The Delusion of AI — Episode 2
AI does not only consume electricity and water. It also consumes the internet that made it possible.
Imagine taking a sharp, colorful photograph of a mountain landscape.
The sky is blue. The trees are detailed. The snow on the mountain has texture. The image feels alive.
Now place that photograph on an old office copy machine and press print.
The first copy looks acceptable.
But it is already slightly worse.
The colors are flatter. The edges are softer. Some details are missing.
Now take that copy, place it back on the glass, and copy it again.
Then copy the copy.
Then copy that copy.
After enough repetition, the mountain disappears.
What remains is not a landscape.
It is a gray, blurry, lifeless imitation of something that once had depth.
That is the photocopy paradox.
And a version of it is now unfolding across the internet.
In Episode 1 of The Delusion of AI, I wrote about the physical cost of artificial intelligence: electricity, water, data centers, cooling systems, land, and the hidden infrastructure behind the illusion of a weightless digital tool.
But there is another cost.
A quieter one.
AI is not only consuming power.
It is also consuming the internet itself.
More specifically, it is consuming the human-made knowledge system that kept the internet alive for nearly three decades.
And if we are not careful, the open web may slowly become a copy of a copy of a copy.
Larger than ever.
Faster than ever.
But less human than ever.
The old internet had a rough contract
For most of the internet’s life, there was a simple but imperfect agreement.
Humans created things.
Writers wrote essays.
Journalists investigated stories.
Programmers answered technical questions.
Cooks shared recipes.
Designers published tutorials.
Artists uploaded their work.
Bloggers explained niche topics that no large media company cared about.
Independent experts made the internet useful because they added human experience, judgment, memory, taste, and effort.
Search engines organized that work.
They did not create most of the value themselves. They found it, ranked it, and sent people toward it.
That traffic mattered.
A click was not just a click.
It was attention.
It was ad revenue.
It was a subscriber.
It was a customer.
It was proof that the work had value.
The creator gave the internet knowledge.
The internet gave the creator visibility.
This system was never perfect.
It had spam, clickbait, SEO manipulation, platform dependence, and low-quality content long before generative AI arrived.
But even with all its flaws, the basic exchange still existed:
Humans made the internet useful, and the internet sent humans back an audience.
Generative AI changes that exchange.
Now, when someone asks a question, the answer can appear directly in a chat box or at the top of a search page.
The user gets the summary.
The source gets ignored.
The creator’s work is absorbed, compressed, repackaged, and delivered without the user needing to visit the original page.
That is not a small design change.
It is a structural change in how attention moves online.
And when attention stops moving back to creators, the internet starts starving itself.
The answer economy has a hidden victim
From the user’s side, AI summaries feel obviously convenient.
Why open ten tabs when one answer can appear instantly?
Why read a full article when a model can summarize it?
Why watch a tutorial when AI can give the steps?
Why search through a forum when a chatbot can compress the discussion?
The convenience is real.
But the cost is also real.
The original creator may receive no visit.
No payment.
No subscriber.
No recognition.
No relationship with the reader.
The platform captures the interaction.
The AI interface captures the trust.
The creator becomes invisible.
This is one of the biggest problems facing writers, journalists, educators, researchers, publishers, and independent creators:
Their work may still be used.
But the path back to them is disappearing.
The internet was built on links.
AI is building an internet of answers without exits.
That sounds efficient from a user perspective.
But it is dangerous from an ecosystem perspective.
Because if enough people stop visiting original sources, the people who produce those sources lose the economic reason to keep producing them.
And if original work becomes less rewarding, less visible, and less sustainable, the open web does not become smarter.
It becomes thinner.
AI slop is filling the empty space
When original creators lose traffic, income, and motivation, many of them make a choice.
Some reduce output.
Some move behind paywalls.
Some shift to closed communities.
Some stop publishing openly.
Some leave completely.
But the internet does not simply become quiet.
The empty space gets filled.
Not always by better work.
Often by cheaper work.
This is where AI slop enters the picture.
AI slop is low-effort, mass-produced content created mostly to fill feeds, manipulate algorithms, attract cheap clicks, or imitate usefulness without real substance.
It appears as shallow articles, fake images, recycled motivational posts, generic comments, synthetic books, empty tutorials, spammy SEO pages, and videos that feel like they were made by nobody for nobody.
The problem is not that one bad article exists.
The problem is scale.
A human spammer has limits.
A machine can produce thousands of pieces of content before lunch.
That changes the texture of the internet.
Feeds become noisier.
Search results become thinner.
Comments become less trustworthy.
Images become harder to verify.
Articles become harder to distinguish from filler.
The internet becomes larger, but less alive.
More content.
Less meaning.
More language.
