The Next Dataset Is Us

Companies took the internet first. Now the next dataset is us.

You probably used one of these tools today. You asked it something that mattered. About your work, your money, your health, a person, a decision you are trying to make. It helped. That is where this begins. Not with fear. With one honest question: while it was helping you, what was it learning about you?

Start with how we got here. The pages. The posts. The code. The images. The public record of human work.

They trained on it. They sold products on top of it. They argued about it in court. They called it innovation, scale, personalization, safety, fair use, better products.

Some of that was true.

That is the uncomfortable part. These companies did build useful things. Search helped people. Social products connected people. AI tools help people write, code, learn, research, and move faster. I use them. I have worked in and around this world. I am not writing this from the outside.

But a useful product can still be built on a broken bargain.

And the bargain is broken.

The old bargain

The old technology bargain was simple:

Use the product. Give us the data. Trust us with the rest.

It looked free. It was not free.

People paid with attention. Then with their social graph. Then with location. Then with their photos, messages, searches, homes, routes, interests, friendships, preferences, and private patterns.

The companies got the durable value. The public got convenience.

Sometimes that trade created real good. I do not want to flatten the story. Technology has created access, opportunity, connection, and leverage for millions of people.

But the pattern is also clear. When the public did not understand the trade, the value still moved upward. When users were told they had control, that control often turned out to be weaker than they believed. The regulatory record is full of examples: Facebook's privacy settlement, Google Buzz, Safari tracking, Ring videos, sensitive location data brokers.

The point is not one company.

The point is the operating model.

Take the human signal first. Explain the bargain later. Make the product useful enough that people keep coming back. Then convert the accumulated human data into power the public cannot easily inspect, negotiate with, or walk away from.

AI makes the bargain deeper

AI is not just another version of this. It goes deeper.

The old internet was mostly static. Pages. Posts. Links. Images. Public records.

The new AI layer wants living data.

Your questions. Your documents. Your code. Your calendar. Your screen. Your meetings. Your voice. Your feedback. Your research prompts. Your drafts. Your private reasoning. Your team's unfinished thinking.

A deep research request is not just a query.

It is a map of what you care about. It reveals what you know, what you do not know, what decision you are about to make, what future you are moving toward.

At scale, that is not search.

That is a map of human intent.

And this is the part people are still underestimating: when you use these systems, you are not only receiving intelligence. You are also revealing intelligence.

The prompt is often the least valuable part.

The valuable part is the context around the prompt: why this matters, what tradeoff you are making, what your taste is, which answer you accept, which answer you reject, what kind of reasoning makes you move, what kind of mistake you catch, where your group disagrees, what your team is trying to become.

This is why the next dataset is not just the internet.

The next dataset is us.

Continuously. In parallel. Across millions of people.

The hypocrisy is now visible

Now the companies are fighting each other.

Publishers say AI companies used their work without permission or payment. Authors have sued over books. Reddit has sued over allegedly scraped community content. The New York Times has sued OpenAI and Microsoft. Courts are still working through what is fair use, what is sourcing, what is copyright, what is competition.

At the same time, AI companies are accusing other AI companies of extracting from them.

OpenAI reviewed allegations that DeepSeek used distillation from OpenAI models. Anthropic has accused Alibaba-linked operators of large-scale Claude distillation. Whether every allegation is true is not the point here. Some of these fights will be decided in courts. Some will be settled. Some may remain contested.

The important thing is the language.

Same act. Two vocabularies. The difference is who owns the intelligence.

That tells us something.

The industry understands that intelligence has value when it is their intelligence.

The public needs to understand the same thing about ours.

The real theft is leverage

I am using theft here in the moral sense.

The legal system will decide specific legal questions. Copyright. Contract. Fair use. Terms of service. Privacy law. Consumer protection. Those distinctions matter.

But moral theft is simpler:

Taking value from people before they have a meaningful choice, then using that value to reduce their bargaining power.

That is the part I care about.

The thing being taken is not only data. It is leverage.

If a system learns from your work, your judgment, your taste, your team, your mistakes, and your context, then the output is not neutral. It is partly made of you. If the company owns that intelligence and you do not, the company now has leverage over the future that your own work helped create.

Then it sells the product back to you.

Sometimes it sells it back as convenience. Sometimes as productivity. Sometimes as automation that replaces the very people whose work trained the system.

That is the broken bargain.

You create the raw intelligence. The company captures the compounding value. Then you rent access to a compressed version of what people like you already produced.

Group intelligence is a separate problem

Up to now, this has been about you. But you are never only you. You think inside teams, meetings, threads, families — and those rooms create something of their own.

