Article based on discussions with Arthur on 30th September 2026…

Artificial intelligence is usually discussed as a contest between corporations and states, or as a threat to jobs and human control. A more useful question is who will own the systems, knowledge and productive power that AI is creating, and whether they can be organised for the public good.

A political question disguised as a technical one

Debate about artificial intelligence tends to concentrate on spectacular possibilities: machines escaping human control, mass unemployment, disinformation, cyberattack and the prospect of conscious non-human intelligence. These questions matter, but they can obscure the political issue already before us. AI is being built inside an internet dominated by a small number of corporations and increasingly shaped by competing states. The immediate struggle concerns ownership, access and control.

The same technologies could deepen corporate power and state surveillance, or help create a society with far less compulsory labour and much wider access to knowledge. Neither outcome follows automatically from the technology. The result will depend on institutions, political organisation and social struggle.

This is why the argument cannot be reduced to whether AI is good or bad. The more important distinction is between proprietary systems controlled from above and public systems that can be examined, reproduced and improved by a wider community. That distinction also affects the resilience of the internet, the future of copyright, the development of robotics and the distribution of the wealth that automation can produce.

The fragile corporate internet

The internet was designed as a distributed network able to route information around broken links. In everyday life, however, most people reach it through a few corporate gateways. Google, Apple and Microsoft mediate access through phones and computers, while large social-media companies determine much of what users see. The network may be decentralised in principle, but its practical use has become highly concentrated.

This concentration creates political as well as technical vulnerabilities. A severe cyberattack could damage communications, banking, payment systems and other essential infrastructure at the same time. Governments might respond by restricting communication in the name of suppressing disinformation or subversion. Even without a catastrophic failure, algorithmic feeds encourage people to remain inside self-reinforcing circles, consuming material selected to keep them engaged rather than to support serious discussion.

Concerns about AI-generated deception are legitimate. Powerful systems can manufacture convincing text, voices, photographs and video on a vast scale. They may be used for fraud, political manipulation and coordinated attacks on communications networks. The predictable official response will be demands for tighter control over digital identity, software and online communication. Centralised monitoring will be presented as necessary to distinguish authentic information from manipulation.

The infrastructure created to suppress fraud can also suppress dissent. Most people access the digital world through devices controlled by large companies. Their operating systems determine what software can be installed and what communications are permitted. The hardware contains components and low-level software that ordinary users cannot inspect or change. Some restrictions serve real security purposes. Digital signing, for example, helps prevent malicious software from impersonating a trusted application. Yet the same machinery can ensure that only officially authorised programs operate.

A serious cyberattack could provide the pretext for expanding such controls. If banking or communications systems were disrupted, authorities might restore access only through approved channels. Political communication could then be monitored, restricted or traced far more easily. Freedom of information must therefore be treated as part of AI safety. A society protected from rogue software but unable to debate how it is governed would not be safe. It would be authoritarian.

Resilience requires more than stronger corporate security. Essential communication should not depend upon a single provider. Open operating systems such as Linux and decentralised social networks based on independent relays offer a possible foundation. Users should be able to choose the feeds they follow, receive them chronologically and maintain contact without relying on a company that owns the whole service. Engineers also need secure, independent channels through which they could coordinate restoration work if normal corporate systems failed. This is not an assertion that collapse is inevitable. It is a recognition that public communication should not depend entirely upon platforms organised around advertising, data collection and capital accumulation.

A modern republic of letters

A democratic information system would amount to a modern republic of letters: an international network through which people communicate, exchange knowledge and develop ideas without depending completely upon governments or corporations. During the seventeenth and eighteenth centuries, intellectuals living under different monarchies corresponded across borders. Their letters allowed scientific and political ideas to circulate despite censorship and religious division. The network was restricted to a small educated elite, but it demonstrated the value of communication independent of any single state.

Today most of the world’s population is literate and tertiary education is far more common. Billions of people could participate in an expanded republic of letters. That would require decentralised methods of authentication and communication. People should be able to establish that messages come from a consistent source without needing an identity issued by Google, Microsoft or a government. Cryptographic signatures can provide continuity while still allowing pseudonyms where they are necessary for safety or political freedom.

Open protocols could relay material among many independent servers and devices. If normal networks were blocked, information could still be moved physically on portable storage – a sneakernet. No such arrangement would make surveillance impossible, but it would make centralised suppression more difficult. People need not all become computer experts, yet they do need a basic understanding of the systems through which they communicate. Political movements that rely entirely upon opaque corporate platforms remain vulnerable to sudden exclusion, monitoring and manipulation. The struggle for democratic AI is also a struggle for open standards, user-controlled devices and independent channels of communication.

