Artificial intelligence is no longer merely an emerging technology: it is now a strategic infrastructure, destined to profoundly redefine the mechanisms of production, knowledge systems, social organisation, and the distribution of political and economic power. It is becoming the new backbone of global transformation. Sam Altman, in his essay The Gentle Singularity, imagines a future in which AI will be a ubiquitous resource, accessible at marginal cost, and integrated everywhere in daily life, with effects analogous to those historically produced by electricity and the internet. A prospect that is no longer science fiction: in the United Arab Emirates, ChatGPT Plus has been made available free of charge to all residents, anticipating scenarios of universal access, yet managed through centralised agreements and control-oriented logic.
It is in this context that I want to propose a radical change of perspective: treating artificial intelligence not as private property to be monetised, but as a collective good, to be governed according to criteria of equity, transparency and distributive justice. The crucial problem is not only what we can do with AI, but who decides, who controls, who benefits. And today we are delegating these fundamental choices to a small number of private companies, guided by profit logic and opaque powers, removed from democratic oversight.
What is needed, therefore, is an alternative theoretical framework, solid and capable of inspiring new models of governance. Economic theories on the management of public goods represent a valuable foundation: they propose hybrid solutions that go beyond the opposition between State and market, and place commons, participatory governance and multi-level structures at the centre. From Ostrom to Lindahl, from Olson to North, these theories offer concrete tools for rethinking AI as a shared infrastructure oriented towards the common good.
Let us begin with an observation: artificial intelligence has characteristics that make it assimilable to a public good. It is (potentially) non-rival: the same model can be used by multiple actors without being depleted. And it can be non-excludable, if the data and algorithms are accessible and reusable. Like scientific knowledge or clean air, AI generates widespread benefits only if it is not fenced in by exclusive ownership logic.
This intuition is found in the classical vision of the pure public good. Erik Lindahl suggested that such goods can be financed in proportion to the benefit each actor perceives. Applying this principle to AI means building a system in which large companies — which obtain significant profits from the use of proprietary models — contribute to the sustainability of the AI ecosystem, enabling access also for those with lesser economic capacity: schools, local authorities, citizens.
However, equitable access is not enough. The issue of concentration of value must be addressed. AI platforms hold structural advantages: privileged access to data, unlimited computational power, the ability to influence markets and norms. Without redistributive mechanisms, the risk is that these rents consolidate, worsening pre-existing inequalities. This is where instruments such as a Windfall Tax on extraordinary rents come into play, or the introduction of shared profit clauses, obliging dominant players to invest part of their revenues in educational, health, environmental or social innovation projects.
But how to make these mechanisms operational? Solid and credible institutions are needed. The New Institutional Economics reminds us that the effectiveness of collective management depends on the quality of the rules, the capacity for enforcement, and trust in decision-making structures. For AI, this means creating open standards, independent audits, interoperability between systems, protection of sensitive data, algorithmic transparency. What is needed is a distributed institutional infrastructure, capable of operating on a global scale but with local sensitivities.
We cannot, however, blindly trust the State. The Public Choice Theory of Buchanan and Tullock warns that even public institutions can be inefficient, subject to capture by private interests, and inert in the face of change. Big tech knows how to exploit these weaknesses: they act as lobbyists, influence public opinion, and define the rules of the game. It is therefore indispensable to provide civil oversight mechanisms, instruments of radical transparency and genuine decision-making pluralism, which give space to communities, independent researchers, associations and citizens.
It is in this tension between centralised power and participatory governance that the vision of Elinor Ostrom proves illuminating. For those who, like me, identify with an anarchist and mutualist culture, Ostrom represents a powerful reference. She demonstrated empirically that communities, when equipped with clear rules, monitoring tools and inclusive decision-making processes, can manage common goods without central authority. AI can be approached with the same logic: cooperative consortia, civic data trusts, local ethics committees. As Édouard Jourdain suggests in The Self-governance of Collective Goods (Elèuthera, 2024), artificial intelligence can be not only for the community, but of the community.
But all this also presupposes mobilisation. The Theory of Collective Action of Mancur Olson highlights a recurring problem in public goods: individual disengagement. If everyone benefits, few feel incentivised to contribute. To overcome the free-rider risk, it is necessary to activate selective incentives — symbolic and material recognition, reputational mechanisms. In the AI world, this can mean rewarding those who share quality data, those who contribute to improving open-source models, those who participate in public decision-making processes.
Looking at the global scale, AI cannot be treated as a national technology. It is a planetary infrastructure. Models are trained on global corpora, impacts fall everywhere, and rules must reflect this complexity. Hence the need for polycentric governance: a system with multiple autonomous but interconnected decision-making centres. States can be read as Ostrom’s “communities” at macro scale, called upon to coordinate to guarantee equity, transparency, and interoperability.
In parallel, the logic of Network Governance is imperative: a model in which public, private and civic actors collaborate, building fluid alliances and common projects. In AI, this means co-designing regulatory tools, technical standards and ethical guidelines. Overcoming the verticality of command and replacing it with a web of shared responsibilities.
Some practical examples help to imagine this model: a European public platform for open and transparent AI models; a participatory local ecosystem for deciding the use of AI in sensitive areas such as health and education; an international fund for access to AI in low-income countries, financed by a global tax on digital rents.
The stakes are not only technological efficiency. They are systemic justice — the possibility of building an artificial intelligence that redistributes value, opportunity and power. An AI that is not an infrastructure of domination, but a tool of collective emancipation. Like every public infrastructure, AI too needs to be designed, financed, regulated and maintained. But above all it needs to be removed from extractive logic and brought back within a democratic, participatory, multi-level horizon. The challenge is enormous. But it is also a great opportunity to radically rethink the relationship between technology and society. And to imagine a future in which innovation truly serves the common interest.
