Reading · The Streetlight

In 1900, Kodak changed the world when it released the Brownie, a one‑dollar cardboard‑and‑wood box camera that was simple, durable, and used Kodak roll film.

The Brownie was a massive success, with more than 100,000 units sold in the first year alone, and it popularized the idea of the “snapshot.”

Before that, cameras were expensive and hard to use. Only professionals and a small number of serious hobbyists could afford them, and they controlled what most people ever saw.

“You press the button, we do the rest.” That was Kodak’s simple and radical promise. Overnight, ordinary workers could afford to freeze moments that had never been recorded before. What used to be a luxury skill became something almost anyone could do.

Around 1999, phone cameras arrived with tiny sensors, about 0.1 megapixels of resolution, and terrible image quality. At first, their blurry photos were almost a joke that professionals mocked. Then the iPhone turned billions of people into photographers.

By 2025, flagship phones had roughly 200‑megapixel sensors, and the processing on the phone quietly did the work that used to require a Kodak studio with several trained photographers.

Today, your phone will take multiple shots, blend them, fix the lighting, correct the color, and sharpen the image before you ever see it. You gave up manual control of aperture and shutter, and in return you got a competent photo nearly every time, from a device you already carry.

Professional photography did not vanish. But the market changed. At the top end, wedding, fashion, and commercial photographers, the ones hired for their vision and reliability, still got work. In the middle, where people once paid for simple portraits, stock photos, or basic event coverage, the floor collapsed. The phone was “good enough,” so budgets shrank or disappeared entirely.

Most important, the camera phone blew the doors open for two groups. It unleashed people with great artistic instincts who had no money for gear, and it gave everyone else the power to capture their lives. Along with that came an explosion of terrible photos. We get more genius and more garbage at the same time. That is the price we pay for access.

Traditionally, writing at a professional level required years of practice, fluent language skills, and often a lot of schooling. If you had a brilliant idea but clumsy words, you were stuck. You could hire a writer if you had money. If not, your idea probably died in your notebook.

For years, the best technology could offer was the ability to dictate to your device and use something like Grammarly to help fix your writing. Helpful if you already had some skill, but not enough to get most people’s ideas into coherent content.

Generative AI changed this. A large language model is trained on massive amounts of text so it can predict and generate fluent language. You type a prompt in plain speech, and the system gives you drafts, story ideas, outlines, code, marketing copy, essays, and more. Image generators work the same way for pictures.

You no longer have full control over every word or pixel, but you get speed and fluency on demand. This is the same basic tradeoff as the phone camera. You give up fine control and get broad access.

Generative artificial intelligence is doing to language and creativity what the camera phone did to photography. It’s taking something that used to require skill, money, and time and putting it into everyone’s hands.

In many ways that’s a good thing. But history tells us that whenever a tool is democratized, someone loses control, someone else gains it, and those in power race to bend the new system back toward their interests.

Generative AI is a reality and it is reaching more people with every passing day. It’s time we asked deeper questions than “is AI good or bad.”

We must ask: who does AI make more powerful, who does it make more vulnerable, and what are we going to do about it? Who gets to speak, who gets drowned out, who profits while the rest of us scroll through an endless feed of noise?

In a major study published in the journal Science Advances in 2024, researchers Anil Doshi and Oliver Hauser worked with 293 writers and 600 evaluators. Some writers got to ask a powerful language model for up to five story ideas before they wrote. The evaluators did not know which stories involved AI.

Writers with AI assistance produced stories that judges rated about 8 percent more novel and about 9 percent more useful.

For writers who scored low on an initial creativity test, story novelty went up around 10 percent and “well written” ratings jumped roughly 26 percent. People who used to be locked out by their weaker writing suddenly put work on the table that looked a lot like the work of “naturally creative” peers.

But there was a catch. The same paper measured how similar the stories were to each other. AI‑assisted stories were about 9 to 11 percent more similar than stories written without AI.