A global public library for the age of AI

Freedom of information should extend beyond the ability to communicate. It should include universal access to humanity’s accumulated knowledge. A global public digital library should therefore stand at the centre of any democratic AI policy. It would preserve books, research, newspapers, music, film and other cultural material while supplying verified sources and computing infrastructure for public AI. Distributed across several countries, it should remain available even if particular networks or data centres failed.

The idea is technically plausible because much of the world’s published knowledge has already been digitised. Technology companies, universities, state institutions and unofficial archives hold enormous collections. AI companies have assembled vast bodies of text and images for training, often without conventional permission from every copyright holder. File-sharing networks and shadow libraries have also shown that digital material can be copied and preserved globally. Yet this knowledge is divided among private corporate datasets, legally contested archives and public institutions that lack the resources or authority to combine it.

The public still encounters paywalls, incompatible catalogues and copyright barriers, while AI companies sell services built partly upon knowledge produced collectively over generations. A coalition of governments could create a properly catalogued library in which a reader could locate a work by author, title, ISBN or other identifying information. AI answers could link directly to the relevant record and underlying source, allowing users to verify what they had been told. This would address a persistent weakness of current systems, which can provide citations that are inaccurate, irrelevant or invented.

Universal access need not require writers, researchers, artists, editors and publishers to work without payment. Australia and other countries already operate public lending right schemes under which eligible authors and publishers are paid for library use. An international version could remunerate creators from public funds or levies on AI companies, rather than requiring every reader to purchase access to each work.

The details would be difficult. Any formula based on downloads or consultations would need protection against manipulation. Different forms of writing, research and cultural production may require different payment methods. Selection, editing and recommendation would still be necessary. These are practical problems to solve, not reasons to preserve a copyright system increasingly at odds with digital abundance. The guiding principle should be universal access combined with fair remuneration for creative and intellectual work.

Such a library would also challenge the advertising-funded media economy. Search engines and social platforms have absorbed much of the advertising revenue that once supported newspapers and publishers, while media organisations remain dependent on attention-grabbing content and proprietary access. A public library linked to public AI would separate access to knowledge from both advertising and monopoly ownership. It would replace artificial scarcity with a public system of access, verification and payment.

What genuinely public and open AI requires

A global digital library could provide the knowledge base for an international, publicly funded AI network. The leading models now are American corporate AI and Chinese state-directed AI. Neither is an adequate choice for the rest of the world. A third centre could consist of public institutions in several countries, cooperating through common standards and dedicated to research, education and human development rather than advertising, surveillance or military advantage.

Its software should be genuinely free and open source. The term open source is used loosely in AI. A conventional open-source program provides human-readable code from which the executable software is built. Many so-called open AI models provide a file of numerical parameters. Users may run or modify the model, but they cannot reconstruct its development or see precisely which data and decisions produced it.

Open parameters are useful, but they are not the same as a reproducible system. A genuinely public model would disclose its source code, training data, training methods, evaluation procedures and, where practical, intermediate checkpoints. Independent experts could test its behaviour and attempt to reproduce the final system. The public would gain more than assurances from a company’s own safety team.

The preferred standard should be free libre open source software under a reciprocal licence. Anyone could study and improve the system, but anyone distributing an altered version would also have to publish those improvements. Knowledge would accumulate as a commons rather than being drawn from public research and enclosed inside competing firms.

A public AI network could attract researchers who do not want their work controlled by a small number of companies engaged in a reckless race. Corporate salaries may remain higher, but many scientists and engineers would value the opportunity to work openly for institutions with a public purpose. Such a system would not make private AI disappear, nor would it prevent military agencies from developing secret models. It would nevertheless create a credible alternative and give democratic governments an independent source of expertise.

Open systems also require responsibility

Openness does not remove danger. A model that can be modified may be altered to eliminate safeguards. Someone with specialist knowledge might retrain it to assist biological weapons research, cyberattacks or other destructive activity. Completely closed systems concentrate power in unaccountable institutions, while completely unrestricted systems may be adapted for serious harm. There is no simple technical formula that eliminates this tension.

Public systems need rigorous evaluation, controlled access to especially dangerous capabilities and international cooperation. One part of the solution is authentication. Software can be digitally signed so that users know who produced it and whether it has been altered. A similar system could identify recognised public AI models and significant modifications.

Anyone distributing a modified model could be required to identify the changes and sign the new version. Independent organisations could test it and certify that it met defined standards. Users could then distinguish an authenticated public release from an anonymous or potentially dangerous modification. A decentralised web of trust would be preferable to placing all authority in one government or corporation. Different institutions and researchers could verify one another, creating a transparent record of responsibility.

