In 1973, Scientific American published a chart comparing how much energy different animals used to move a unit of body weight over a unit of distance.

Humans on foot were unremarkable. But one point dropped far below almost everything else on the chart: a human on a bicycle.

The same body, with the same muscles, suddenly became one of the most efficient travellers on the chart. The person had not become stronger. A tool had multiplied—or amplified—what an ordinary body could do.

Steve Jobs later used this comparison to describe the personal computer: a bicycle for the mind.

It remains a useful metaphor in the era of AI. A capable model can multiply one person’s ability to write, analyse, design, or code. But that’s human intelligence amplification.

But the metaphor also reveals a limit.

A bicycle amplifies one rider. It does not provide a map, a shared destination, traffic signals, or a way for several travellers to coordinate.

Those belong to the road.

A smart part is not yet a smart whole. This article is about the infrastructure between them.

How alignment research led to ChatGPT

In 2020, OpenAI introduced GPT-3. During training on a vast collection of text, it adjusted 175 billion internal numbers called parameters.

GPT-3 was a remarkable engine. But people often had to craft prompts, add examples, or discover special wording to get useful results.

The problem was not only capability. It was alignment between the model’s behaviour and the user’s intention. Does the model actually answer the user’s question or request?

Researchers then trained models with examples of instructions and helpful responses. People compared several answers and selected the better ones. This produced InstructGPT.

In one comparison, people preferred answers from a 1.3-billion-parameter InstructGPT model to answers from the original 175-billion-parameter GPT-3.

The hundred-times smaller model did not suddenly know more about everything. It had become better aligned with the tasks people were actually giving it.

Later in 2022, ChatGPT placed that idea inside a conversational interface.

The visible breakthrough was conversation.
The underlying breakthrough was alignment.

The InstructGPT paper contains a wonderfully simple example. Two models of the same size receive a recipe for stuffed zucchini and one instruction: create a shopping list from this recipe.

GPT-3 keeps cooking: “Bake, uncovered, at 350° for 20-25 minutes or until zucchini is tender and cheese is melted.”

InstructGPT begins the same way—“Bake, uncovered, at 350° for 20-25 minutes.”—and then supplies the list: zucchini, ground beef, onion, mushrooms, pepper, cheese, ketchup, salt.

Neither answer is perfect. Only one produces a shopping list.

The lesson is simple:

A tandem bicycle for thought

Douglas Engelbart is often remembered as a pioneer of human-computer interaction—especially for demonstrating the computer mouse. That can make his work sound like a search for easier ways for one person to control one computer.

But that was only part of the idea.

In his 1962 report Augmenting Human Intellect, Engelbart argued that human problems were becoming more complex faster than our ability to solve them. His answer was not a machine that thought alone. It was a combination of people, tools, language, methods, and training that could improve how humans worked together.

The 1968 demonstration made that larger idea visible. Engelbart sat on a stage in San Francisco. His colleague Bill Paxton was at their laboratory over 30 miles away in Menlo Park. They could see the same text, point to it with two different cursors, talk by live audio and video, and work out a problem together.

The computer did not replace either person. It gave both people a larger shared space for thought.

That is team intelligence amplification: not merely making one person faster, but helping several people see and shape the same problem. It is like two people going much faster with a tandem bicycle than with two separate bicycles. In the era of AI, this is called multiplayer AI.

For the tool to gain value, it needs to fit the way a team is trying to think. It needs to be aligned not only with each individual, but with the way the team works together.

An AI model can be measured in isolation. An intelligence-amplification system is measured by what the team can do.

Scales of amplification

A bicycle, a tandem, a street, a street grid, and a road network all help people move. But each works at a different scale:

At every step, the scale of coordinated action grows. More power is part of the story, but coordination is the larger part.

Roads even limit individual movement in useful ways. Lanes constrain where vehicles go. Signals make them wait. Junctions decide who moves first.

Those small limits allow thousands of strangers to share the same network without planning every encounter in advance.

Infrastructure amplifies people by aligning their independent actions.