When many people leaned on the same system, their work started to sound alike. The authors called this a “social dilemma.” Individually, using AI is the smart move. Collectively, it flattens the creative landscape.

A 2025 study in Nature Human Behaviour found the same pattern in brainstorming. In one experiment, people designed a new toy with help from ChatGPT. Ideas poured out fast. But 94 percent of these AI‑assisted ideas overlapped. Nine different people even ended up naming their toy “Build‑a‑Breeze Castle.”

Across five experiments, AI help almost always meant less diversity of ideas.

Another psychology study in Frontiers in Psychology saw that AI can generate many creative‑sounding ideas quickly, but it tends to stick close to familiar patterns and struggles to separate truly original ideas from safe variations.

AI raises the creative floor for individuals, especially for those who start out struggling. At the same time, it pulls the collective output toward the same center, making our culture more homogeneous.

Humans, especially experts, are still better at spotting the rare ideas that actually change things.

The companies selling AI rarely talk about the other side. They want you to believe this is mostly about empowerment and productivity, but they underplay the flood of low‑value content that buries human work and pollutes the information ecosystem we all rely on.

“Slop,” Merriam‑Webster’s Word of the Year in 2025, has become shorthand for digital content that is generated mainly by AI, in bulk, with almost no editorial care or originality. It is cheap, fast, and everywhere.

By April 2025, roughly three out of four new web pages contained some AI‑generated text. In Google’s top twenty search results, AI‑written pages went from about 1 percent to nearly 19 percent in just a couple of years.

On YouTube, a study that simulated 500 new users found that about 21 percent of first‑recommended videos were fully AI‑generated slop channels. Another 33 percent were tagged as “brainrot” content designed purely to hijack your attention.

Researchers identified 278 such slop channels among the top 15,000 globally, with a combined 63 billion views, 221 million subscribers, and an estimated 117 million dollars in yearly revenue.

That’s real money flowing to faceless operators who hit “generate” and let the algorithm do the rest.

Before Gutenberg, copying books was slow, expensive, and tightly controlled. Monks and scribes spent months on a single manuscript. A single book could cost the equivalent of a year’s wages. Religious and political authorities effectively decided what knowledge circulated.

The printing press shattered that monopoly. Within about fifty years of its spread through Europe, book production grew roughly fivefold. Much of the new material was low quality: sensational pamphlets, crackpot medical claims, wild rumors.

Traditional scribes were furious. In one recorded incident, scribes in Paris physically attacked a printing press in 1476.

Yet the chaos opened space for the Reformation, the Scientific Revolution, and the kind of scientific and political debates that remade the modern world. Over time, new gatekeepers emerged: publishers, editors, censors, universities.

New tools like peer review and standardized spelling turned the wild flood into something more usable.

Generative AI is repeating the same script at internet speed. Instead of monks, we have editors, copywriters, and designers. Instead of pamphlets, we have feeds and recommendation systems.

The new gatekeepers are not scribes with ink on their hands but platform algorithms, corporate policy teams, and regulators. The question is whether they really understand how the systems actually work, and whose interests they are serving.

In 1957, science fiction writer Theodore Sturgeon coined Sturgeon’s Law: “90% of everything is crud.” Every creative field is mostly mediocre, with a small layer of brilliance on top.

Generative AI does not change that ratio. It multiplies the volume. When the cost of creating content approaches zero, you get more genius and far more garbage.

When AI systems are repeatedly trained on data that already contains a lot of AI‑generated content, this leads to what researchers call “model collapse.” Over successive training rounds, the model loses information about rare and unusual patterns. Those are the tails of the distribution, the edges where novelty lives. They effectively disappear and cannot easily be recovered.

If we let AI fill the internet with slop and then train future AI on that polluted data, we slowly lobotomize the very tools we are told will drive the next century of progress.

