1. Public Does Not Mean Free for Unlimited Extraction
We often make a simple assumption: if something is public, everyone can use it. But public access has always existed within a social context. A public library allows people to read its books. A public park allows people to use the space. Neither example automatically answers every question about large-scale commercial extraction.
You cannot arrive at a public park with mining equipment, extract valuable resources for private profit, and defend yourself simply by saying, “But the park was public.” The distinction is between public access and private extraction of economic value.
The open web raises a similar question. People publish material publicly for many different reasons—education, research, journalism, discussion, entertainment, personal expression or business. Public availability does not necessarily mean that every creator explicitly agreed in advance to every future commercial use of that material.
This is not, by itself, a claim that AI training is legally unlawful. It is a question about the economic and social assumptions surrounding public knowledge.
2. The Transformation Argument
AI companies can make an important argument: training is transformative. And it is. A model does not simply copy the Internet and return it unchanged. It processes enormous quantities of information, identifies patterns, adjusts its parameters and develops capabilities that did not previously exist in that form.
But there is a distinction we should not lose:
The process is transformative. The input is not.
Transformation describes what the AI system does with the material. It does not explain where the material came from. A refinery transforms crude oil. A factory transforms raw materials. Neither transformation means that the underlying resources were created by the refinery or factory.
Likewise, AI transforms information. The interesting question therefore moves beyond whether the process is transformative: What social obligation arises when a transformative commercial system is built upon an enormous intellectual inheritance accumulated by humanity?
3. The Gatekeeper Problem
A gatekeeper begins by providing access. It builds tools, improves navigation and makes a resource easier to use. Eventually, however, the gatekeeper may discover that the resource flowing through the gate has enormous economic value. The gatekeeper can then become more than a facilitator. It can become the dominant private beneficiary of the resource.
This is the gatekeeper problem. The question is not whether AI companies are literally thieves. That would reduce a complicated economic issue to an accusation. The better question is: Can the gatekeeper of humanity’s intellectual commons become its largest private extractor of economic value?
4. The Rented Farmland Test
Imagine renting farmland at an agreed annual rate. The landowner provides the land. The farmer provides the seeds, labour, experimentation, decisions and risk. After a year, the farmer produces a bumper crop.
The landowner cannot reasonably say, “The land belongs to me, therefore the crop belongs to me.” The payment was for access to the land. It did not automatically transfer ownership of everything produced through that access.
The AI analogy is important. We are not simply renting intelligence. We are using infrastructure and services. Computing capacity, servers, software, interfaces, models, engineering and capital make the work possible. But payment for that infrastructure does not automatically mean that everything a user creates through it belongs to the infrastructure provider.
The principle is broader: Paying for infrastructure does not automatically transfer ownership of the intellectual output created through that infrastructure.
5. The Navier–Stokes Test
A controversy unfolding on September 8, 2026 provides a striking real-world test of the boundary between AI infrastructure and intellectual contribution. Mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge published AI-assisted, Lean-verified results on fluid equations, while Buckmaster also described a dispute with OpenAI over a separate internal result concerning Navier–Stokes.
According to Buckmaster’s account, after information about their work reached OpenAI, OpenAI researcher Sébastien Bubeck told him that an internal model had produced a roughly 100-page proof related to Navier–Stokes. Buckmaster says he asked whether the model had been trained on, or had access to, their Codex sessions containing drafts from the project. He says he was told the model did not look up user data, but that his follow-up question about training was not answered. These are Buckmaster’s allegations, not established facts. Bubeck has publicly described the allegations as “false and inflammatory,” and the reported internal proof has not been publicly verified. Today’s Navier–Stokes controversy coverage
That distinction is important. This controversy is not proof that OpenAI used private research sessions for training, nor is it proof that OpenAI has solved the Millennium Prize problem. But it raises an extraordinary question: imagine a mathematician paying for AI infrastructure while using it to attack one of the hardest problems in mathematics. The researcher supplies the intellectual direction, experiments, failed attempts, discoveries and research risk. The AI infrastructure accelerates the work. What happens when the provider’s own researchers subsequently produce a potentially valuable result in the same research territory?
The question is not simply who owns the servers. It is: Where does providing the infrastructure end and appropriating the intellectual opportunity begin? That is the Rented Farmland Test in a real-world setting.
The farmer rents the land. The farmer grows the crop. The AI researcher rents the infrastructure. Who owns the intellectual harvest?
The more powerful the infrastructure becomes, the more important it is to establish clear boundaries around the intellectual value created through it.
6. The Marketplace Test
Now imagine a marketplace. You pay to participate. You discover an opportunity, develop a product, take the risk and build demand. The marketplace sees what happens because it operates the platform. Then imagine that the marketplace uses its position to launch a competing product based on what it has learned from successful sellers.
The marketplace may own the shop. But owning the shop is not the same as creating the merchant’s product. This is why concerns surrounding marketplace platforms and private-label competition provide a useful analogy. The point is not to accuse any particular company of wrongdoing in every case.