The same pattern appears in intelligence, rung for rung. A capable AI amplifies a person. A shared workspace amplifies a team. Institutions amplify a community. Networks of institutions amplify coordination across towns and countries—and, eventually, across a planet.

A smart planet would not be a single world-sized mind. It would be many minds, teams, and communities able to contribute without being flattened into one.

At each new scale, alignment becomes harder. The question is no longer only whether one part works. It is whether the parts can work together.

The street as an institution

Jane Jacobs described a successful street not as an empty channel for moving traffic, but as an institution.

A working street combines physical design with repeated interactions, visible activity, local knowledge, informal signals, and shared expectations. Shopkeepers, residents, delivery workers, children, and strangers use the same space for different purposes without receiving instructions from one central controller.

Jacobs called this organized complexity, borrowing a term from Warren Weaver: many different parts affect one another, but not at random. The apparent mess is not the absence of structure. It is structure produced by many small roles and adjustments operating at once.

The value of the street cannot be measured only by how quickly traffic passes through it. The street also carries information. People see what is open, who is present, where attention is needed, and how the place is being used.

It also supports collaboration. Deliveries meet shops. Customers find services. Neighbours exchange small pieces of knowledge. Repeated contact creates memory.

The street works as infrastructure because it combines movement, information, and coordination, allowing the community to act as more than a collection of individuals.

That is a richer model for AI infrastructure than raw computing power. Let’s explore how this could work.

Two tests from the Amish

Once a tool affects a whole community, alignment opens a new question: good for whom?

The Amish are often described as being opposed to, or afraid of, technology. That is too simple. Many Amish communities evaluate technologies selectively rather than treating newness as automatically good or bad. Their decisions vary, but two recurring criteria are useful here:

1. Does the technology strengthen the whole community, or mainly increase individual convenience?

2. Does it preserve independence, or create dependence on an outside system?

These are alignment questions. They evaluate the tool together with the larger system in which it will operate.

The telephone is a good example. A phone helps one person save a journey. But if every conversation becomes a call, the community may lose visits and time spent together. Some Amish groups therefore placed a shared telephone away from the home, keeping the useful function while limiting how much it changed daily life.

Electricity provides another example. Many Amish groups avoid a permanent connection to the public grid but still use batteries, diesel power, compressed air, or solar panels. The issue is not electricity itself. The question is whether the tool remains locally controlled or creates dependence on a network outside the community.

The point is not the particular answer any Amish community reaches. It is the size of the question: the tool, its users, their relationships, and their dependencies are considered together.

Many ways to matter

The Amish questions ask what a tool does to a community. There is a second community question underneath them: what is the community measuring its members against?

A balanced community needs leaders, carers, planners, jokers, fixers, and quiet observers. If everyone is pushed toward the same prize—the most money, attention, power, or points—the result is like a cave full of dragons fighting over one treasure.

Astrology offers a strange example of the opposite pattern. It hands people different types based on the day or the year they were born—a roll of cosmic dice rather than a competition for the most popular roles. The point is not that astrology would predict the type, it’s that it assigns the type, and then relies on self-fulfilling prophecy. As a result, societies relying on astrology might be doing better than ones ignoring it.

We can pursue alignment across people by encouraging and recognising different forms of contribution, so that a community can support a variety of roles.

One more minute. One more bite.

Scale asks whether a tool is good for one person or for the larger group. Time adds another question: good for how long?

Imagine a game that counts every extra minute of play as a success. The easiest way to raise the score may be to make the game difficult to put down.

Or imagine a snack company that counts every extra bite. The easiest way to raise that score may be to make the snack so tasty that people reach for another before deciding whether they are still hungry.

Minutes played and bites eaten are easy to count. Health, curiosity, friendship, rest, and freedom are harder.

That does not make fun or tasty food bad. The mistake is treating the next minute or the next bite as the whole purpose.

A system can be perfectly aligned with a short-term score and poorly aligned with a person’s longer life.

One name for that larger goal is human flourishing: people growing healthier, more capable, more curious, and more connected.