Even platforms that have happily pushed engagement‑bait for years, like YouTube, are suddenly talking about cracking down on slop. They are not doing this out of love for truth. They are doing it because their own models and recommendation systems start to rot if they feed too heavily on machine‑generated junk.

While so‑called AI leaders misdirect our attention with promises of Artificial General Intelligence and Artificial Super Intelligence, and others fear‑monger about Skynet and Armageddon, we keep drowning in slop.

At first, generative AI’s output was laughable to professionals. Now it’s “good enough” for huge numbers of tasks, especially at the low and middle ends of content work.

This is what Clayton Christensen called disruptive innovation. A product starts at the bottom: cheaper, worse, easy to ignore. Over time it gets better until it pushes out older products entirely.

Once that happens, the Jevons paradox takes over. The more efficient a process becomes, the more total demand and usage go up.

When AI lets one worker produce ten times as much text or imagery, the world does not settle for the old amount produced more efficiently. It orders ten times as much content.

Even if the amount of high‑quality content grows, it gets harder to find amid the flood. Nassim Taleb describes this as the signal‑to‑noise problem. AI, by multiplying output, magnifies this problem. More data usually means more noise.

Like every innovation that creates new value, AI also destroys some existing jobs, firms, and ways of life. This is the process of creative destruction.

Meanwhile, more options make people less satisfied and more anxious. Using AI to generate fifty versions of a blog post or hundreds of images with a few prompts leaves you more overwhelmed, and the value of trusted filters and curators goes up.

Our time and focus are finite. Platforms are tuned to maximize engagement, not truth or meaning. AI generates content that is perfectly designed to feed that machine.

All the while, energy experts warn that powering all this could push AI‑related data center electricity use toward twenty percent of US power demand by 2030, not to mention the massive consumption of precious drinking water.

The marketing story says AI “levels the playing field.” That is half true. For some people it really does. For others, it tilts the field even more steeply.

Generative AI is a lifeline for people who have strong ideas but weak language or design skills. People who start out with lower creativity scores benefit most, sometimes producing work that looks as strong as that of more naturally creative peers.

A scientist in Cairo or Lagos who struggles to write in English can use AI to draft clear, polished papers and grant proposals. A small business owner can produce credible marketing materials without hiring an agency.

Consulting firms like McKinsey estimate that generative AI could add between about 2.6 and 4.4 trillion US dollars to the global economy each year. Much of that potential lies in small and medium‑sized organizations that can now automate tasks and punch above their weight.

But the people most exposed to the downside are not abstract “workers.” They are freelance writers, designers, video editors, translators, and photographers whose work has been turned into training data and then used to undercut them.

A study by researchers at Washington University in St. Louis looked at real data from freelance platforms before and after the release of major AI tools.

For writers, job volume dropped about 2 percent and earnings dropped about 5.2 percent. For image freelancers, job volume fell about 3.7 percent and income fell about 9.4 percent.

That hurt everyone, but top performers were hit hardest. For each extra 1 percent of previous earnings, freelancers saw an additional 0.5 percent drop in jobs and a 1.7 percent drop in income.

A 2026 survey by the American Society of Journalists and Authors found that about 40 percent of freelancers said AI had cut their income. At the same time, roughly a quarter of freelancers who were advanced users of AI said their income had gone up. Around three quarters expected AI to reduce opportunities over time.

AI is not just replacing workers, it’s sorting them into those who can wield it as leverage and those who are left behind.

A 2025 field study published in the Journal of Applied Psychology followed 250 employees using AI at work. The study found that AI improved creativity and performance only for people who used what psychologists call metacognitive strategies.

Metacognition means being aware of how you think, planning your approach, monitoring your progress, and adjusting when something is off.

Workers who consciously decided when and how to use AI, and who critiqued its output, got better. Workers who used AI passively, copying outputs without much thought, did not improve and sometimes did worse.

People in lower‑income countries face unreliable power, weak connectivity, little training, and models that do not speak their languages well.

