invention of the paragraph. The people most fluent in that older, human flavor of slop have the most to lose if anyone makes the comparison honestly.

The academic version of this complaint is the most absurd. Researchers from a system whose foundational papers fail to replicate more than half the time, whose journals are flooded with paper-mill fabrications, whose retraction rate is climbing every year, whose top-cited “discoveries” routinely turn out to be artifacts of p-hacking — these people are now positioning themselves as the guardians of epistemic integrity against the machines. It would be funny if it weren’t so revealing.

The credential is the moat. The peer review is the moat. The institutional affiliation is the moat. None of them is, on the evidence, a quality signal. They are all economic mechanisms for restricting the supply of cognitive output and charging rent on access. AI threatens the rent. That is the actual complaint, dressed up in the language of standards.

The Absurdity Test

Look at the world built by unaugmented human cognition. Look at it carefully.

Two world wars. The Holocaust. Stalinism. The Khmer Rouge. Slavery on three continents. Every famine that wasn’t caused by weather. Every preventable pandemic. The Atlantic Triangle. The Bengal Famine. Mao’s Great Leap Forward. Nuclear arsenals built by physicists who confessed they had become Death. Tobacco companies that knew. Oil companies that knew. Pharmaceutical companies that knew. Bureaucracies that filed and watched. Lobotomies. Eugenics. Leaded gasoline. Trans fats. Thalidomide. The food pyramid. Every preventable disaster, executed by a human looking at clear evidence and choosing wrong.

This is the species we are being asked to defend against the cognitive contamination of large language models.

The argument that human thinking has been historically good for civilization is, at this point, a mystical position. It survives only because we keep grading the wetware on a curve where its failures don’t count as evidence against it and its rare successes are treated as the rule. Strip the rationalization away and what remains is a brain that was good at not being eaten by leopards and bad at almost everything we have asked it to do since.

If a manufacturer released a product with the failure rate of human cognition, the FTC would sue them. If a software vendor shipped a system with the bias profile of the human mind, no enterprise customer would buy it. We have accepted the wetware’s failures because we had no choice — except now, for the first time, we do.

The Real Question

It isn’t whether AI produces slop. It does. So does every cognitive system in history, including yours and mine.

The interesting question is whether AI raises the floor — whether it expands the share of humans who can produce competent work, the way the calculator expanded the share of humans who could do arithmetic. Whether it can interrupt the bias loops that produced the body count above. Whether it can take the millions of people stuck in the artificial-moat economy of credentialed mediocrity and let them produce something better, or at minimum get out of the way of the people who would have produced something better if cognitive labor weren’t being rationed by guilds.

The early evidence — in oncology, in pathology, in mammography, in protein folding, in materials science, in structured decision-making — points to yes.

The cavemen who drew the bad cave paintings still drew them. The clerks who copied error-ridden manuscripts still copied them. The bankers who packaged subprime mortgages still packaged them. The intelligence analysts who signed off on Iraqi WMDs still signed off. The researchers who p-hacked their findings still got tenure. The slop has always been the baseline. The question has always been what we manage to build on top of it.

If your worry about AI is that it will flood the world with mediocre cognitive output — friend, look around. The 188 biases were already in your skull when you got here. The opioid crisis, the housing collapse, the climate inaction, the wars, the unreplicable research, the wrong food pyramid, the lobotomies, the leaded gas, the trans fats, the eugenics movement — all of this was on the ledger before any model was trained. The flood started 40,000 years ago. We are standing in it. We have been standing in it the entire time.

The question isn’t whether we’ll keep producing slop. We will. The question is whether, for the first time in our species’ history, we finally have a tool that runs the pump in the other direction — and whether we will use it, or whether we will let the people whose paychecks depend on the slop economy talk us out of the only cognitive upgrade we are ever likely to get.

The choice is that simple. The defenders of the wetware status quo are not on the side of human flourishing. They are on the side of their own market position. Don’t let them dress it up as anything else.

Several data points in this piece — particularly on the 2008 crisis, the cognitive bias literature, the opioid ledger, and AI medical performance — are drawn from Brian James Curry’s “The Augmented Species: Why Human Cognition Has Failed and AI Is the Only Way Out” (Medium, 2026), which assembles the underlying sources in detail. Original references include the Federal Reserve and Federal Reserve Bank of San Francisco, the Lancet Countdown, the Potsdam Institute (Nature, 2024), Nature Medicine (2025), Joseph Stiglitz’s Iraq War cost estimates, the Open Science Collaboration’s 2015 Reproducibility Project, Begley & Ellis’s 2012 Nature paper on cancer biology reproducibility, and the published cognitive bias literature.

Brian James Curry is an Enterprise AI Strategist, Data and Marketing Scientist, and Applied Researcher. He is the founder of Vector1 Research, an independent research lab focused on marketing economics, causal systems, cognitive AI architecture, and the economic implications of intelligent systems. He also founded KC AI Lab, a data science community and incubator for applied machine learning education.