Stop blaming AI for human slop

AI slop is real. But the machine didn’t pick the topic, skip the fact-check, or hit publish. A person did. However, the loudest AI-haters have started punishing the wrong thing.

LinkedIn’s rolling out a button that says “Seems like AI slop.” I want that button. Badly!

I use AI every day, for hours, and I still cringe seeing my feed filled with the same immaculate airport encounter, the same fourteen one-line paragraphs, the same leadership lesson nobody asked for, closing on a question that reads less like curiosity and more like a hostage negotiation for engagement.

So yes. Some of it has earned the button. But the thing is, the phrase is escaping its box — like every other latest SOTA model escaping its sandbox.

“AI slop” started out describing low-effort, mass-produced, vaguely human-shaped content. Now it’s turning into something else entirely: AI touched this, therefore it’s fake, and the person is a fraud.

Even LinkedIn’s own chief product officer drew that line while announcing the feature: “AI and slop are not the same thing.” He’s right. But most of the people mashing his button won’t care.

Where I come from, the old advice is to hate the bad deed, not the bad doer. The internet has pulled off something genuinely impressive here. It hates neither. It hates the hammer.

Don’t hate AI. Hate the human sloppers.

That’s the whole thesis. What’s drowning your feed isn’t AI slop. It’s ‘human slop’ — a topic chosen by a person, generated on command, skimmed for thirty seconds, and shoved live by someone who wanted the reach more than they wanted to be right.

Yes, the Flood Is Real. No, I’m Not Defending It.

I ain’t here to defend the flood. Generative AI made generic output cheap, fast, and infinitely copyable. The market sold instant expertise. Platforms paid out for volume. People took first drafts from a machine and shipped them as finished work.

Articles nobody checked. Images nobody art-directed. Code nobody understood. Videos with no head or tail.

And if you paste a prompt into a chatbot, copy the first 1,200 words, and publish them under your name as expertise, you didn’t adopt anything early. You outsourced the one thing the reader actually showed up for — your judgment.

Using AI heavily and hating AI slop isn’t a contradiction. It’s the same as owning a camera and hating bad photography. Or having an email address and hating spam.

The tool isn’t the standard. The person holding it is.

We’ve Done This Before. We Just Blamed the Right People.

Photoshop was sorcery in the nineties. Sorry, Adobe — it looks quaint now, but at the time it may as well have been witchcraft. And while most of us were wrestling a mouse to draw a lopsided cat in MS Paint, a small group of nerds were bending and blending reality into composites that looked photographed.

Not everyone was good at it. Plenty of people (ahem) bought the same software, produced genuine garbage, and were fully convinced they’d made a Picasso. MS Word didn’t turn anybody into a novelist either. It turned a lot of people into people who owned a word processor.

Then came the spam. Photoshop spam. Word spam. Email spam — the delivery van for both. And the software companies were right there the whole time, selling every one of those tools as the must-have shortcut to instant success and unstoppable growth.

Any of this ringing a bell? LLMs? Autonomous agents? Magic pills? Anyone?

Here’s what we didn’t do in 1997, though. We didn’t put Photoshop on trial (pun intended). Nobody built a ‘seems like Photoshop slop’ button. We blamed the guy with the airbrushed unicorn and the seventeen lens flares. We blamed the spammer. The tool sat there, morally inert, the way tools do.

Somewhere between then and now, we lost the plot and started prosecuting the software instead.

The Companies Selling You Sentience Made This Confusing on Purpose

To be fair to the shamers, half this confusion was manufactured — and not by them.

AI companies market their products as assistants, copilots, agents, coworkers, creators, digital employees. The interface says “I.” The demos compress hours of setup, selection, correction, and cleanup into one sparkling command. They’re selling you something supposedly alive.

Ask the model itself, and it’ll usually tell you the truth. It’s a large language model. It doesn’t feel things. It isn’t alive.

The machine is more honest than the people who built it. Sit with that one for a second.

So one side sells personhood and the other side prosecutes it, while the actual humans in the middle — the ones who picked the goal, set the quota, and approved the output — quietly leave the room as everybody argues about whether matrix multiplication has a soul.

None of which makes the vendors innocent. A car is a tool. That doesn’t excuse a manufacturer who hides faulty brakes. NIST’s AI Risk Management Framework is aimed squarely at the organizations designing, deploying, and using these systems, because responsibility doesn’t evaporate at the prompt box. Vendors own their training data, their safeguards, and every word of their marketing.

They just don’t own your publish button.

AI Shaming Is Manufacturing the Dishonesty It Claims to Hate

Now the part that’s going to cost me followers.

‘AI shaming’ — the reflex to treat any disclosed AI use as a confession of fraud — isn’t quality control. It’s a costume. It looks like ethics and it functions like a purity test.

Start with the detection lab the internet built for itself. Em dash? AI. Clean structure? AI. A sentence shaped like “It’s not X, it’s Y”? Straight to AI jail.