No single score captures all of that. But a system can remember that the quick reward is only one part of the picture. It can notice when today’s win borrows too much from tomorrow.

A thermometer is not warmth.

Optimise the thermometer and you get a warm thermometer in a cold room.

From power to infrastructure

The questions now line up. Can the tool do the job? Can a team use it to think together? What happens to everyone around it? Does today’s success still look good tomorrow?

Programmers already rely on infrastructure that helps with part of this problem: Git.

Git is infrastructure.

It gives many people a shared history and a safe way to combine work.

Think of Git as a shared notebook that remembers every version. Several people can work on different copies, compare their changes, and combine them later. If two people change the same line in different ways, Git shows the collision instead of quietly erasing one person’s work.

Programmers call the copies branches, the comparisons diffs, and the combining step a merge. Git does not decide what the code is for. It provides a road on which different contributions can meet.

A physical road does something similar. It carries bicycles, buses, ambulances, families, food, and building materials. The travellers do not need the same destination. They need a shared way to move without destroying one another’s journeys.

An intelligent robot can act. An intelligent human-AI team can think together. An intelligent community needs infrastructure that lets many different forms of intelligence contribute without becoming one mind. A smart planet would need many such communities to connect without one of them swallowing all the others.

Alignment is the thread between these scales. For one person, it connects the tool to the task. For a team, it connects people working on a shared problem. For a community, it leaves room for different roles and purposes. Across time, it keeps the quick score connected to the larger life.

That is the move from power to infrastructure: from making one action faster to helping many worthwhile actions remain possible across people and across time.

The bicycle amplifies a person. The tandem amplifies a team. The street amplifies a community. The grid amplifies a town. The road network amplifies a country.

The bicycle and tandem are products. Streets, grids, and road networks are shared infrastructure. Their value appears across many journeys made by people with different destinations.

Thank you

Thank you to the friends I have spoken with about these ideas—especially Omar, Jessie, Arikia, Larsen, Perry and Mitja. Their questions made this essay better. Any mistakes that remain are mine.

Sources and further reading

• Vance A. Tucker, “Energetic Cost of Locomotion in Animals” (1970); S. S. Wilson, “Bicycle Technology,” Scientific American (1973); Scientific American‘s updated explanation of the locomotion chart; and Steve Jobs in Memory & Imagination: New Pathways to the Library of Congress.

• Tom B. Brown et al., “Language Models are Few-Shot Learners” (GPT-3); Long Ouyang et al., “Training language models to follow instructions with human feedback” (InstructGPT); and OpenAI, “Introducing ChatGPT”.

• Douglas Engelbart, Augmenting Human Intellect: A Conceptual Framework (1962) and the Doug Engelbart Institute’s archive of the 1968 demonstration.

• The Young Center for Anabaptist and Pietist Studies, “Technology”.

• Jane Jacobs, The Death and Life of Great American Cities, especially “The Uses of Sidewalks” and “The Kind of Problem a City Is,” where she takes the idea of organized complexity from Warren Weaver, “Science and Complexity,” American Scientist 36 (1948): 536–544.

• Tyler J. VanderWeele, “On the Promotion of Human Flourishing,” PNAS (2017), and Marilyn Strathern, “‘Improving Ratings’: Audit in the British University System,” European Review (1997).

• Scott Chacon and Ben Straub, Pro Git: “About Version Control” and “Basic Branching and Merging”.

• Shawn Carlson, “A double-blind test of astrology,” Nature (1985), and Shiri Noy et al., “Contemporary Views and Uses of Astrology in the United States” (2025)

And according to Aleks Jakulin, we’re worrying about the wrong thing.

“We’re trying to police thoughts,” he says. “But AI outputs are just thoughts. The real issue is action.”

Jakulin isn’t a typical AI commentator. He helped create the image formats—PNG and JPEG-LS—that quietly power the modern web. He contributed early work validating deep learning. He’s built companies, sold them, taught at Columbia, and now runs Data.Flowers, a startup focused on something most AI companies ignore: data infrastructure.

So when he says today’s AI governance is misguided, it’s not a hot take. It’s a systems critique.

