AI governance needs more than safety rules. Treating its knowledge base and benefits as the common heritage of humankind offers a practical way to protect epistemic diversity and prevent a global crisis of never-skilling.
Artificial Intelligence (AI) is presented as the product of a handful of companies. Yet its capabilities were built from a much larger inheritance: generations of scientific discovery, public investment, language, art, labour and cultural knowledge. Much of the information used to develop AI was scraped without creators’ knowledge or compensation; some training libraries even contained pirated works. The US Copyright Office has concluded that some commercial uses of copyrighted material for AI training may exceed fair use. The governance question is therefore not simply who owns an AI model. It is who may extract humanity’s accumulated knowledge, convert it into private power, and decide how its benefits are distributed.
The principle of the common heritage of humankind offers a useful principle to correct this – and provide guidance for its governance for future generations. This principle already governs the deep seabed beyond national jurisdictions under the United Nations Convention on the Law of the Sea. Its core ideas are non-appropriation, international administration, peaceful use, equitable benefit-sharing and stewardship for future generations. AI is not legally recognised as common heritage, and the analogy should not erase copyright, sovereignty or Indigenous rights. But these principles can supply the normative foundation that fragmented AI initiatives currently lack.
Plural Knowledge Requires Shared Authority
The first problem is epistemic. UN Trade and Development reports that 118 countries, mostly in the Global South, were absent from the seven major international AI governance initiatives it examined. Their exclusion shapes not only who writes the rules around AI, but what AI is capable of knowing. A system may communicate in hundreds of languages while reproducing narrow assumptions about autonomy, rationality, wellbeing, and even what constitutes legitimate knowledge. Multilingual AI is not necessarily epistemically plural AI.
This distinction is crucial for Indigenous knowledge. Concepts such as Country in Australia or Vanua in some parts of the Pacific do not simply describe land. They express complex relationships among people, ancestors, place, spirituality, and responsibility. They are based on different temporal and experiential registers. Converting such knowledge into retrievable data may preserve information while destroying the relationships that give it meaning. There are also crucial questions about whether Indigenous concepts are fully translatable into Western linguistic and classificatory categories.
Inclusion alone is therefore inadequate. Communities must help determine how their knowledge is classified, interpreted, accessed, and used. The CARE Principles for Indigenous Data Governance – collective benefit, authority to control, responsibility and ethics – provide one workable foundation. If coupled with a common-heritage approach, this could turn epistemic pluralism from aspiration into institutional authority.
Global rules should require disclosure of significant training sources, community-defined protocols for culturally sensitive data, continuing rights to contest or withdraw wrongful uses, and representation for source communities in the bodies governing shared datasets and foundational models. Common heritage must not mean placing everyone’s knowledge in an unrestricted global public domain. Rather, it means shared stewardship that recognises the preferential rights of the communities from which knowledge originates.
The Labour Crisis and Never-Skilling
Much discussion of global AI inequality still focuses on access: who has connectivity, computing power, digital skills, or access to advanced models.
These matter. But access alone can actually conceal a deeper dependency regarding labour. Here, commentary usually focuses on displacement or deskilling. A deeper danger is never-skilling: people may use AI to perform tasks before acquiring the capabilities that those tasks once developed.
Consider Timor-Leste. Factors found to be associated with high national AI uptake include income, digital infrastructure, human capital, English fluency and a young population. Timor-Leste scores poorly on most of these measures except youth, yet AI use is already substantial in the country. This is not necessarily evidence of technological readiness. Rather, as observed by Wayan Vota, “It shows a young population reaching a free product over borrowed pipes.”
A country can therefore become a highly proficient downstream user of AI while possessing little capacity to determine how those systems are built, what knowledge they privilege, or whether access might eventually be restricted. This can lead to new vulnerability: a young population accessing powerful technologies through infrastructure, models and knowledge systems controlled elsewhere.
This danger of AI is unevenly distributed internationally. Wealthier states may experience occupational restructuring while retaining strong universities, technical industries, and regulatory capacity. Developing states face the possible narrowing of labour-intensive development pathways as automation weakens the advantage of lower-cost labour. The International Labour Organization estimates that one in four jobs is potentially exposed to generative AI, with transformation more likely than wholesale replacement.
Transformation without bargaining power or domestic capability can deepen dependency rather than development. Here, common heritage adds more than a demand to share profits. Equitable benefit-sharing should be defined as capability-building. States and companies benefiting from globally sourced knowledge should contribute to an international AI commons fund supporting public universities, local-language datasets, public-interest compute, vocational training, worker transition and independent regulatory expertise in developing countries. Access to models is not enough. Although open-sourcing can widen access, enable independent scrutiny and reduce dependence on proprietary providers, it is not a substitute for AI governance. Openness alone cannot ensure accountability, fair benefit-sharing, data rights, safety or meaningful participation in decisions over how AI is developed and deployed.
Countries need the capacity to adapt them, audit them, and decide when not to use them.
Intergenerational stewardship also requires a human-capability test for AI deployment. Education systems, employers and governments should ask which capabilities an application augments, which it substitutes, and whether learners and junior workers retain protected opportunities to practise underlying skills. Human development requires opportunities to acquire, exercise and renew skills throughout the life course – not merely retraining after technological displacement. Procurement and workplace agreements could require human-led training stages, worker consultation and evidence that automation improves job quality. Benefit-sharing would then preserve the social conditions in which future generations can become capable agents rather than permanent downstream users.
How to Make the Principle Practical
States can advance this agenda through the United Nations, UNESCO and its regional partnerships. They can begin by treating Indigenous data sovereignty as a governance principle, not an afterthought; supporting smaller states to develop local-language infrastructure and negotiating expertise; and advocating international benefit-sharing tied to capability development.
The United Nations and the wider multilateral system should advance this agenda supporting states – particularly smaller and developing countries – by establishing international benefit-sharing mechanisms tied to lasting capability development. Having an international, democratic body for deliberation may seem a lofty ambition but is essential. This would complement the UNESCO Recommendation on the Ethics of Artificial Intelligence, which already links AI governance to cultural diversity, inclusion and human agency, by translating these commitments into coordinated institutions, resources and enforceable responsibilities.
A genuinely common heritage cannot be created by declaring unlawfully extracted resources common after the fact. It requires restitution where extraction was wrongful, authority for communities whose knowledge is involved, and institutions that distribute both benefits and technological capability. The objective is not to make every algorithm globally owned. It is to prevent the accumulated knowledge of humanity from becoming the exclusive infrastructure of a few corporations and powerful states.
The decisive question is not how intelligent AI becomes. It is whether artificial capability enlarges or diminishes the capacity of people and societies to understand, create, work and determine their own futures. No company and no state should answer that question for humanity alone.
Dr. Shannon Brincat is a Senior Lecturer in Politics and International Relations at the University of the Sunshine Coast. His research focuses on critical international relations theory, climate change adaptation, dialectics, recognition, and the imagination in global politics. He has undertaken projects related to climate and community organisation in India, Mozambique, Fiji, Papua New Guinea, and Timor-Leste. He has edited or written 10 books and over 50 papers in international journals. His most recent book is Dialectical Dialogues in Global International Relations published in the Voices in IR Series at Oxford University Press, brings different dialectical traditions into critical dialogue to illuminate the tensions, contradictions, and possibilities for transformation within global politics. He is also a co-editor of Global Discourse. For more information see: shannonbrincat.com
This article is published under a Creative Commons License and may be republished with attribution.