Protecting one person's data is already hard.

Protecting what a group makes together is harder.

A team meeting is not just ten individuals speaking. It contains the pattern between them: disagreement, trust, taste, priorities, tradeoffs, history, tension, sequence, who notices what, who misses what, how decisions actually happen.

A company thread is not just messages. A research group is not just documents. A family chat is not just text. A founder's conversations with a team are not just prompts.

Groups create intelligence that no single person owns alone.

When AI captures that, summarizes it, trains from it, generalizes it, and turns it into product intelligence, we should treat it differently from the ordinary usage data a company collects about one person at a time. This is closer to a collective mind than to the data trail one person leaves behind — and that is why it needs legal attention, not just a settings page.

This is where the danger grows bigger than Facebook. Facebook mapped the links between us — who we know, what holds our attention, how we behave. AI can capture something deeper: how we think and decide together — our reasoning, our taste, our disagreements, the moves we are about to make. That is a far deeper map: not of individuals, but of how groups create value.

And that is the part no one person can sign away. If a company captures the intelligence of a team, a community, a profession, or a culture, one person's consent is not enough. The value never came from one individual. It came from what happens between people — and it should take all of them to let it go.

The illusion of control

Some of these companies will tell you that you are already protected.

They will point to a settings page. A toggle to turn off training. A button to delete your history. And they are right that these exist. Some of them genuinely help.

But I have worked on the inside of how these systems get built. So let me be honest about what most of these controls really are.

First, they come late. Privacy controls are almost never the first thing built. They arrive after the product is launched, popular, and already trained on years of human data. By the time you get the switch, the taking has mostly already happened. The control is a patch on a house that was already emptied.

Second, they are off by default. The burden is on you to discover the switch exists, understand what it does, and find where it is buried. Most people never will. That is not an accident. A default is a business decision.

Third, they punish you for using them. Turn the protections on and the product often gets worse. Delete your history in some assistants and they don't just forget. They lose their memory, their integrations, their usefulness, and quietly turn into a duller tool that can barely read a link you paste in. You are offered a trade: keep your data, or keep your product.

Fourth, even when you say no, your data can still be taken somewhere else. A screen. A log. A cache. A connected app. A "to improve our services" line you never read. There are many doors, and you were only ever shown one.

This is not control. It is the appearance of control.

A switch you are allowed to flip while the value keeps moving in the same direction. Real control is not a buried toggle you have to fight for. Real control is the default.

This is not anti-AI

It is too easy to make this anti-AI.

I am not anti-AI.

AI can create real value. It can help people think. It can reduce busywork. It can make expertise more accessible. It can help people with disabilities. It can help small teams move faster. It can help someone who has no network, no staff, no specialist, no time.

The question is not whether AI is useful.

The question is: who owns the intelligence it creates from humans?

Does the system make the human more powerful? Or does it make the human more extractable?

That is the line.

The standard should be different

We need a new standard for AI systems that learn from people.

Not buried settings. Not fake consent. Not "we may use your data to improve services" written in language no normal person can reason about. Not one toggle that decides the fate of years of your work.

The standard should be clear. This is the test I would ask of any AI product — tap each one and check it against the tools you already use.

Can I see what you learned from me? Can I take it with me? Can I delete it? Can I stop you from using it outside my benefit? Can my group decide what happens to group intelligence? Do I get the value back?

If the answer is no, you are not only the user.

You are the supply chain.

My line

I have worked in and around some of these companies.

I have benefited from this world. I have seen the good. I have also seen the direction.

A concrete version of that direction looks mundane from the inside. I have seen product reviews where a privacy or control question was translated into a product-quality question: if this is off, what memory disappears, which recommendation gets worse, which workflow breaks, which retention number drops? The data did not live in one box. It moved through event logs, caches, connected-app traces, feedback loops, and model-quality reviews. Nobody had to announce a bad intention. The default had gravity. The control had friction.

And I do not want to spend the next decade building another machine that takes more from humans, learns more from humans, sells more back to humans, and leaves humans with less power.

That is not the future I want.

The direction I care about now is the opposite:

Your intelligence, your control.

Systems that learn your work to serve you. Not systems that learn from you so they can serve everyone else while you lose leverage.

Systems that prepare the next move and wait for your approval. Not systems that quietly turn human thought into corporate power.

People should be more powerful because of AI. Not more exposed. Not more replaceable. Not more dependent on the same companies that captured their data in the first place.

The future should not be large corporations owning the compounded intelligence of humanity.

The future should be people and groups owning systems that make their own intelligence compound.

That is the line.

Your intelligence, your control.