If AI companies can commercially learn from humanity’s knowledge, should humanity receive a return?
1. The Question We Are Asking
The debate over AI and intellectual property is often reduced to one question: Is AI training fair use? That matters, but it may be the wrong place to begin. Before asking whether the transformation is lawful, we should ask what was acquired, under what rights, and who ultimately benefits from turning humanity’s accumulated knowledge into commercial intelligence.
2. The Word Test
Consider something ordinary. I pay Microsoft every month for a Microsoft 365 subscription and use Word to write a book. Microsoft owns Word; I merely rent access to the software. Yet Microsoft does not become the owner of my book. The tool belongs to Microsoft; the intellectual work I create with it belongs to me. A calculator provides an even simpler example: it performs my calculation, but it doesn’t acquire ownership of my reasoning.
Generative AI is obviously different because it can participate much more deeply in the creative and intellectual process. It can analyze, explain, reorganize, suggest and generate. But that makes it more important—not less important—to distinguish ownership of what the user creates from the rights granted to the AI provider to process what the user submits.
If I pay for an AI service and deliberately opt out of model improvement, my reasonable understanding is that I am paying to use the intelligence, not donating my private research to improve the intelligence sold to somebody else. That is a very different issue from what happens to material gathered from the public Internet.
3. Public Is Not the Same as Licensed
One of the most convenient shortcuts in the AI debate is: “It was publicly available, therefore AI companies could use it.” Public accessibility tells us that information could be accessed. It does not tell us that every possible commercial use was licensed.
A work available online might be public domain, CC0, CC BY, CC BY-NC, CC BY-ND, subject to a publisher’s license, covered by a software license, or fully copyrighted with no open license. The fact that all of these works can be found on the Internet does not make their legal status identical.
The ORCID example makes the distinction obvious. ORCID can make its public metadata available under CC0 while the actual papers represented in those records can carry completely different rights. The metadata being open does not turn every underlying paper into CC0. Likewise, the Internet being accessible does not turn the Internet into a universal AI-training license.
4. The Farmer Test
Consider an illiterate farmer who develops an extraordinary agricultural technique. He knows nothing about copyright, Creative Commons, licensing or AI. He records videos demonstrating his method and puts them online because he wants people to learn from it. An AI company subsequently collects those videos, processes them and incorporates information derived from them into a commercial system that millions of paying customers use.
Where did the farmer agree to license his work to the AI company for commercial training? The answer cannot simply be “He put it on the Internet.” He made his work visible; that does not necessarily mean he consented to every future commercial exploitation of it.
That does not prove that the AI company’s use is unlawful. Fair use exists precisely because copyright does not prohibit every unlicensed use. But it demonstrates why public availability cannot be the entire legal analysis.
5. The Missing Step
This is where the word “transformative” deserves closer examination.
1. Someone creates the work.
2. The AI company acquires and processes it.
3. The material enters training.
4. The company develops a commercial intelligence.
5. Customers pay to use that intelligence.
When someone says “Training is transformative,” they are primarily describing Step 3. But the controversy begins earlier. How was the material acquired? What was its copyright status? What license applied? Was it public domain? Was it CC BY? Was it CC BY-NC? Was it proprietary? Was there an opt-out? Was a licensing market available? And then comes another question: What economic effect does the resulting AI system have on the original creator?
Transformation may be an important part of a fair-use analysis. It cannot logically answer every question about acquisition, rights and economic consequences.
6. Transformation Does Not Mean Creation
AI may transform the material. But the AI company did not create the underlying civilization from which the training material emerged. It did not invent mathematics. It did not conduct every scientific experiment. It did not write every textbook. It did not develop every farming technique. It did not create every useful piece of software.
It built an extraordinarily powerful machine for processing patterns contained in an enormous body of human-created information. That is an extraordinary engineering achievement. But building the machine is not the same as creating everything the machine learned from.