The point is to expose a recurring economic problem: Infrastructure ownership can become confused with ownership of the value created through that infrastructure.
AI could make this problem vastly more consequential.
7. The Jackpot Test
Consider another example. Suppose 100,000 people buy $10 coupons, with the stated possibility that one coupon will become a $1 million jackpot. One person gets the winning coupon. The seller then says, “You can have your $10 back. Actually, I’ll give you $20. But the million-dollar upside belongs to me.”
Assuming the original bargain did not reserve that upside for the seller, something about the arrangement immediately feels wrong. The payment purchased a chance. It did not necessarily purchase the seller’s right to confiscate the winner’s extraordinary future upside.
The analogy is not a legal argument about lottery contracts. It illustrates a broader economic intuition: Payment for access does not automatically mean surrender of every extraordinary outcome produced through that access.
8. The Three Contributions
There are at least three different contributors to the AI economy. The first is the creator: people who produce books, research, software, discoveries, experiments, photographs, explanations, ideas and other intellectual work.
The second is the AI company: the people and institutions providing capital, engineering, computing infrastructure, research, data processing and enormous organizational effort.
The third contribution is much harder to put on an invoice: humanity’s accumulated intellectual inheritance. Newton did not create mathematics from nothing. Einstein did not create physics from nothing. Modern engineers, scientists and researchers inherit knowledge from countless generations.
Every generation stands on previous generations. AI is doing something historically unusual with that inheritance: industrializing the ability to learn from it.
9. The Strange Economic Outcome
Now consider where this could lead. Humanity collectively accumulates knowledge. AI companies transform increasingly large amounts of that knowledge into increasingly powerful productive systems. Those systems potentially reduce the need for human labour. Economic value increasingly flows toward AI, capital and whoever owns the productive infrastructure.
And then humanity is told: “Perhaps you will need Universal Basic Income because AI has made your labour less valuable.”
There is something deeply strange about that outcome. Humanity created the intellectual inheritance, the science that made computers possible, the institutions, infrastructure and networks that made the digital economy possible, and generations of labour and creativity. After all of that, humanity could potentially end up asking the owners of AI for an allowance.
10. From Ownership to Begging
There is a profound difference between income and ownership. A person receiving income is a recipient. A person owning an asset is a stakeholder.
This distinction becomes increasingly important if AI eventually performs a large share of economically valuable work. If labour becomes less necessary, labour’s bargaining power may decline. If wages become less important, ownership of productive capital becomes more important.
So the fundamental question should not simply be, “How much income will people need after AI takes their jobs?” It should be: “Who owns the productive economy after AI transforms it?”
Why should the people who lose bargaining power through automation also lose their stake in the productive systems that replace their labour?
11. The Knowledge Dividend
Suppose commercial AI companies generate extraordinary profits partly because they can transform humanity’s accumulated intellectual inheritance into extraordinarily productive systems.
One possible response is to establish a Knowledge Dividend.
Under this proposal, 50% of AI-company profits would return to society.
The 50% is a policy proposal, not existing law. It is also not intended to place an individual price on every fact, webpage or piece of knowledge. The idea is different. If an industry can legitimately transform an enormous common intellectual inheritance into extraordinary private wealth, then extraordinary commercial success should carry a corresponding obligation of reciprocity.
Prosper from humanity’s inheritance. Return a substantial share of that prosperity to humanity.
12. Why 50%?
Why 50%? Because the argument is not that AI companies contribute nothing. They contribute enormous amounts: capital, engineering, research, computing infrastructure, risk and organizational capability. But neither did the companies create the entire intellectual foundation from which they are building.
The 50% figure therefore represents a deliberately substantial sharing rule. It says that extraordinary AI profits should not be treated as if they emerged solely from corporate capital and engineering. A significant part of the value comes from standing on an enormous accumulated human inheritance.
The exact percentage would ultimately be a matter for economic and political debate. The principle is more important than the number: Extraordinary private extraction from a common inheritance should produce extraordinary public return.
13. From Dividend to Universal Basic Equity
But distributing money is only the beginning. The deeper idea is Universal Basic Equity.
Universal Basic Income says, “Here is money so you can survive.” Universal Basic Equity says: “You have a continuing economic stake in the productive economy.”
That is a radically different relationship. The Knowledge Dividend could become the funding mechanism for a permanent pool of productive assets held broadly on behalf of society.
Instead of simply receiving and spending a monthly payment, people could accumulate an enduring economic stake. The assets could generate returns, compound over time and provide benefits to future generations.
The objective would be to create ownership, not merely assistance.
14. UBI Versus UBE
The distinction can be stated simply. UBI: AI may make your labour unnecessary, so here is money to compensate you. UBE: AI transformed the economy, and you retain a stake in the economy it created.
One treats people primarily as recipients. The other treats them as owners.
That difference becomes enormous if productive capital becomes more powerful than human labour. A society of workers can bargain through work. A society in which machines perform much of the work may have to bargain through ownership.
15. The Dystopian Divergence
This is where the AI future can split.
