Access Is Not Power
A Response to “The Future Is for Everyone”
Mark Zuckerberg’s vision of personal superintelligence is compelling because it begins with an idea that is difficult to oppose: the future should belong to everyone. Artificial intelligence should expand human possibility rather than diminish it. It should help ordinary people invent, learn, create businesses, improve their health, and exercise greater control over their lives. Its benefits should not be confined to governments, corporations, or a small technological elite.
These are admirable principles. The problem is that the argument quietly substitutes access for power.
Giving everyone access to a tool does not mean giving everyone control over it. Billions of people already have access to Facebook, Instagram, and other digital platforms. That does not make them joint owners of those platforms, participants in their governance, or equals of the companies operating them. They may use the system, but they do not control its infrastructure, algorithms, economic incentives, rules, or continued availability.
The same distinction will matter far more with artificial intelligence. A personal agent may feel like an extension of the individual while still depending entirely on infrastructure owned by someone else. The company operating that infrastructure may determine which model the agent uses, what it remembers, what it is permitted to do, how much compute it receives, what it costs, and whether it continues to exist.
That is widespread access, but it is not necessarily distributed power.
“The Future Is for Everyone” presents a choice between two possible worlds. In one, superintelligence is centralized in the hands of a few institutions. In the other, it is distributed to billions of individuals through personal agents. But this is a false choice. A technology can be widely distributed at the level of use while remaining extraordinarily concentrated at the level of ownership.
If billions of personal agents depend on a handful of companies controlling the data centers, chips, models, energy supplies, operating systems, and distribution networks, then superintelligence has not truly been decentralized. It has merely acquired billions of user interfaces.
The essay’s historical argument is similarly attractive but incomplete. Previous technological revolutions eventually produced greater prosperity, health, and freedom, so we are told that AI is likely to do the same. But history does not move so cleanly. Industrialization increased human productivity, but it also produced generations of brutal working conditions, environmental destruction, monopoly power, colonial extraction, and displaced communities. The benefits were not distributed automatically. They were distributed through political conflict, labor organization, regulation, public investment, education, and changes in ownership.
Technology created wealth. Society determined who received it.
The fact that humanity eventually benefited from earlier inventions does not prove that every new technology will benefit everyone, that the gains will arrive quickly, or that the institutions creating the technology should be trusted to determine its distribution. History gives us reasons for hope, but it does not give us a guarantee.
The essay’s stories of the Wright brothers, Michael Faraday, and the young computer entrepreneur in a garage make the same mistake on a smaller scale. They demonstrate that outsiders can produce great inventions. They do not prove that giving everyone AI will create broadly distributed prosperity. These stories remember the rare individuals who transformed the world while overlooking the millions of equally ambitious people whose circumstances prevented their ideas from succeeding.
Invention does not occur through intelligence alone. It requires time, capital, materials, legal rights, social networks, manufacturing, distribution, and the freedom to fail. A superintelligent agent may help a person design a product, but it cannot guarantee that the person will own the resulting intellectual property, receive financing, reach customers, compete with established companies, or retain the economic value created.
The economic argument contains another unsupported leap. People have unlimited desires, the essay says, so new technologies will produce new businesses and jobs. That may happen, but human desire does not necessarily create demand for human labor. If AI can perform both the old work and the newly invented work, an economy may produce more goods and services while employing fewer people.
Productivity and employment are not the same thing. Productivity and broadly shared prosperity are not the same thing either.
A company with ten employees and thousands of AI agents may be highly productive. A million such companies might create extraordinary economic output. But that does not tell us whether most people will earn a living, own productive assets, or possess meaningful bargaining power. The economy can become more entrepreneurial in appearance while ownership becomes more concentrated in practice.
The question is not simply whether AI will create value. It almost certainly will. The question is who will own that value.
The essay attempts to answer this concern through a theory of balance. If only one person has a superintelligent lawyer, that person possesses an unfair advantage. If everyone has one, justice supposedly becomes more equal. If everyone has cybersecurity superintelligence, systems become safer. If every business has AI, markets become more dynamic.