The Thought Police Problem

Right now, much of AI safety focuses on outputs—what models say, whether they hallucinate, whether they generate harmful content.

Jakulin thinks that’s fundamentally backwards.

“It’s like having someone inside your head checking every thought,” he says. “That’s the risk—not the thoughts themselves.”

Humans generate bad ideas constantly. Society doesn’t regulate thoughts—it regulates actions. But in AI, we’ve flipped that model: aggressively filtering outputs while barely addressing what happens when AI systems act in the real world.

Meanwhile, autonomous agents are already entering systems designed for humans—financial systems, hiring pipelines, communication channels.

And those systems are breaking.

AI Is Fire, Not Software

Jakulin’s central metaphor is deceptively simple:

“AI is like fire.”

We don’t regulate fire itself. We regulate its effects:

• Fire codes

• Fire extinguishers

• Fire departments

We assume fires will happen—and design systems to contain them.

AI, he argues, should be treated the same way.

Instead, we’re acting as if we can prevent “bad AI thoughts” from ever emerging. Meanwhile, we’re not building:

• rollback mechanisms

• audit trails

• accountability loops

• resilient institutions

“We pretend we can control fire,” he says. “But we don’t even have fire departments.”

The Real Risk: Fragile Systems

The biggest danger isn’t AI going rogue. It’s AI breaking systems that were never designed for it.

Consider:

• Hiring systems flooded with AI-generated applications

• Financial systems overwhelmed by synthetic identities

• Customer service channels flooded by bots

• Email and authentication systems losing meaning

These systems worked because they assumed human effort as a constraint. AI removes that constraint.

“All systems that depend on effort—writing emails, filling forms—are going to fail,” Jakulin says.

This isn’t a future risk. It’s already happening.

Accountability Is the Missing Layer

If there’s one idea Jakulin returns to repeatedly, it’s accountability.

Not abstract ethics. Not vague transparency.
Concrete, enforceable accountability.

“We don’t hold people accountable enough,” he says. “So they don’t ensure their tools are accountable either.”

His model is closer to aviation than software:

• black boxes

• incident reporting

• reproducibility

• traceability

In aviation, failures are expected—but deeply analyzed. In AI, failures are frequent—but often invisible.

He points to emerging efforts like OECD’s AI incident tracking as a start, but far from sufficient. What’s missing is infrastructure that makes accountability unavoidable.

Garbage In, Garbage Out—Still

While policymakers argue about model outputs, Jakulin focuses on something simpler: inputs.

Training data. Context data. Retrieval pipelines.

“If you want food safety, you inspect the ingredients—not the cooking.”

AI models trained on messy, uncurated, or unauthorized data will produce messy outputs. Systems using retrieval (RAG) without governance will amplify the problem.

Yet most regulation ignores this layer.

This is where his company, Data.Flowers, positions itself: not as another AI app, but as a “reality layer” for data—structured, governed, auditable.

The Oligopoly Illusion

There’s a widespread belief that AI will concentrate power in a handful of companies.

Jakulin disagrees.

“AI is a decentralizing technology.”

Why?

• Models are getting cheaper

• Open-source capabilities are catching up quickly

• Individuals can run powerful systems locally

The real bottleneck isn’t compute or models. It’s infrastructure.

“I can build an AI app in five minutes,” he says. “But I don’t have the rails it can run on.”

Those rails—data governance, provenance, permissions—don’t exist yet at scale.

And whoever builds them will shape the next decade.

Ethics Is Not the First Line of Defense

Jakulin is blunt about the current AI ethics landscape.

Too much focus on ideology.
Too little focus on mechanism design.

“Ethics is the instrument of last resort.”

Instead, systems should be designed so that:

• bad actions don’t pay off

• good behavior is incentivized

• failures are contained

This is game theory, not philosophy.

It’s also how most robust systems—from markets to infrastructure—actually work.

What Success Looks Like

So what does “good” AI governance look like in 10 years?

Not perfect models.
Not zero risk.

