This article is translated from an excerpt of a conversation between Prof. Yin Zhiguang of Fudan University and Prof. Li Hansong of American University. The conversation was originally conducted in Chinese.

Prof. Yin Zhiguang:
We can see that there is a highly differentiated landscape not only within the AI industry itself, but also at the intersection between AI and other industries.

So the question arises: everyone is now discussing global AI governance, especially how to govern this new space created by AI. Since we have already recognized that there are highly differentiated patterns of development within this space, are there new ways of thinking that can help us understand these differences and, on that basis, build a new framework for global governance?

Prof. Li Hansong:
This question goes beyond inequality within the narrow sense of the global AI industrial chain. It touches upon the broader issue of global epistemic injustice. The governance of AI in this new era must be inclusive and participatory. And the most important expression of inclusiveness is data inclusiveness. This is the second dimension.

Artificial intelligence, especially large language models, appear to be built upon massive amounts of data. But in reality, the data used to train these models represents only an extremely limited portion of the totality of human experiential data. A vast amount of knowledge, experience, and culture continuously generated by people in their everyday lives has never been collected, let alone incorporated into the process of model training.

Therefore, AI itself is not an inherently inclusive system. On the contrary, from the very beginning of its data sources, it already possesses a certain degree of exclusivity. What kinds of data can enter databases? What kinds of data are recognized by major North American technology companies and engineers as “valid data”? This process itself involves choices, and those choices represent power.

One of the key tasks of AI governance, therefore, is to incorporate as much as possible the data that has long been excluded from mainstream databases, allowing more countries, regions, languages, and social groups to participate in the construction of AI knowledge systems.

I have seen some very interesting cases in Africa. For example, Uganda has witnessed the emergence of many local AI startups. Instead of relying on existing datasets from Europe and the United States, they use local data to train models, enabling AI to understand Uganda’s indigenous languages, culture, and social environment.

Another example is Ethiopia, which has 80 or 90 languages across the country. The development of local AI startups there is increasingly focused on these indigenous languages, rather than simply replicating large language models developed in the English-speaking world.

These practices demonstrate that AI development does not necessarily have to follow the path established by North American technology companies. Different regions can fully rely on their own data resources, linguistic environments, and social needs to explore development models that better suit their local realities. This diversification of data sources is itself an indispensable expression of inclusiveness in global AI governance.

Prof. Yin Zhiguang:
So, it is not merely about linguistic diversity.

Prof. Li Hansong:
Exactly. More importantly, it is about local data and local knowledge. Different regions possess different historical experiences, social environments, and knowledge systems, and all of these influence the sources of AI data. Different data produces different training outcomes for algorithms, and ultimately generates different forms of knowledge and judgment.

Therefore, this is not merely a technological issue; it is an issue of epistemic justice. Whose data can enter models? Whose knowledge can be learned and expressed by AI? And who is excluded from knowledge production? These questions directly concern fairness in knowledge production in the AI era.

Thus, AI development itself contains a form of epistemic inequality and injustice. The data entering models often reflects the experiences of only a small number of countries and social groups; the design of algorithms is also primarily controlled by a small number of technology companies and technical elites.

Therefore, the first task of global AI governance is to address the issue of inclusiveness. We must continuously expand the sources of data and knowledge, allowing more countries, languages, cultures, and social groups to participate in knowledge production. Only then can the perspectives generated by AI—especially when they involve value judgments and normative judgments—truly remain diverse, rather than becoming a “monologue” controlled by a handful of companies and countries.

The third dimension concerns future imaginaries.

Today, whether we are talking about human-machine integration or AI-assisted empowerment with humans remaining in control, these are actually not the most radical visions of the future. Some technological elites in Silicon Valley have already begun imagining far more extreme futures: for example, a small number of people migrating to Mars while leaving Earth to decline on its own; or a tiny minority using artificial intelligence, embodied robots, and other technologies to control the vast majority of resources, thereby dominating or even ruling over other human beings.

What these visions present is, in reality, a highly technocratic and hierarchical vision of the future. They also reflect a form of future imagination infused with elements of the “Dark Enlightenment.”

Prof. Yin Zhiguang:
To some extent, this is essentially the same kind of libertarian utopian imagination represented by Ayn Rand’s Atlas Shrugged. It suggests that ultimately only a small number of intellectual elites and technological elites deserve to survive, while the vast majority of people, in their view, have little value beyond obstructing technological progress and the advancement of human civilization. If that is the case, what should they do? They should simply construct a utopia of their own.

In Rand’s Atlas Shrugged, that utopia is an isolated valley separated from the outside world. In the techno-futurism represented by figures like Musk, this utopia becomes Mars. Only the chosen few are qualified to travel to Mars; as for Earth, it appears that it can simply be abandoned and left to its own destruction.