The inclusive trajectory: Human knowledge → AI transformation → massive productivity → Knowledge Dividend → Universal Basic Equity → broad ownership.
The dystopian trajectory: Human knowledge → AI transformation → massive productivity → concentrated ownership → reduced labour bargaining power → extreme wealth concentration → permanent elite power.
The second possibility is more serious than ordinary inequality. Historically, economic power could change hands. New industries created new fortunes. Workers could become entrepreneurs. New businesses could challenge established companies. Political movements could challenge entrenched economic interests.
But what happens if AI makes productive capital extraordinarily powerful while making human labour comparatively unnecessary? The traditional bargaining mechanism weakens. If ownership is already concentrated, the owners of productive AI systems could capture an increasing share of the economic surplus.
And then AI could produce something more dangerous than inequality: A divergence in which economic elites acquire the structural ability to remain elites indefinitely.
16. AI Could Democratize Abundance—or Concentrate It
AI becoming powerful is not necessarily the problem. A world in which machines can produce extraordinary abundance could be one of the greatest achievements in human history.
The question is: Who owns the abundance?
If AI produces ten times more economic value, humanity could potentially become enormously more prosperous. But ten times more productivity does not automatically mean ten times more prosperity for everyone. If ownership remains concentrated, it could instead mean vastly greater leverage for the people who own the productive systems.
This is the difference between technological abundance and shared abundance.
17. The Permanent Elite Problem
The deepest danger may therefore not be that a few people become extremely wealthy. It is that the economic system begins losing its ability to redistribute ownership itself.
If human labour is no longer essential, wages cannot remain the primary mechanism through which ordinary people participate in economic growth. If capital becomes increasingly productive, capital ownership becomes increasingly important.
And if that ownership is concentrated before AI reaches its full potential, the resulting wealth can reinforce the very concentration that created it. Money can buy more capital. Capital can buy more AI. AI can produce more wealth. Wealth can purchase more influence. Influence can protect the ownership structure. This is the broader concern explored in the permanent-elite problem.
That creates a potential feedback loop: Capital → AI → productivity → wealth → ownership → influence → more capital.
If that loop becomes sufficiently strong, the concern is no longer simply rich versus poor. It becomes: Can economic power still change hands?
18. The Social Contract
This suggests a new bargain between AI and society.
AI companies should have freedom to innovate, access knowledge subject to law, commercialize their technologies and earn substantial private returns. Society, in return, should receive transparency, protection for creators, preservation of the knowledge commons, a Knowledge Dividend and broad participation in the wealth created by AI.
The objective is not to punish AI companies. It is not to prevent technological progress. It is not even to prevent extraordinary private fortunes.
The objective is to prevent technological progress from becoming a mechanism through which humanity’s accumulated inheritance is converted into permanently concentrated ownership.
19. The Question We Should Ask Now
We are still early enough to make these choices. Once enormous fortunes, infrastructure and political power have accumulated around AI, changing the economic architecture will become much harder.
That is why the question should not wait until millions of people have already lost bargaining power.
We should ask now: How do we make sure people own part of the economy AI is creating before ownership becomes extremely difficult to redistribute?
This is also why transparency matters. If society is going to debate the economic return from humanity’s knowledge, we need to understand what is being used, how it is acquired and under what rights.
A useful provenance framework would be: Source → Work → Rights status → License → Acquisition route → Date collected → Training use → Opt-out status.
20. The Real Choice
There are potentially two very different AI futures. In one, AI becomes a great equalizer. Machines create extraordinary abundance, while society broadly owns a meaningful portion of the productive systems generating that abundance.
In the other, AI becomes the great divergence. Machines create extraordinary abundance, but ownership becomes so concentrated that economic and political power increasingly belongs to a small permanent elite.
The technology could be largely the same. The difference is ownership.
That is why the AI debate cannot remain only about intelligence, jobs and safety. It must also become a debate about capital, property, bargaining power and who owns the future.
21. Don’t Turn Humanity Into a Beggar
Humanity has spent thousands of years building the intellectual foundation of civilization. AI may finally give us machines capable of exploiting that foundation at industrial scale.
We should not be afraid of that possibility. We should be amazed by it. But we should also ask whether the economic architecture surrounding it is fair.
A future in which humanity supplies the accumulated intellectual inheritance, AI industrializes it, a small ownership class captures most of the resulting wealth, and everyone else receives basic income to survive would be a remarkable historical inversion.
The people who built the intellectual house would become tenants in it.
That is not the future we should automatically accept.
The alternative is not hostility toward AI. It is reciprocity.
Let AI learn. Let AI transform. Let AI innovate. Let AI prosper. But if extraordinary AI prosperity is built partly upon humanity’s extraordinary accumulated inheritance, let humanity prosper with it.
Not merely through welfare. Not merely through Universal Basic Income. Through ownership.
Universal Basic Equity: not compensation for becoming unnecessary, but ownership of the future.
And perhaps that is the social contract we should establish before the machines become powerful enough to establish one for us.