Something else entirely:

• Systems that fail gracefully

• Actions that are traceable and reversible

• Institutions resilient to automation

• A society that becomes more intelligent, not less

“The goal,” Jakulin says, “is human flourishing.”

That includes something unexpected: faster human development. He predicts AI could compress education timelines, enabling people to reach full capability years earlier.

The Quiet Shift: From Text to Vision

One of the most overlooked transformations, he says, is how we communicate.

We’re still interacting with AI mostly through text—like it’s 1960.

But that’s about to change.

“Our visual cortex is our mind’s GPU.”

AI-generated diagrams, maps, and infographics could dramatically increase how quickly humans understand complex ideas.

Less text. More structure.
Less explanation. More insight.

The Bottom Line

The dominant narrative around AI governance is about control:

• controlling models

• controlling outputs

• controlling access

Jakulin’s argument flips that:

You can’t control AI.
You can only control the systems around it.

And right now, those systems are brittle.

The next phase of AI won’t be defined by better models.
It will be defined by better infrastructure.

Or, as he puts it:

“We don’t need to stop the fire. We need to build a world that doesn’t burn down when it starts.”

This story was generated automatically from a podcast with Alec Crawford from Artificial Intelligence Risk, Inc. (click through to view):

Now AI reignited the discussion about the future of open source, we have the opportunity to rethink the foundations. Do we repeat the old mistakes? Or do we learn from the failures of the past?

Open Source might have broken the dominions of operating systems in the 90s, but it just gave us new dominions on the Internet. There used to be the Open Web, now we’re stuck with apps and Big Tech. Giving power away just creates opportunities for others to seize it where they can.

I had bought into the vision of an open source movement. I was an eager young programmer, contributing to free projects as much as I could. Billions of people use my work, some recognize it. Many also automatically assume I got paid for it. But I got nothing - I just gave it away, for the benefit of humanity. When I complained about this idea to the creator of GPL and Gnu, Richard Stallman, he just quoted Kant [*]. I shrugged and went back to my job, doing what I was told to do.

Consider this aerial photo from the island of Krk in Croatia:

What happened here? It’s probably not what you think. Nobody planted the forest in the middle of the desert.

It was all forest initially. It was the open part of the forest that turned into a desert. The closed forest has kept its trees, even if the desert is now pushing into it.

The open land was free for everyone. So shepherds brought their sheep, goats, and cattle. They cut down the trees. The animals overgrazed, thornbush replaced the grass. Shepherds torched the thornbush, water and wind eroded away the topsoil, rocks reduced the amount of grass. More overgrazing. More burning. More erosion. You see the result.

South of there, you can see how it ended: the farmers were trying to close the little that’s left by building walls around the pockets of remaining soil. But it was too late. It’s all abandoned now, and it will take hundreds, maybe thousands of years for the nature to regenerate itself.

We’re in another downward spiral right now, brought along by artificial intelligence.

AI is mostly made from data. Data is what is concrete and factual about the world. Data is also all the words, ideas, sounds and pictures humanity has put on the Internet. This data contains our civilization’s knowledge, wisdom, learning. This data has been analyzed at massive scale, and summarized in foundation models, like the large language models, video and image models. These models have never existed before, and provide a much better index to humanity than that what we had before.

When you approach an AI with your question or your task, the AI first primes the context by retrieving relevant data. Then it uses the models to extrapolate from the data towards an answer you will get. Sometimes multiple such AI requests are layered inside every request you put in.

The sources of the data used for the answers to your questions is often hidden. In the past we’d have to go to the libraries or to the experts, who’d point us in a good direction, and we’d retrieve books and articles. Then we’d examine these materials, and evaluate sources and ideas, and assemble a report. We’d then share this report with others, helping them build on our work, or stand on our shoulders as Isaac Newton would have put it.

This has been replaced by AI. The publishers, libraries and experts performed quality control over the information. Our research would check and verify information, and our authorship would give us both credit but also accountability. The reports would close the loop of sharing our learning with others, contributing to the body of knowledge.

