If the public provides the raw material that powers artificial intelligence, the next step requires building a container to capture the value of that material. The previous argument established that machine learning models rely on human knowledge to function. The companies building the software constructed the delivery pipe. The public filled the reservoir over thousands of years. Acknowledging the public origin of the water does nothing if the water keeps flowing through a privately metered pipe with no mechanism to share the output.
The question now is how a society moves from recognizing the problem to constructing the plumbing that redirects a portion of the flow back to the village.
What Counts as the Public Resource
The first problem involves drawing the boundaries of the resource. Not all data looks the same, and different kinds of input may require different rules.
A novelist writing a book creates a specific product with clear authorship. When an artificial intelligence model trains on that book, the consumption is recognizable and the ownership trail is obvious. Copyright law already provides a partial structure for compensation, even though technology companies have largely ignored it.
Other inputs look nothing like a book. When a million drivers navigate a city grid while carrying smartphones, their collective movement trains a navigation model that becomes worth billions. No individual driver authored anything. No single trip has commercial value on its own. The commercial value emerges from the pattern created by millions of people moving through the same grid simultaneously, and no existing legal category cleanly captures who owns that pattern.
A hospital system anonymizes ten million patient records and feeds them into a diagnostic model. The model learns to identify cancerous tissue more accurately than a human radiologist. The patients whose scans trained the model receive no compensation and in most cases have no idea their data was used. The hospital may not have been compensated either. The company selling the diagnostic software captures the full commercial value of a product built on the biological information of millions of strangers.
The public resource powering machine intelligence contains published creative works alongside behavioral data, medical records, scientific research, satellite imagery, government filings, conversation transcripts, and the accumulated cultural output of every civilization that ever committed its knowledge to a recordable medium. Defining this resource clearly is the first requirement of any institutional response, because the boundaries of the resource determine the boundaries of the public claim.
The Social License
Private technology companies contribute enormous capital to the process. They hire talented engineers, consume huge amounts of energy, and build physical infrastructure. The public argument does not dismiss the value of these contributions. The pipe is real.
The argument is that building the pipe does not grant permanent ownership of the reservoir or unlimited control over the water supply. A social license sets the terms of that relationship. Private firms develop and operate the software. Their right to profit from collective human knowledge remains conditional on returning a share of that value to the society generating it. The company earns a return on its investment while recognizing that the public owns the underlying resource.
This structure has historical precedent beyond the technology sector. Broadcast television and radio operate under federal licenses that impose public interest obligations in exchange for access to a shared public resource, the electromagnetic spectrum. Mining companies pay royalties to governments for extracting minerals from public land. Pharmaceutical companies receive patent protections that eventually expire, returning the knowledge to the public domain.
The pattern is consistent. When a private company profits from a shared public resource, the society granting access to that resource retains the right to impose conditions. Machine intelligence fits the same pattern at a much larger scale because the "resource" in this case is the entire recorded output of human civilization.
Three Ways to Build the Institution
A society could approach the collection and distribution of this wealth through at least three different paths, each with distinct advantages and vulnerabilities.
The first path relies on an automation tax or royalty. A company deploying a machine learning model pays a levy based on its revenue, the computing power it consumes, or the number of human jobs it displaces. The money flows into a public treasury and funds universal services. South Korea introduced a version of this concept in 2017 by reducing the tax deductions available to companies that replace workers with automation. The European Parliament has debated a "robot tax" multiple times. The advantage of this approach is administrative simplicity. It uses existing tax collection structures and requires no new ownership arrangements. The weakness is that it treats the symptom rather than the cause. A tax redistributes revenue after extraction has already occurred, without giving the public any structural ownership of the resource or any governance role in how the technology develops.
The second path creates a publicly owned permanent fund. This structure mirrors the Alaska Permanent Fund, which has distributed oil revenue directly to residents since 1982. A national machine-intelligence fund would collect royalties from technology companies and distribute a universal payment to every citizen. Norway operates a similar model through its Government Pension Fund, which channels oil revenue into a sovereign wealth pool worth over 1.7 trillion dollars. The advantage of a permanent fund is that it converts a depleting public resource into a growing financial asset that compounds over time. The weakness is that a cash dividend alone does not address the governance problem. The public receives money while the private companies retain full control over the models, the training data, and the decisions about which products to build and which to suppress.
The third path explores cooperative or public ownership of foundational models. Under this approach, a government or international body builds and maintains large-scale models as public utilities, similar to how many countries treat electricity grids, water systems, and postal services. The public retains audit rights, transparency requirements, and limits on how much power any single private entity can concentrate. France invested 2.5 billion dollars into sovereign AI development in 2024 to reduce dependence on American private models. The advantage of public ownership is that it addresses governance, distribution, and concentration simultaneously. The weakness is the scale of capital and talent required to compete with private companies already operating at enormous size.
These three paths are not mutually exclusive. A society could layer an automation tax on top of a permanent fund while investing in publicly owned foundational models. The strongest institutional design probably combines elements of all three, capturing revenue in the short term while building public infrastructure for the long term.
Who Receives the Benefit
A universal cash payment is one option. Others include expanded public healthcare, education, housing, shorter working hours, transition grants for displaced workers, or shared ownership stakes in automated enterprises. The strongest model likely combines a universal dividend with universal basic services, distributing both cash and access to foundational infrastructure that no individual can reasonably purchase alone.
The distribution question matters because it determines whether the public dividend functions as a genuine structural correction or as a token payment that leaves the underlying power imbalance untouched. A small annual check does very little if the recipient still cannot afford housing, healthcare, or education. A comprehensive approach treats the dividend as part of a broader public settlement that includes access to the services automation was supposed to make cheaper.
The Danger of Waiting
Reversing the concentration of wealth becomes far more difficult after the infrastructure hardens. Once a small number of companies control the largest models, the computing infrastructure, the distribution channels, and the accumulated behavioral data of billions of people, those companies gain the ability to shape public opinion, influence elections, and block regulatory reform.
Every historical precedent confirms this pattern. Standard Oil became harder to break up the longer it operated. The tobacco industry delayed regulation for decades after the science was settled. Financial institutions became "too big to fail" because regulators waited until the concentration was irreversible. The primary challenge facing the public in 2026 centers on establishing claims to this abundance before private extraction becomes permanent.
The reservoir took thousands of years to fill. The pipe was built in less than a decade. The window for deciding who controls the water is closing now.