US, China, and EU: Competing Visions for AI Governance

The three leading global AI actors—the US, China, and the EU—have divergent, often conflicting, approaches to AI governance, driven by distinct agendas and motivations for global leadership. These differences are a primary barrier to establishing a unified global mechanism.

• US Approach: Characterized by a market-driven, less restrictive, and decentralized model, prioritizing competitive advantage and innovation. Governance is often sector-specific, with influence from industry actors and a reliance on existing mechanisms.
• China Approach: An extension of its authoritative political system, centrally-planned and rule-based, aiming for global dominance in AI by 2030, including setting norms and ethical standards. It leverages AI for local governance, economic development, and geopolitical power.
• EU Approach: Focuses on "trustworthy AI" with a risk-based regulatory framework (e.g., AI Act), prioritizing human-centered ethics, data privacy, and fundamental human rights. The EU seeks to exert a "Brussels Effect" globally, exporting its values through regulatory influence.

The Perils of Non-Cooperation: Barriers and Risks

The non-cooperation between the US, China, and the EU stems from geopolitical ambitions, ideological divides, and divergent preferences for international cooperation. Each seeks global leadership in AI, making consensus on common standards difficult. The EU's innovation-stifling reputation, coupled with US-China tech decoupling, further exacerbates tensions.

The risks of this non-cooperation are severe and multifaceted:

• Heightened Nationalism & Geopolitical Fragmentation: Prioritizing national interests over global concerns, leading to protectionist policies and worsening inequalities.
• Weakening Multilateralism: Undermining global cooperation, fostering 'minilateral' groupings, and enabling 'forum shopping' for least restrictive regimes.
• Inability to Address Cross-Border Risks: Hampering joint research and coordinated efforts to mitigate risks from open-source AI, nefarious applications, and frontier AI models (e.g., cyberattacks, existential risks).
• Race-to-the-Bottom Dynamic: Prioritizing speed over safety, leading to increased risks and cutting corners in AI development, potentially escalating an "AI arms race" with dangerous military applications and value alignment issues.

Collective Action Framework for Global AI Governance

Framing global AI governance as a common good—non-excludable but rivalrous—allows for the application of collective action insights. Ostrom's design principles for Common-Pool Resources (CPRs) and Jagers et al.'s large-scale collective action framework offer critical guidance.

Key considerations for designing effective global AI governance institutions include:

• Polycentric Multilevel Arrangement: More effective than a single centralized mechanism, allowing for flexibility, local responsiveness, and learning.
• Clear Boundaries & Incentives: Defining participation rights and incentivizing key players (US, China, EU) to cooperate by focusing on shared challenges.
• Congruence & Adaptability: Rules must consider local contexts, cultural variations, and allow for collective modification to cope with rapid AI developments.
• Robust Enforcement: Implementing mutually agreed-upon monitoring, graduated sanctioning, and conflict-resolution mechanisms, supported by international organizations for information provision and trust-building.