By 2025, one major global survey estimated that roughly 16 percent of people worldwide had used generative AI at least once, but adoption in the Global North was much higher than in the Global South.

So even in this “democratized” world, there are hidden filters. Simply giving people AI tools does not automatically empower them.

Without serious effort, AI will deepen, not close, global inequality.

While companies race ahead, lawmakers are scrambling to catch up.

The European Union has passed a wide‑ranging AI law called the EU AI Act. It sorts AI systems by their risk and imposes obligations accordingly. One section, Article 50, will require AI‑generated content to be labeled in a way machines can read starting 2 August 2026.

That will not magically fix the flood, but it creates the legal basis for platforms to detect and disclose when content is machine‑made. The law allows for fines as high as 3 percent of a company’s global revenue, which is large enough to get the attention of tech giants.

Europe is also drafting a Code of Practice on how to mark and label AI‑generated content.

In the United States, individual states are moving on their own. California’s AI Transparency Act, SB 942, will require large platforms to watermark or otherwise label AI‑generated content starting in 2026, with core provisions taking effect on January 1, 2026.

Another California law, SB 53, already in effect, forces companies that train very large models to publish transparency reports and safety plans. Colorado’s AI Act, which will be enforced from June 30, 2026, requires organizations to assess the impact of high‑risk AI on things like hiring, housing, healthcare, and education.

A federal Executive Order signed at the end of 2025 told the US Department of Commerce to review state AI laws and flag any that are overly restrictive or conflicting. Around the same time, the Federal Trade Commission is expected to issue its own view on how federal law interacts with state AI rules.

These moves could be used to challenge or harmonize state laws, shaping how much power local communities have to set their own AI boundaries.

The C2PA standard, created by a coalition of tech and media organizations, defines a way to embed “content credentials” in files. These are cryptographic tags that record how and where a piece of content was created or edited.

The risk is that these tools can be used either to protect the public from deepfakes and slop or to create new ways to track, control, and monetize creators.

It is easy to make predictions about where generative AI is going. It is extremely hard to know which predictions will materialize.

S&P Global calls early 2026 one of the hottest periods for AI funding and acquisitions. Anthropic launched Claude Cowork and Claude Code, allowing AI to be integrated into team workflows. OpenAI rolled out a cheaper tier called ChatGPT Go and is testing ads in its free and low‑cost versions, dropping the price of access even further.

Netflix bought Ben Affleck’s AI film‑tech company InterPositive, pointing toward AI‑driven film production, not just marketing trailers. Oracle is planning deep job cuts, in the tens of thousands, as it pours money into AI data centers. Broadcom is publicly challenging Nvidia in AI chip forecasts by projecting more than 100 billion US dollars in AI chip sales by 2027.

We are likely to see more new tools impacting the cost of creation. The world gets flooded with content, most of it mediocre or worse. Old gatekeepers continue to panic and try to hold on.

People look for new filters, rules, and institutions. Some are legal, like the EU AI Act or state transparency laws. Some are technical, like watermarking and content provenance standards. Some are social, like norms around disclosing AI use.

A new kind of gatekeeping emerges to help us cope with the flood of content as the internet breaks apart into more isolated fractals and communities.

We get used to the new norm as these tools vanish into the background. Everyone uses AI as part of their work, and the barriers between what is real or not no longer hold. Tools for finding and verification arise as a new category of protection that comes with its own issues and challenges.

Trust goes through a period of collapse before we get the world back into some semblance of equilibrium.

Our battle was never with technology. AI is a brilliant technology when used in the right way, for the right reasons, and in a way that adds value to all stakeholders.

The question is whether we, the people, will fight for our vision of the world to dominate. To amplify human creativity and judgment. To build tech that serves us, that adds value to our lives, and to refuse to let a handful of platforms drown the world in slop, destroy our natural resources for profit, and call it progress.

Slopnet Is the Real Skynet

Comments