Prof. Li Hansong:
Yes. So I think this is actually a very particular form of technological theology. What is even more noteworthy is that this technological theology has also merged with certain Schmittian and Straussian political ideas, forming an elitist narrative about the future of society.

I believe this issue has never been adequately addressed or resolved within the United States itself. And because the current development of AI has long been centered around North America, this dark vision of the future has continued to spread globally.

Precisely because of this, China’s emergence has created another possibility. As China continues to open up new technological spaces and development pathways, if this space can continue to expand, it may help break the monopoly of the “Dark Enlightenment”-style dystopian imagination over visions of the future.

Prof. Yin Zhiguang:
This point particularly interests me. You have just raised a very important question: as AI spreads globally, what exactly is the mechanism through which it spreads? Earlier, we discussed that diversity must be incorporated into AI governance. This means both developing large language models capable of covering multiple languages and developing localized AI models that can integrate the actual conditions of different regions. But all of this rests on an even more fundamental prerequisite: the material foundation.

First, large language models themselves must become more accessible. Algorithms need to become affordable and more open. Second, the energy costs supporting AI development—especially electricity—must also become sufficiently low.

Otherwise, the diversity, inclusiveness, and even global governance we discussed earlier would merely be a “superstructure built in the air.” Only when people can afford AI, have access to stable and inexpensive energy supplies, and when large models themselves become sufficiently low-cost and open-source, can these governance goals truly become achievable.

And precisely in these foundational conditions, the efforts China has made in recent years deserve close attention.

Prof. Li Hansong:
Yes. Beyond open-source development, what is even more important is that AI must truly become integrated with social activities and economic production. In a sense, we may even slow down the extreme competition over the capability limits of generative artificial intelligence (GenAI) and large language models. Instead, we should pay more attention to what these technologies actually bring to ordinary people’s lives and what value they create for society. This is one direction.

Another direction is sustainable energy. AI development is highly dependent on electricity, meaning renewable and sustainable energy itself is also a crucial foundation for AI development. On this front, China has continued to invest heavily, while the United States has relatively lacked such a long-term strategy.

In recent years, changes in the international situation, including events such as the Strait of Hormuz crisis, have further highlighted the importance of energy security. Countries such as South Korea and Japan have also gradually recognized the vulnerabilities created by long-term dependence on imported energy and have begun increasing investment in renewable energy. The United States, by comparison, has made relatively limited progress in this area.

Therefore, AI development must not only be deeply integrated with society and the economy; it must also be built upon a more diversified and sustainable energy system.

Beyond this, there is an even more fundamental question: where does society actually want AI to go? This is essentially the AI governance issue we discussed earlier.

At present, Singapore is promoting ASEAN’s exploration of a regional AI governance framework. The African Union is also discussing related issues, and countries such as Indonesia are actively participating as well.

By contrast, the United States has long tended to allow innovation to run ahead like a “wild horse,” intervening with regulation only after problems have already become extremely serious.

Prof. Yin Zhiguang:
In a certain sense, this continues the governance path and institutional inertia formed during the neoliberal era in the United States. The emphasis has been placed more on the self-driven development of markets rather than on shaping the direction of technological development in advance through public governance.

From an epistemological perspective, this is actually what we often describe as a “castle in the air.”

It is like someone who eats six bowls of noodles and only feels full after the seventh bowl, and therefore mistakenly believes that only the seventh bowl actually mattered. Or it is like the parable from Mencius, where someone wants to build only the second floor while forgetting that the first floor is the foundation of the entire structure.

Many people seem to believe that as long as they possess the most advanced technologies, all other problems will naturally be solved. But in reality, development has always been a systemic undertaking.

Therefore, bringing human existence back into discussions of technological development and global governance, and truly grounding technology in the foundations of human social development, is, in my view, the most fundamental issue.

Prof. Li Hansong:
Yes. Today, we see such severe polarization within American society, yet people rarely reflect fundamentally on whether this very polarization is precisely the result of building “castles in the air” for so long while ignoring the real conditions of ordinary people at the bottom of society.

When more and more people are excluded from technological progress, they will naturally express dissatisfaction with this model of development. Even though these grievances have now transformed into very concrete political consequences, American society still lacks a systematic political reflection on the issue.

I believe this is deeply disappointing.

Prof. Yin Zhiguang:
Then, looking from another perspective, if the Global South has come to recognize this problem, how should we promote a form of technological development and global governance that places human agency at its center?

Prof. Li Hansong:
The key lies in collaborative governance.

For a long time, resources and discursive power have been concentrated in the hands of a small number of countries and enterprises. As a result, many international rules have not truly reflected the interests of the majority of developing countries.

During the American War of Independence, a famous phrase was put forward: “No taxation without representation.” Yet for a long period of time, Western countries have dominated global rules without broad representation, while allowing other countries to bear the corresponding costs.