Less thought.
This is not just a content quality issue.
It is a trust issue.
When users no longer know whether they are reading a real human experience, a copied summary, a fake review, a bot comment, or an algorithmically generated article, the web loses one of its most important qualities:
Believability.
Humans are becoming harder to find online
There is another layer to this problem: automation.
Bots, scrapers, automated agents, and AI-driven systems now make up a massive part of internet activity.
Not every bot is bad.
Search engines use crawlers.
Security tools scan websites.
Companies automate useful processes.
Many systems depend on automation to function.
But the trend still matters.
When automated traffic grows faster than human activity, the internet becomes less like a public square and more like a machine environment where humans are only one type of user.
Bots scrape pages.
Bots post comments.
Bots generate fake engagement.
Bots attack websites.
Bots imitate real users.
Bots feed data into systems that generate more content for more bots to interact with.
This is why the “dead internet” idea has become emotionally powerful, even if the internet is not literally dead.
The fear is not that no humans are online.
The fear is that human presence becomes harder to identify, harder to trust, and harder to reward.
That is a serious cultural problem.
Because the internet is not valuable simply because it contains information.
It is valuable because real people contribute experience, disagreement, humor, pain, expertise, memory, taste, and originality.
Without that human layer, the web becomes a warehouse of synthetic noise.
Organized, searchable, endless noise.
But still noise.
When AI eats its own output
Now we reach the strangest part of the story.
AI systems need data.
Not just any data.
They need high-quality human-generated data: books, journalism, code, essays, conversations, research, art, documentation, and millions of examples of how humans think, write, explain, argue, solve, and create.
But what happens when the internet becomes heavily polluted with AI-generated content?
Future AI systems may train on the output of previous AI systems.
This is recursion.
A model learns from data produced by a model that learned from data produced by another model.
At first, the damage may be subtle.
Rare ideas become less visible.
Unusual writing styles get averaged out.
Minority viewpoints become easier to miss.
Niche expertise gets flattened.
Strange, messy, specific human details are replaced by smooth generalities.
Then the degradation grows.
The model becomes more confident but less grounded.
More fluent but less original.
More polished but less alive.
This is the core concern behind model collapse.
When generative models are repeatedly trained on synthetic data without enough fresh human input, they can gradually lose contact with the richness of the original human data.
In plain language:
If AI keeps learning from copies of copies, the signal decays.
The mountain becomes a smudge.
Human work is not outdated
This is the mistake many people make when they talk about AI replacing creators.
They think human creativity is just an outdated production method.
Slow.
Expensive.
Messy.
Emotional.
Inefficient.
But that messiness is exactly what makes human work valuable.
A human writer does not only arrange words.
A human writer notices things.
A journalist takes risks.
A researcher spends years developing judgment.
An artist carries memory, culture, pain, failure, and obsession into the work.
A programmer solves problems inside real constraints.
A teacher understands confusion because they have seen students struggle.
A founder understands pressure because they have had to make decisions without enough information.
A real person brings context.
AI can imitate the surface.
But the source material still comes from human life.
If we remove the economic and cultural reasons for humans to keep producing original work, we are not making AI stronger.
We are cutting off its food supply.
That is the hidden contradiction of the AI race:
The machine looks independent only because it is standing on a mountain of human work.
If that mountain stops growing, the machine eventually starts recycling itself.
The internet cannot survive on imitation alone
This is the second delusion of AI.
We look at AI tools generating essays, code, images, summaries, captions, strategies, and videos in seconds, and we think we have built an infinite engine of intelligence.
But maybe we have built something more fragile.
An engine that depends on the very creators it may weaken.
A system that extracts from the open web while reducing the reasons to contribute to it.
A tool that can produce endless language while slowly damaging the human ecosystem that gave language its meaning.
The danger is not that AI will immediately destroy creativity.
The danger is slower and more ordinary.
Creators lose traffic.
Publishers lose revenue.
Independent voices disappear.
The open web fills with slop.
AI systems train on synthetic leftovers.
Everything becomes smoother, faster, and emptier.
A copy of a copy of a copy.
Until the mountain is gone.
The question is not whether AI can generate content.
It clearly can.
The question is whether the internet can remain meaningful if original human contribution becomes less visible, less rewarded, and less trusted.
Because if we replace the writers, thinkers, journalists, artists, teachers, researchers, builders, and independent voices, what exactly will the machines learn from tomorrow?
Next in the series
In Episode 3, we leave the internet and look inward.
What happens to our own brains, our critical thinking, and our ability to solve problems when we remove too much mental friction from daily life?
Are you noticing more AI slop in your feeds, search results, or comment sections?
And more importantly:
Do you still feel like the internet is made by people?
View my website.