7. The Loney Example
S. L. Loney provides a useful illustration. I learned mathematics from Loney. Another generation learned from other mathematicians. Today’s students can learn the same mathematics from teachers, books, computers or AI.
That chain of learning is not theft. Knowledge advances because people are allowed to learn from what came before them.
The interesting change is economic. The AI is no longer merely another student. It can become a commercial intermediary between humanity’s accumulated knowledge and millions of new students. A student pays the AI company to obtain explanations, solve problems and learn concepts whose intellectual ancestry may stretch back centuries.
Knowledge itself must remain capable of becoming a commons. The new question is whether a company that industrializes access to that accumulated knowledge should capture essentially all of the economic value created by doing so.
8. Fair Use Is Not a Universal License
A Creative Commons license is permission from a rights holder under specified conditions. Fair use is a legal doctrine that can permit certain uses without permission. They are not interchangeable.
Commercial use can sometimes be fair. Nonprofit use can sometimes infringe. Fair use is evaluated using multiple factors.
Therefore: Transformative ≠ automatically fair.
Publicly accessible ≠ automatically licensed.
9. Who Owns the Knowledge?
Human knowledge is cumulative. A mathematical theorem does not become permanent private property simply because somebody discovered it. A scientific fact is knowledge. A farming technique is knowledge. A mathematical relationship is knowledge. A historical fact is knowledge.
Copyright generally protects expression rather than the underlying facts, ideas, methods and systems. That distinction is essential.
Human civilization progresses because people are allowed to learn from what came before them. I learned from Loney. Someone else learned from me. Another generation will learn from them. That is civilization.
10. The Economic Transformation
AI introduces something new. A book allows one person to learn. A teacher can teach hundreds. The Internet can reach millions. A sufficiently capable AI can potentially provide individualized intellectual assistance to hundreds of millions or billions of people simultaneously.
That is an enormous economic transformation. The company providing the AI can charge for access to this capability.
So the relevant question becomes: Who should benefit when machines industrialize humanity’s accumulated intellectual inheritance?
11. The Knowledge Dividend
Paying every contributor individually would be nearly impossible. Imagine calculating how much one textbook, one research paper, one Wikipedia editor, one programmer, one teacher or one farmer contributed to a model whose capabilities emerge from millions of interconnected sources.
A different solution is a Knowledge Dividend. Commercial AI companies would remain free to innovate and make substantial profits, but a significant portion of those profits would return to society through a dedicated tax or levy and be distributed directly to people.
50% of AI-company profits → direct human dividend
The 50% figure is a policy proposal, not an existing legal requirement. Its purpose is to make the principle clear: if commercial AI can build extraordinary value from humanity’s accumulated intellectual ecosystem, humanity should receive a meaningful economic return.
12. Why 50%?
Five percent can look like philanthropy. Ten percent can look like corporate charity. Twenty percent can look like a negotiated concession.
Fifty percent says that humanity is a stakeholder.
AI companies would still retain substantial resources for research, infrastructure, employees, investment, expansion, innovation and shareholder returns. But the assumption that all economic surplus should accrue to private owners would be rejected.
This becomes particularly important if AI dramatically increases productivity while reducing demand for human cognitive labor. Productivity can rise enormously while the distribution of income becomes increasingly unequal.
13. And What About AI Executives?
A second proposal is an AI-company executive compensation ceiling of $1 million per year.
This must mean total company-linked compensation—not merely salary. Otherwise an executive could receive a $1 million salary plus a $100 million bonus and $500 million stock grant and technically claim compliance with a salary cap.
The proposed ceiling would therefore cover salary + bonuses + options + equity grants + other company-linked remuneration.
This is not an argument that AI researchers, engineers, founders or investors should work for free. They deserve substantial rewards. The argument is that building the machine does not mean creating the entire intellectual foundation from which the machine derives its capabilities.
14. Show What Was Used
There is another essential requirement: training-data transparency.
