Open-source AI and the Choice Before Us

Throughout my engineering career, I’ve examined how complex systems operate under uncertainty, such as how an aircraft detects structural damage, how a power grid forecasts failures, and how models trained on incomplete data still generate valuable insights. A key lesson from this work persists as artificial intelligence (AI) transitions from research to daily applications: the trustworthiness of a system hinges on the assumptions it contains, and these assumptions depend on the quality of data, institutions and incentives that shaped them.

This is why the discussion on open-source AI warrants much more attention beyond expert circles. It goes beyond technical details and is fundamentally a debate about power and, specifically, who controls the assumptions built into AI systems and, ultimately, who benefits from them.

As AI plays an increasingly vital role in economic growth, public administration, healthcare, education and scientific discovery, the issue of openness will be crucial in deciding whether AI serves as a global public good or just another means of technological concentration.

The case for openness

Open-source AI appeals because it enables the broad community of researchers and developers to examine, test, challenge, and improve the code, model architecture, and underlying weights. This transparency shifts innovation from private decision-making to public scrutiny, fostering accountability. Additionally, it empowers countries, institutions, and communities to tailor AI systems to their specific needs instead of depending solely on foreign-designed technologies.

Whether it is a health ministry building an epidemic-response platform, a university aiming to develop local expertise, or a startup working in resource-limited settings, open systems provide a key benefit: autonomy. They can be customized to suit local conditions and deployed without ongoing licensing fees, which often influence access to advanced technologies.

In my view, the success of any intelligent system relies on the variety of information it can access. Open architectures increase this variety, while closed systems limit it to those with sufficient resources and authority. Therefore, openness isn’t just a technical decision; it broadens involvement in knowledge creation.

Openness without governance is not enough

However, openness alone isn’t enough. My experience with complex engineering systems shows that transparency doesn’t remove risk. A system that anyone can enhance can also be misused. An open model without proper safeguards might bring public benefits, but it can also be exploited for harmful aims.

We have seen open, loosely regulated AI systems used for disinformation, synthetic media and other activities that erode public trust faster than institutions can respond. 

he real challenge isn’t choosing between openness and governance but ensuring they develop together.

Open-source software communities have long recognized this principle. Successful projects rely not only on innovation but also on maintenance, security audits, documentation, quality control and well-defined governance structures. AI is no exception.

The future of open-source AI relies equally on institutions that can provide stewardship and on the engineers who code it. Accountability, auditing and ongoing oversight are not barriers to innovation; rather, they are essential for making innovation sustainable.

The representation problem

Openness alone cannot address a deeper issue. The development of most leading AI systems, whether open or closed, is concentrated within a few countries and institutions. As a result, these systems embody the languages, experiences, priorities and assumptions of those specific environments.

I have previously discussed the risks of interpreting statistical patterns in data as definitive truths about human societies. A model trained mainly on data from the Global North will naturally absorb many perspectives, biases, and blind spots present in that data. This influence can extend to language representation, cultural understanding, and assumptions related to economic activities, governance and social norms.

As a result, communities with minimal influence on AI development tend to face its impacts most directly.

Meeting this challenge demands more than just open code; it calls for open knowledge ecosystems. It involves establishing research infrastructures that promote inclusion from underrepresented regions, languages and communities. Most crucially, it means acknowledging that those impacted by AI should be active contributors to its development, rather than simply passive consumers of technologies created elsewhere.

From consumption to capability

This challenge has major implications for development policy. The world is often seen as divided between countries with advanced AI and those without. Yet, a more crucial divide may emerge among the nations that have access to AI technologies.

Some countries will primarily adopt systems developed abroad, whereas others have the skills, institutions, infrastructure and research capacity to develop their own systems independently. The difference between these two approaches is important.

Countries that only consume AI risk becoming dependent on others’ technological decisions. Conversely, those involved in creating and regulating AI gain more control over their digital future, foster stronger innovation ecosystems and build greater resilience in a fast-evolving global economy.

Open-source AI is one of the few ways to lower barriers to entry. It enables countries, universities, startups and public bodies to move from merely using AI to actively developing it. However, this shift will only occur if openness is paired with investments in education, research, digital infrastructure and governance.

Building the institutions that matter

At the United Nations University, many of our institutes aim to tackle this challenge. We are strengthening governance frameworks, expanding digital public goods, developing local research capacity and training public officials to understand the opportunities and risks of emerging technologies across regions and disciplines.

This work rarely receives the same attention as the release of a new model or a breakthrough benchmark. Yet history shows that institutions often have a greater impact than the inventions themselves. For technologies to truly transform societies, they need governance systems that can guide them to serve the public interest.

The future of AI depends not only on algorithm development but also on the quality of the institutions that govern it.

A choice about the future

The Secretary-General, António Guterres, emphasized the danger of a world split into AI ‘haves’ and ‘have-nots,’ a warning that remains pressing. We must also address a second division: the gap between those who develop AI and those who only use it.

Open-source AI offers an opportunity to reduce that disparity. It can democratize access to knowledge, boost participation in innovation, and promote broader sharing of AI’s benefits. However, these outcomes are not assured.

Technology itself doesn’t shape its future; societies do. The key question is whether openness will serve as a basis for genuine global engagement in AI or simply a means to perpetuate existing inequalities more efficiently. This outcome hinges not only on the technologies we develop but also on the institutions, skills, and governance structures we establish alongside them.

Open-source AI is more than a software model; it serves as a test of whether the emerging digital economy will prioritize concentration or participation, dependence or capability, and exclusion or inclusion. The outcome of this test will influence not only the future of AI but also the overall direction of development.

This article is based on a keynote address delivered during Open Source Week at UN Headquarters in New York in June 2026.