Therefore, when we formulate international rules again in the era of AI and the knowledge economy, one fundamental principle must be: joint formulation and joint governance.

Only through genuine participation in rule-making will countries recognize the legitimacy of these rules and be willing to follow them together. Otherwise, if the rules were not created with my participation, why should I accept them or abide by them?

I believe this is the most fundamental principle for future global AI governance.

Pror. Yin Zhiguang:
This also happens to align perfectly with the themes of the World Artificial Intelligence Conference currently being held in Shanghai. The emphasis of this year’s conference on collaborative governance, inclusive development, and related issues are precisely the key concepts we have been discussing.

Prof. Li Hansong:
Yes. The key question is how to form a shared wisdom regarding artificial intelligence, use this shared wisdom to gradually build common understanding, and ultimately establish norms that everyone can follow together.

Of course, these norms cannot be perfect from the very beginning. They will inevitably undergo continuous adjustment and refinement. But as long as this process of accumulation continues, they will gradually mature and eventually form a truly universal governance mechanism.

In the past, a Hegelian view of history held that the West represented universality, while other civilizations were merely local or marginal. They might be interesting, but ultimately they were only “interesting” and nothing more.

Today, if the Global South can genuinely participate in constructing universal rules, then it will no longer merely be a recipient of rules. It can also become a co-creator of universality.

Prof. Yin Zhiguang:
I agree. However, I do not think our goal is to reproduce a new kind of “Hegelian universality.” It is not about replacing someone else’s universality with our own, nor is it simply about “bringing the Global North into the fold.”

What I think deserves more attention is that, in today’s discussions of global governance and AI governance, we are bringing back a dimension of moral philosophy.

Fairness, justice, the pursuit of goodness… These concepts may appear abstract, but in reality, they run throughout the development of human civilization. Regardless of civilization or culture, human beings cannot escape the pursuit of goodness, fairness, justice, and shared coexistence.

Unfortunately, since the 19th century, and especially since the 20th century, development theories and global governance practices led by the United States have increasingly emphasized instrumental rationality, attempting to exclude these moral questions—which may appear vague but actually embody humanity’s deepest values—from consideration.

Today, whether through the World Artificial Intelligence Conference or various initiatives and white papers on global AI governance, there is an effort to bring values such as fairness, justice, freedom, and democracy back into discussions of AI governance.

Prof. Li Hansong:
Precisely because of this, morality itself possesses universality.

Human beings are inherently beings who pursue moral ideals. We naturally care about fairness and justice, and we care about how we can live together. Therefore, the “pulling in” or “bringing along” we mentioned earlier does not refer to coercive inclusion. Rather, it refers to a kind of moral attraction.

For those countries that were initially unwilling to participate in collaborative governance, when a governance framework emerges in the world that is more inclusive and possesses greater moral legitimacy, they will have a new choice.

Ultimately, everyone will choose the governance system they are willing to participate in based on the values embodied by different norms. This is not coercion in the realist sense; it is closer to what we might call “normative realism.”

Prof. Yin Zhiguang:
Or perhaps we could say that it is not about being forced to join, but rather being gradually attracted, incorporated, and ultimately forming new consensus through shared practice.

Prof. Li Hansong:
Exactly. As the saying goes: “If those far away are not convinced, cultivate virtue and refinement so that they will come.”

Through continuously improving one’s own institutions and values, more people will voluntarily choose to participate.

Prof. Yin Zhiguang:
This is a kind of harmony that naturally emerges through practice. I think it is particularly meaningful, and it is also very suitable as the conclusion of today’s discussion.

Prof. Li Hansong:
Because ultimately, it comes down to the word “common” — common governance, common development, and common life.

Prof. Yin Zhiguang:
And behind this, there is an even deeper foundation: morality.

Prof. Li Hansong:
In fact, before the rise of instrumental rationality, the Western tradition itself was also deeply grounded in moral concerns.

This includes the tradition of natural law, as well as the basic principles concerning human rights and humanitarianism in modern international law. To a large extent, these ideas all originate from natural law thinking.

Ultimately, these traditions are addressing the same question: How can humanity live together better?

Prof. Yin Zhiguang:
This is actually a very interesting point.

Europe has a tradition of natural law, while China likewise has its own pursuit of universal moral principles. Although China does not have the concept of “divine law,” it has always emphasized concepts such as the “Way of Heaven” (tiandao) and “following nature” (fazi ziran), while also caring about universal human values such as fairness, justice, and sustainability.

These values are precisely the greatest common ground among different civilizations.

At the same time, we have always emphasized the need to understand the relationship between freedom and unity, conflict and cooperation, through a dialectical perspective, and to understand the world through its interconnected whole.

I think that although today’s discussion may have appeared to cover many different topics, it has ultimately revolved around one central question:

Under the existing structure of inequality, how should we think about the future of human order?

Editor: Zhiyu Wang