If an AI company says, “Our use of copyrighted material is lawful,” then creators and courts should have some meaningful ability to determine whether their work was actually used and under what circumstances.
This does not necessarily require publishing billions of copyrighted files. But there should be meaningful provenance:
Source → Work → Rights status → License → Acquisition route → Date collected → Training use → Opt-out status
Without provenance, there is an enormous information imbalance. The creator asks whether their work was used, while the company may be in the only position to know the answer.
Transparency is therefore not anti-AI. It is accountability.
15. The Three Contributions
A sensible AI policy should recognize three different contributions.
The creator contributes the original work.
The AI company contributes technology, capital, engineering and infrastructure.
Humanity contributes the accumulated intellectual commons on which the entire system ultimately rests. This question of human contribution is explored further in The Evolution of Humans in the New Economy.
None of these contributions cancels the others.
An engineer can legitimately say: “I built the machine.”
A creator can legitimately say: “I created this work.”
Humanity can legitimately say: “The knowledge from which civilization’s progress emerged did not begin with either of you.”
16. The Social Contract
The bargain could therefore be explicit.
AI companies receive: freedom to innovate; predictable rules; access to knowledge subject to applicable law; the ability to commercialize transformative systems; substantial private returns.
Society receives: training-data transparency; protection for legitimate creator rights; continued access to the knowledge commons; a direct AI Knowledge Dividend; safeguards against extreme concentration of AI-generated wealth.
This is not a demand to stop AI from learning. It is a demand to decide what happens after AI learns.
17. The Real Question
The AI debate has become trapped between two extremes: “AI is stealing everything” and “Everything was publicly available, so AI can use everything.”
Neither is sufficiently precise. Some material is public domain. Some is openly licensed. Some is restricted. Some is commercially licensed. Some is fully copyrighted. Some uses may qualify as fair use. Some may not. Different jurisdictions may reach different conclusions.
That complexity is precisely why the solution should include provenance, transparency, legal accountability and a broader economic framework.
18. The AI Future Should Not Belong to a Few
Suppose courts ultimately decide that particular forms of AI training qualify as fair use. The debate still doesn’t end.
Because legal permission and economic fairness are different questions.
A use can be lawful while society decides that the resulting economic surplus should be shared more broadly. See also Are We Heading Toward a Selective Utopia for the Rich?
AI could become the greatest productivity engine in human history. It could also become the greatest wealth-concentration engine in human history.
Which one happens is not determined by the technology alone.
It depends on the rules we create around it.
19. A New Social Contract for AI
The principle is simple:
Let AI learn from humanity. Let AI transform knowledge. Let AI companies innovate and prosper. Protect legitimate intellectual property. Make training provenance transparent. And return a meaningful share of AI-generated prosperity to humanity.
A 50% AI profit dividend and a $1 million executive compensation ceiling are not the only possible mechanisms. They are deliberately strong proposals intended to force the underlying question into the open:
If humanity created the knowledge, should the economic prosperity created by machines learning from that knowledge belong exclusively to those who own the machines?
20. Let AI Prosper — With Humanity
The goal should not be to prevent AI from learning. It should be to prevent humanity from becoming merely the unpaid substrate of a privately concentrated intelligence economy.
I want the farmer to have the AI. I want the child without a university nearby to have the AI. I want the scientist to have the AI. I want the programmer to have the AI. I want the writer to have the AI. I want everyone to have access to the extraordinary intellectual capabilities this technology can provide.
But if AI becomes extraordinarily profitable because it can process and commercialize humanity’s accumulated knowledge at unprecedented scale, then the benefits should not flow only upward.
Humanity created the knowledge.
AI can multiply its usefulness.
The economic system should multiply its benefits too.
That is not anti-AI. It is a proposal for an AI social contract.
Let AI learn. Let AI innovate. Let AI prosper. But let humanity prosper with it.