If your detector is a punctuation mark, a tidy triad, and the phrase “here’s the thing,” congratulations. You’ve built a horoscope for writers.

The real software isn’t doing much better. A 2023 study in Patterns ran seven GPT detectors across 91 TOEFL essays written by non-native English speakers. Average false-positive rate: 61.3 percent. Human-written essays, flagged as machine output, six times out of ten. A 2026 paper in the Journal of Higher Education Policy and Management pushes further, arguing detector scores can’t be independently verified in real cases and shouldn’t be used to prove misconduct at all.

Seven tools, one study, three years ago — not a permanent law of nature. Fine. But it still means the accusation lands hardest on people writing in their second language, which is a hell of a thing to be casually confident about from a comment box.

AI detection is a clue. It’s never once been proof of authorship.

Labels were supposed to solve this, and the good ones genuinely try. YouTube asks creators to flag realistic synthetic content and openly exempts minor edits. Meta adds “AI info” context and generally leaves the post standing. The open C2PA standard treats provenance like a nutrition label — where this came from, what changed along the way.

Context. That’s the job. Not a verdict. “Made with AI” doesn’t mean false, and “No AI” has never once meant true.

Here’s the consequence nobody in this crowd wants to own. When disclosure reliably gets you publicly humiliated, people stop disclosing. That isn’t a moral failure on their part. That’s arithmetic.

So the shaming crowd took an ecosystem where people were starting to volunteer how their work got made, bolted a social penalty onto it, and built a world where everybody hides it instead.

You wanted transparency. You made honesty expensive. Nice work.

A No-AI Rule Is a Boundary. It Isn’t a Moral Rank.

Let me be clear about what I’m not saying, because somebody’s already typing.

If you hire a hand-lettering artist and they quietly hand you generated lettering, the problem isn’t your fear of technology. It’s the lie. If a documentary commissions an image of a real event and gets a photorealistic invention, that’s not a creative shortcut — that’s false evidence. If a university wants to test whether a student can reason unaided, it should lock that assessment down and mean it.

Legitimate boundaries, all of them. And universities are already drawing them with more nuance than your average comment section. Sydney separates secure assessments from open ones where AI is allowed with acknowledgment. Canberra sorts tasks into permitted, guided, and restricted. UWA allows it only where the coordinator explicitly says so.

Different answers, same grown-up move. Define the boundary for the task. Say what has to be disclosed. Test unaided skill, where unaided skill is the actual point.

That’s a policy. “Real writers don’t use AI” is a bumper sticker.

And to the purists specifically, I have a dare. Take whichever model you like and produce something genuinely good in one shot. One prompt, no steering, no revision, ship it. You’ll get the exact generic mush you’ve spent six months complaining about — because directing these things well is a skill, and you don’t have it yet. That’s not an insult. It’s just the part of your argument you needed to be false.

Painters made their peace with digital artists. Publishers embraced eBooks. Musicians took the VSTs and never looked back. Handmade paintings still sell for absurd money. Vinyl turned into a taste symbol.

Nobody’s craft died. So what exactly is the holdup with one more machine?

Four Questions That Beat Shouting “AI” at a Page

Stop asking whether AI touched the work. It’s the least informative question available to you.

Ask what the person actually brought — reporting, experience, taste, a real problem worth solving, or a prompt that said, “Make this go viral.” Ask whether they were allowed to use it, because the contract, the assignment, and the platform policy all get a vote. Ask what they hid: faked experience, invented citations, borrowed identity, a process they promised to perform and skipped. Ask what they verified — did they open the sources, run the code, check the claims, think for ten seconds about who gets hurt if it’s wrong.

Then ask the one that settles it. When this breaks, do they stand behind it or point at the model?

Permission, representation, verification, ownership. That test catches the lazy booster who wants credit without responsibility. It also catches the purist who thinks typing every word by hand is a guarantee of truth. It isn’t. Manual work can be plagiarized, misleading, incompetent, or dull enough to make a loading screen feel emotionally complex.

Method matters. It just doesn’t deliver the verdict on its own.

I’ve argued before that offloading too much of your thinking to AI hollows out the skill you meant to sharpen and that research, testing, and lived experience must never get blurred together. Both still stand. Defending AI users doesn’t require pretending the risks are imaginary. It requires putting the blame where the decisions got made.

The Machine Didn’t Lower the Standard

AI makes skilled people faster. It makes lazy people louder. That’s the entire story, and it’s the one nobody’s monetizing.

The model doesn’t absolve the user. The user doesn’t absolve the company. Responsibility can be shared, split, and argued over — it just can’t be handed to a noun and left there.

I don’t need human fingers on every keystroke. I need human judgment at the center and a comment section willing to go find it instead of yelling at the software.

So next time something hollow lands in your feed, hold off on the button for one second and ask who approved it.

The machine generated it. A person signed their name to it.