But equal access to a category of technology does not produce equal power. One person may have a free agent operating under strict limits, while another corporation can deploy ten thousand agents using vastly greater compute. In court, wealthy parties would still possess more time, influence, data, and capacity to overwhelm opponents. In cybersecurity, attackers need to find only one weakness while defenders must protect entire systems. In business, the company controlling the infrastructure may observe the whole market while each individual sees only a fragment of it.
Giving everyone a weapon does not necessarily create peace. Sometimes it creates an arms race.
Democratic checks and balances do not work merely because opposing parties possess comparable abilities. They depend on laws, institutions, enforcement, transparency, legitimacy, and limits on behavior. Competing AI agents do not automatically reproduce constitutional government. Without common rules, competition among powerful systems may magnify conflict rather than balance it.
The essay is especially vulnerable when it treats adoption as evidence of alignment. It argues that people will not adopt personal agents unless they trust them. Therefore, if billions of people adopt these agents, the systems must have become aligned with individual interests.
History provides little reason to believe this.
Billions of people use products that exploit their attention, influence their behavior, collect their data, or act against their long-term interests. People often adopt technologies because everyone else has adopted them, because their employment requires them, because alternatives have disappeared, or because the harm is difficult to see. Commercial success is not proof of safety, trustworthiness, or alignment.
Nor is following a person’s stated goals necessarily the same as serving that person’s interests. People have contradictory desires, incomplete information, destructive impulses, and conflicts between what they want today and what they hope to become tomorrow. An agent designed to maximize immediate preferences may be personally obedient while remaining profoundly harmful.
There is also a contradiction at the center of the privacy promise. The proposed agent will understand a person’s health, finances, career, relationships, home, interests, and private communications. It will monitor behavior, remember personal history, place orders, communicate with outside services, and operate continuously.
This is compared to the encryption used by WhatsApp. But encrypting messages is not the same technical or institutional problem as operating an agent that must continually act on private information. The promise may someday be achievable, but the analogy does not prove it. We are being asked to entrust a corporation with the most comprehensive representation of ourselves ever assembled while accepting the corporation’s assurance that even it will be unable to see it.
The geopolitical argument introduces an additional danger. Any delay by an American company may allow a foreign competitor to move ahead. Therefore, even a one-month delay could endanger American leadership. Risk must be addressed, but not in a way that slows deployment.
This is the logic of an arms race. Every participant can acknowledge the danger while insisting that restraint is impossible because someone else may continue. Under that logic, greater risk becomes an argument for greater speed. If AI is safe, it should be deployed. If it is dangerous, it must be deployed quickly so that responsible actors are not overtaken. Almost every possible observation leads to the same conclusion: accelerate.
A theory that recommends the same action regardless of the evidence is not a meaningful safety theory.
The deepest contradiction is that the essay warns against allowing a small number of institutions to determine humanity’s future while presenting Meta as the institution best positioned to prevent that outcome. It asks us to distrust centralized power in principle while trusting one of the most powerful corporations in history to build the models, provide the agents, operate the infrastructure, define alignment, set the prices, protect the data, cooperate with governments, and decide when a system is safe to release.
An independent corporate board is not distributed public power. Affordability is not ownership. Open access is not democratic governance. A free version of a product is not a constitutional right.
None of this means that the vision should be rejected. Personal AI could become an extraordinary instrument of human creativity. Broad access is preferable to exclusive access. Privacy, pluralism, open inquiry, and protection against concentrated power are worthy principles.
But if the future is truly for everyone, people must possess more than permission to use it.
They need meaningful rights over their data, the ability to move their agents between providers, genuine model transparency, independent oversight, protection from manipulation, access that cannot be withdrawn arbitrarily, and some form of participation in the economic value their lives and work help create. Society must discuss ownership, labor, taxation, competition, liability, and democratic governance—not merely access.
The future will not belong to everyone simply because everyone receives an agent.
It will belong to everyone only if the people operating those agents have enforceable power over the systems behind them. Otherwise, personal superintelligence may become the most intimate and capable technology ever placed in human hands while remaining one of the most centralized technologies ever constructed.
That is the question the essay does not answer:
When everyone has a superintelligent agent, who owns the intelligence, who controls the infrastructure, and who can turn it off?
Until those questions are answered, “the future is for everyone” remains a promise made by the people who expect to own the future.