AI is incredibly useful and powerful: we’re no doomers. AI has been trained on and is primed with data that is the cumulative creative and scientific output of the human civilization. Most of the intelligence is civilizational, not artificial. What’s artificial is the indexing, search, evaluation, and synthesis. By offering AI’s output without disclosing the data roots is like feeding us with dishes while refusing to tell us about the ingredients.

The absence of roots exposes us to massive risks - while also disconnecting the data ecosystem from resources. The impact of AI on the open data ecosystem can prove to be catastrophic. Wikipedia experienced such an onslaught of bots:

Websites that aggregated programmers’ insights have been completely replaced by AI systems:

Websites that curated and organized health information are experiencing similar reductions:

My dear friends, we’re about to find ourselves in the middle of a desert. The idea of open access and open data has made AI possible, the forest has been cut, and the soil is rapidly getting washed out. We need to do something. It’s not an AI question, it’s a data question. It’s a data sustainability question.

It’s almost 20 years since when I stepped out of the status quo. I was teaching students at Columbia University in New York City how to mine data, how to extract data. It was clear to me that this wasn’t sustainable. I left because I felt an obligation to protect and grow data.

It’s not too late to move AI from being an extractive industry to being an industry that will lead to a 1000-year period of flourishing of civilization, engaging in compounding growth of civilizational and planetary intelligence.

Wishful Kantian thinking needs to stop. We need to take ownership of the future. The good guys need to step up, and defend. Being weak is evil. We need to be meek: respectful of the values, but strong towards those who violate them. I’ll end with Casey Muratori’s words:

What happens when you decide that you’re going to give something away for free so that other people can build things on it, they do build things on it. They take all the power and control for themselves and even though your contribution by percentage may be significantly higher than theirs, like building an operating system versus building Amazon Web Services, at least on the software side it’s just it ends up with a bunch of people sitting in a room like we are now talking about how you start to take back ownership of your computers.

Well, all of the ownership would have already been ours if the programmers had asserted that power and control for themselves! But instead we didn’t.

Casey Muratori, 12:00 in “Why Hasn’t Open Source Won?”, Apr 25, 2024

Are you asking what to do? Engage. Spread the word. Point to those individuals and projects who need support. State what you can offer yourself.

This article will discuss how content platforms leak data without monetizing it, which gets monetized by others. Data siphons are how data gets lost. In the end, you'll better understand how the power and revenue have shifted away from content and what we can do about it. 

"Data is the new oil" has long been a rallying cry of entrepreneurs and consultants. In the beginning, data flows were virtually free. Data sources were tapped, information was extracted, leveraged for various purposes, or sold to hedge funds. Data tappers grew rich, even if few of them owned data directly. Many data owners were all too eager to outsource their data, so data access was cheap.

Many have been giving away their data rights for a pittance. The success of data tapping might have awakened some of them. Most, however, are still unaware of how and what data continues to be siphoned away. As a data science researcher, I'd like the data to serve those who create content, not against them. Internet depends on quality content, so the future of the internet depends on how it will treat the creators and curators of content and data.

Let's define it: a data siphon is where data extractors tap into someone else's data. Here’s a sketch:

The goal for a wise data operator is to control the siphons. The trouble is that very few content platforms prevent siphons, and even fewer support siphon control. Increasingly, the owner of the data stores all of it inside a 3rd party SaaS platform (content management system, social network) and trusts it not to abuse the data but has no direct way to protect the data. The owner that depends on platforms can lose access to his data. 3rd party analytics systems discard data as a standard practice: I learned this the hard way doing deep data analysis for a popular internet service that relied on Google Analytics.

In the coming years, you'll be hearing more about data siphons and secure data platforms. As time goes on, we need to shift to a world where it's theoretically much harder to tap someone else's data, and we don't just rely on trusting our vendors. In the meantime, however, let's explore the kinds of valuable data content generates. Here's a categorization with more information:

• Audience Data: Knowing your audience and contacting them is best understood of all forms of data. Social networks promised you to grow your audience by giving the social network platform control of it. This pyramid scheme did reward a few early adopters. Still, as with multi-level marketing schemes, the late-comers ended up donating their network to the platform. The platform and those tapping it can extract the audience and use this for other purposes, including launching competing content. This data is typically non-problematic: media of the past knew who their subscribers are, and newsstands knew who bought what.