Authentication would not prevent every malicious act. No security arrangement can. It would, however, clarify responsibility and provide a safer basis for open development. The answer cannot be permanent secrecy imposed by American corporations or dependence upon models released by an authoritarian state. Proprietary systems conceal too much, while parameter releases without reproducible training disclose too little. Public scrutiny must be combined with practical safeguards and enforceable standards.

A coalition of middle powers

The future of AI is often portrayed as a contest between the United States and China. Other countries are expected to watch, choose a side and accept the consequences. This underestimates their power. The AI supply chain is international. Advanced systems depend upon specialised semiconductor equipment, sophisticated chips, high-density memory, data centres, electrical infrastructure, research expertise and access to global markets.

The Netherlands plays an indispensable role in producing machinery used to manufacture the most advanced semiconductors. Taiwan is central to advanced chip fabrication, while South Korea supplies much of the specialised memory used in data centres. Japan, European countries, Canada, Australia, Singapore and other middle powers hold important knowledge, components, resources and financial capacity. Acting separately, they may struggle to influence American and Chinese giants. Acting together, they would possess considerable leverage.

A coalition could establish common rules governing the development and export of advanced AI hardware. Access to crucial machinery and components could be made conditional upon independent safety evaluation, transparency and compliance with international standards. Such a coalition would not control everything. China has a large internal market and is developing alternatives to Western technology, while American corporations possess immense political power. Coordinated action by countries controlling important parts of the supply chain could nevertheless alter the balance.

The same coalition could fund a distributed global library and reproducible public AI models. Each participating centre could examine the others, share improvements and maintain copies of essential data and software. The network could compete with proprietary American systems and Chinese state releases without becoming an instrument of either bloc. It would also create a practical basis for regulation because public authorities would possess models and expertise of their own.

The obstacle is political rather than purely technical. Governments remain reluctant to confront copyright monopolies, media interests and the companies controlling digital infrastructure. Yet AI is already disrupting those arrangements. Middle powers should stop behaving as powerless spectators. Together, they could become an independent force in governing AI and building a public alternative.

AI will not liberate us automatically

Technological progress does not determine its own political outcome. AI could reduce necessary labour and expand access to knowledge. It could also produce mass unemployment, unprecedented surveillance and a concentration of power greater than anything previously seen. The outcome will depend upon political organisation.

There is no organised international left presently capable of governing this transition. Many progressive people distrust parties and formal organisation, often with good reason. Existing organisations have repeatedly become bureaucratic, sectarian or detached from ordinary people. Rejecting failed forms of organisation, however, does not remove the need to organise. If democratic forces are absent, decisions will remain with the corporate executives, military establishments and political leaders who created the present crisis.

Past revolutions required long periods of experimentation, defeat and renewed struggle. AI may accelerate events, but it cannot remove the need for people to learn how to administer institutions and exercise power collectively. Younger generations have grown up amid economic insecurity, ecological crisis and declining trust in established authority. They are also more accustomed to digital collaboration than any previous generation. Whether this produces democratic transformation or a new authoritarianism remains undecided.

The task is not to wait for technology to deliver a new society. It is to build institutions and movements capable of directing technological development towards human emancipation. Public ownership, open knowledge, shorter working time and enforceable international safety standards are not automatic consequences of AI. They are political demands that require organised support.

A program for democratic AI

An immediate democratic program should connect the institutions needed to control risk with the social measures needed to share the benefits. It should include:

• Mandatory independent evaluation of frontier AI systems
• Internationally enforceable safety and transparency standards
• Cooperation among middle powers that control critical parts of the AI supply chain
• Protection of decentralised and encrypted communication
• Greater public knowledge of open operating systems and communication tools
• Creation of a universally accessible global digital library
• A new international system for remunerating writers, researchers, artists and publishers
• Publicly funded AI research centres using genuinely free and open source software
• Authentication and public certification of AI models and significant modifications
• Democratic control over AI in workplaces, government and essential services
• Shorter working time and secure access to income and public services as automation advances

The struggle over the future

These demands are not a complete answer. They are starting points for a political movement that does not yet fully exist. They join safety to ownership, freedom of information to public infrastructure, and technological progress to the distribution of time and wealth.

Artificial intelligence should not belong to a few corporations, military establishments or great powers. The knowledge used to create it should not remain enclosed behind paywalls and proprietary systems. AI is the product of humanity’s accumulated intellectual and social labour. It should be governed as a global public resource.

The technology could reduce drudgery, democratise knowledge and help address problems that have resisted solution for generations. Those benefits will not emerge from a corporate race for profit or a military contest for dominance. They require institutions that can make another future possible.

The most immediate danger is not that machines will inevitably become too intelligent. It is that human beings will continue surrendering their intelligence, knowledge and political power to institutions they do not control. AI could help humanity govern itself, but first humanity must decide to govern AI. The possibilities are bright. The road to them will be tortuous.