The Beijing AI Principles

In the [YOU] on AI Field Guide

[YOU] on AI asks what it means to see the AI transition clearly, without the narcotic of hype or the paralysis of fear. The Beijing Principles enter the cycle as evidence that what we see depends on which tradition we are looking from. A governance framework that begins with individual rights will build different guardrails than one that begins with collective harmony—not because either is obviously wrong but because the ordering of values encodes a theory of what matters most, and different civilizational traditions have different theories. The cycle's portrait of AI as a planetary phenomenon, reshaping not just individual cognition but the social fabric in which individuals think, finds in the Beijing Principles the only major governance document that treats that social fabric as the primary unit of analysis.

The principles also illuminate the cycle's recurring observation that the hardest AI problems are not technical. The gap between stating that AI should be harmonious and specifying what harmoniousness requires of a particular algorithm is enormous—both normative and empirical. You cannot close it with a policy document, however precise. Yi Zeng has spent subsequent years trying to close it with institutional mechanisms and technical tools, and his partial success is a measure of how large the gap actually is.

Origin

By 2019, the global AI governance conversation had crystallized around themes largely set by Western academic and policy circles: fairness, accountability, transparency, safety, privacy. China had been largely absent from this agenda-setting. The Beijing Principles were Yi Zeng's intervention: a document that joined the conversation while reframing its terms. It was not a purely Chinese document—it endorsed openness, cooperation, and global inclusion—but it was distinctly Chinese in its emphasis, introducing Confucian vocabulary (he xie you hao: harmony and human-friendliness as a paired concept carrying freight that safety alone does not) and positioning AI governance within the frame of the UN Sustainable Development Goals.

Most Western commentary either treated the Principles as geopolitical strategy or dismissed them as aspirational boilerplate. Neither reading engages with what the document actually is: an argument that the choice of first principle in a governance framework is a substantive philosophical claim, not a neutral ordering, and that the claim embedded in most Western frameworks—that individual rights are foundational—is one possible claim among several, not a universal default. The Beijing Principles did not resolve this argument. They made it explicit.

Key Ideas

Harmony as first principle. The Confucian concept of harmony (he, 和) is not the absence of conflict but the dynamic equilibrium of genuinely different elements that are mutually constitutive. As a governance principle, it demands attention to systemic relational effects, not merely individual outcomes. An AI system that is perfectly respectful of individual privacy rights while systematically degrading social trust has done something wrong that individual-rights frameworks cannot name. The Beijing Principles name it.

The principles gap. Between a governance principle and a governance outcome lies a translation problem that is both normative and empirical. The Principles were explicitly designed as a constitutional document rather than a regulatory code—a statement of values that should guide more specific rules, without being those rules themselves. This design choice maximized buy-in at the cost of tractability. The subsequent work of translating harmony into measurable technical requirements is the gap that Zeng has spent years attempting to bridge through platforms like AI Governance Online and the AI for SDGs Think Tank.

Inclusion and sharing as redistribution. The principle of inclusion and sharing (bao rong gong xiang, 包容共享) implies an active redistribution of AI's benefits that is more demanding than the Western concept of fairness. Fairness asks whether the same rules apply to everyone. Inclusion and sharing asks whether the benefits of a technology are actually reaching everyone. These are different requirements, and the second is structurally harder to satisfy and to measure.

Global commons framing. The Principles explicitly position AI governance as a global commons problem: the decisions made by AI-developing nations will shape development globally, and most of the world has little influence over those decisions. The document advocates for genuine multilateral participation rather than the export of any single nation's governance framework—including China's. This anti-monoculture position sits in productive tension with the document's own cultural specificity.