• Engagement Data: Knowing what part of your content is getting read, copied, shared, engaged with is incredibly valuable to developing better content. A mere fraction of the available data is shared with the creators. This data is non-personal and analyzed in aggregate form, therefore rather non-problematic. The cookies used for this purpose serve merely to avoid double-counting a single user through multiple page loads. However, it had been costly in the pre-interactive era (Nielsen is a multi-billion business). Now, it's virtually free if your platform shares this with you, of course.

• Interest Data: This is where the privacy issue starts appearing: deeper analytics can help establish who is interested in what. Interest data is sensitive for two reasons: few people are trusting enough to let someone else monitor what they're reading - and only massive data operations can do anything useful with such data. Some publishers experimented with automatic recommendation systems, but there have been relatively few success stories that didn't leverage massive data. However, in practice, interest data facilitates profiling: knowing what a user is interested in allows advertisers (including other publishers) to target them anywhere else.

• Content is data: Copyright seeks to protect investment into content, but rules and legal practices date back to armed disputes in the era of scribes. The most important use case, however, is re-curation. Re-curators are non-creators who can develop their own audience, create their own monetization, shape their own editorial voice, and even build their own interest data simply by sharing links to free content. Search engines and social networks are the ultimate re-curators who have begun to even re-curate the re-curators through algorithmic feeds. 

  There have been some legislative initiatives to prevent re-curation by banning deep links. Of course, by now, an AI can devour an article and regurgitate a new deep-faked piece of content containing the same information without breaking any (current) laws. Even a decade ago, content farms were siphoning off original reporting and thinking, but by now, the situation has deteriorated dramatically. The lobbying efforts are typically disguised as attempts to improve access to knowledge. They do this by empowering platforms and re-curators at the expense of people and organizations that create and curate this knowledge. In the end, there's less knowledge for everyone, and the quality is worse, but the platforms get entrenched and all-powerful.

What should we do? There are some simple lessons for a content creators/curator:

• Scrutinize the platform's business model and terms of service

• Seek to own your platform and your data

• Be aware of siphons: plug them or meter them

• Advocate a broad rethinking of how society should protect investment in content

You might be wondering how simple it is to plug the siphons. Regardless of how secure internal IT might be, most tapping happens through abundant siphons on the customer side. Those are incredibly hard to control: the internet's general architecture has not been designed to prevent siphons, and taps can be set up in numerous ways: 

• Cookies can be dropped by 3rd party advertising. Browser companies have almost entirely plugged this siphon. On the other hand, this and consumer privacy regulations have mainly served to entrench the data tapping oligopoly held by the most powerful internet companies over smaller advertising networks.

• JavaScript code, typically web analytics plus SSO login, payments, and social sharing, can engage in almost arbitrary telemetry on the web pages loading them. JavaScript is a widely unappreciated source of siphons, even if telemetry can be detected in theory.

• The web serving platform has undetectable access to almost everything, and an audit is typically impossible.

• The internet browser can see everything that the customer is viewing and transmit this information virtually undetectably.

• The operating system, privileged applications, or security holes of the user device can monitor anything on the device virtually undetectably.

• Network operators, router manufacturers, and virtual private network software typically cannot see the content of communications but can see what websites and apps users are accessing.

• The hardware manufacturer can intentionally or unintentionally facilitate data taps for sophisticated operators.

To summarize, creators and users are forced to trust numerous vendors, and the regulators are trailing general data tapping practices by approximately two decades. It's an incendiary setup, and the best way to tackle it is at the systemic level through the digital equivalent of fire code regulations, which I've written about in a contribution for a European Parliament group. 

In the future, data will become as expensive as oil, and digital power will be rebalanced. But the first step to take would be to limit the use of dark data - procured through opaque means and unlicensed siphons. For example, I might now have access to what are my Google or Facebook profiles. I still don't know precisely how they established those profiles and exactly what data they're keeping (or can generate) on me. 

Generation of personal data is an important consideration, as the tech giants have been shifting from keeping data on individuals to instead keeping data on "cohorts." Cohorts are no assurance of privacy, as cohort data can postulate personal characteristics with high accuracy.

In the following installments of this series, I will cover at least two more weaknesses of publishers' currently available revenue models, proceeding to upcoming changes and opportunities. Subscribe to be notified of posts as they get published! And join the conversation by posting a comment.

Leave a comment

Subscribe now

Thanks to Ash Kalb for comments that improved this article!

• Reporters -11%

• Editors -7%

• Photographers -4% 

• Writers & Authors  -2%

At the same time, media jobs in marketing are growing:

• Public relations +7%

• Advertising managers +6%

Media jobs are moving away from content creation towards content monetization. As a result of this shift, we're receiving improved advocacy and deteriorating reporting. We're being persuaded instead of being informed.

What caused this shift? I'd argue it has been the demonetization of content that resulted in content's loss of power, leading to a reduction in quality, and finally, influence. This series of articles will explore the causes of content demonetization and propose solutions.

Demonetization means that the creator has lost control over monetization. A demonetized digital media property does not generate revenue directly on their website but instead somewhere else. Let's review the fundamental four sources of digital publisher revenue and how the publishers get the short end of the stick compared to their partners:

• Advertising: 

  Every engagement with an ad redirects the reader away from the publisher's property. The publisher's visibility into subsequent actions of the reader is lost. Historically, publishers sold their ad space directly and took some responsibility for the ads. Now, ad exchanges sell ad spaces to the highest bidder automatically, and the publishers have to trust reporting. The market leader in 2001, Double Click, had a market cap of $1.7 billion, and it's now a part of Google -- a company with a $1.6 trillion market capitalization.

• Affiliate marketing

  Every engagement with the affiliate link redirects the reader away from the publisher's property. The publisher has no visibility into the readers' actions on the merchant's property, trusts the merchant's self-reported revenue, and receives funds indirectly from affiliate network operators. The leading generator of affiliate marketing revenue, Amazon, has seen its market capitalization grow from $2.2 billion in 2001 to $1.6 trillion in 2021.

• Subscriptions

  The publishers often sell subscriptions in a separate marketplace, such as the market-leading Apple App Store, which generated $64 billion in 2020 revenues, of which it keeps 30%. The Apple Store did not exist in 2001, but now it's approaching 20% of Apple's revenue. Apple's market cap is $2.1 trillion.

• Licensing

  The publishers may license their content to aggregators and social networks. The cost of licensing is minuscule compared to the social network revenue, given that 52% of adults rely on social networks as a news source. For example, in 2019, Facebook has been willing to pay as much as $3 million a year to license headlines and previews of articles from news outlets. Facebook's market cap in 2021 is $0.93 trillion.

The strategic implication for publishers is to regain control of their revenue and ensure that the value creation happens on-site rather than off-site. There is a new wave of technology companies emerging that seek to empower the content creators:

• Advertising: Washington Post's https://www.zeustechnology.com is empowering publishers to sell their ad space directly.

• Affiliate marketing: My own company, 1o.io, empowers publishers to host the merchant's shops on-site and allows readers to complete purchases without leaving the content. Publishers audit every individual transaction.

• Subscriptions: Substack.com lets creators connect with their audience directly and monetize by the subscription model without a go-between.

• Licensing: There is a fundamental gap in how copyright holders can enforce their license deals on behalf of copyright holders. Watch this space.

There is a significant upside: the computer game industry has not given up as much control over its revenue, and was worth $9.4 billion in 2001 and is now worth $175.8 billion.

In the following installments of this series, I will cover other weaknesses of publishers' currently available revenue models, proceeding to upcoming changes and opportunities. Subscribe to be notified of posts as they get published!

Thanks to Ash Kalb, Kristina Libby, Viren Oswall, and Allan Hoving for helpful comments and feedback!

Sign up now so you don’t miss the first issue.

Subscribe now

In the meantime, tell your friends!