面對全球人工智能挑戰，中美合作尤為重要。                 —— 美國前國務卿亨利·基辛格

人工智能（AI）已成為中美兩國重要的競爭領域，這使得原本錯綜複雜的雙邊關係更加複雜。隨着特朗普重返白宮，理論派和實踐派對中美人工智能治理的未來持有不同看法。

一種樂觀的看法認為，儘管特朗普團隊中不乏對華強硬派和“鷹派”，但他們的精力主要集中在經濟和軍事領域。在人工智能安全問題上，由於特朗普政府更加關注其產業發展，傾向於將安全風險問題簡單化，因此相信“中國可以與特朗普進行坦誠對話”。此外，美國聯邦政府與中國在人工智能治理方面也保持着合作。例如，2023年11月，習近平主席與拜登總統在舊金山會晤時，雙方同意建立中美人工智能政府間對話機制。2024年5月14日，首次對話在瑞士日內瓦成功舉行，向外界傳遞出中美加強全球人工智能治理合作的積極信號。

一種較為悲觀的觀點則認為，中美圍繞人工智能的談判只是做樣子，並未觸及治理的實質。兩國合作仍處在十字路口。隨着特朗普重返白宮，共和黨將延續以往做法，在芯片、半導體等尖端技術領域對中國進行打壓。中國也將採取反制措施，從而使兩國在人工智能治理方面合作的可能性變得微乎其微。特朗普上任後還宣布將任命一位人工智能官員，主要負責人工智能政策、國家安全和產業發展，強調在人工智能領域與中國進行全面競爭。

事實上，這兩種觀點可能要麼過於悲觀，要麼過於樂觀。一方面，抑制中國在人工智能領域的崛起已成為美國兩黨共識，但中國也在不斷積累實踐經驗和戰略準備。拜登政府大力遏制中國人工智能產業，如斥資530億美元推出《芯片法案》，用於扶持美國半導體產業。相比之下，特朗普政府開始專註於對外國人工智能芯片製造商徵收高額關稅，而非提供國內補貼，以此振興美國半導體產業。這種做法可能促使台積電、三星等公司轉向規模更大、更開放的中國市場。此外，特朗普指責台灣“竊取”美國半導體業務，並要求其支付更多安全費用的立場，可能導致美台在半導體領域關係緊張。鑒於台灣在先進人工智能芯片生產中的關鍵作用，這反而有利於中國。另一方面，對中美人工智能治理的樂觀情緒也不合理。兩國都認識到，控制人工智能技術意味着重塑全球秩序，失去人工智能發展的戰略窗口意味着喪失競爭優勢。因此，人工智能領域的緊張局勢仍將是大範圍且激烈的。

隨着特朗普重返白宮，兩國爭奪新興人工智能技術領域主導權的趨勢愈演愈烈，並將持續下去。中美人工智能治理將面臨新的問題和挑戰。結合“囚徒困境”理論，中美人工智能治理困境主要表現為二階困境（見圖1）。

圖1：中美人工智能治理的“囚徒困境”

首先的困境是，美國的人工智能政策已從“踩剎車”轉向“猛踩油門”，優先考慮國家利益，這可能損害中國等國的利益。這就是博弈論中的一階困境：個體追求自身利益最大化的決策可能損害集體利益。拜登政府強調人工智能治理和監管，特朗普則側重於放鬆管制和快速發展人工智能。特朗普已經撤銷了拜登2023年關於人工智能的行政命令，該命令旨在解決隱私和安全等風險，而特朗普認為過度監管扼殺創新，削弱美國的競爭力。

相比之下，中國的人工智能政策始終重視發展而非治理，是採取“先發展，後治理”的策略。拜登執政時期的“異步異調”競爭轉為特朗普執政時期的“同步同調”競爭，意味着兩國可能在自動駕駛和物聯網等人工智能領域展開激烈的直接競爭。因此，兩國可能淡化人工智能治理，採取放任自流的策略。這種策略雖然有利於創新，但可能忽視關鍵的監管問題，這將加劇人工智能領域的競爭，減少治理方面的投入。

當中美之間的競爭和互不信任阻礙了人工智能的合作治理，就會出現二階困境。按照囚徒困境理論，這種情況下，個體在尋求自身利益最大化的同時，也必須考慮他人的反應和策略，而這種動態會造成一種循環困境：即使雙方都認識到合作有利於集體，但由於信任和互動不足，有效合作受到阻礙，最終損害全球利益。人工智能的風險無處不在，影響着所有國家，因此需要全球合作才能實現有效治理。然而，如果中美兩國繼續陷入互不信任引發的治理困境，就會破壞全球人工智能體系的穩定。儘管互聯網和智能技術促進了開放包容，“陣營對抗”和冷戰思維仍有可能蔓延到人工智能領域。正如一些學者所言，這可能導致“平行世界”的出現，即美國和中國各自領導一套互不相干的人工智能系統。這種分裂將阻礙科學進步和經濟增長，最終導致雙輸，正如諺語所說——“持槍對決沒有贏家”。

為了解決中美人工智能治理的二階困境，兩國必須建立一種交易性“戰略穩定對話”機制。首先，雙方需要明確人工智能治理的共同利益，例如經濟振興、政績和長期戰略目標。網絡安全對話等過往的失敗經驗凸顯了相互理解、避免以自我為中心的談判的重要性。其次，相互尊重和公平貿易的契約精神至關重要。儘管競爭激烈，但兩國優勢互補：中國在人工智能應用領域表現出色，美國在研發領域處於領先地位。這種共生關係可以促進合作而非零和競爭。第三，雙方應保持“錯位競爭”而非共振競爭，在確保談判透明度的同時，充分發揮各自優勢。例如，中國在人工智能專利申請方面領先，美國在人工智能模型方面佔據主導地位。最後，即使在雙邊關係緊張的情況下，對話機制也必須保持開放。冷戰時期美蘇核對話的經驗告訴我們，持續溝通可以增進理解並最終達成協議，哪怕雙方沒有立即建立起信任。漸進的小措施，也可以為人工智能治理過程中的大規模合作鋪平道路。

In the face of global challenges in AI, cooperation between China and the United States is particularly important.                                                                     —— Henry Kissinger, Former Secretary of State, USA

Artificial intelligence (AI) has become a major arena of competition between China and the United States, further complicating the already complex bilateral relationship. With Trump's return to the White House, theorists and practitioners hold divergent perspectives on the future of China-U.S. AI governance.

One optimistic view suggests that although Trump's team includes some hardliners and "hawks" on China, their focus primarily lies in the economic and military sectors. On AI security issues, it's believed that "China can have an honest dialogue with Trump" because his administration is more concerned with industrial development of AI and tends to simplify security risk issues. Also, the current U.S. federal government maintains cooperative AI governance projects with China. For instance, in November 2023, during the meeting held between President Xi Jinping and President Biden in San Francisco, both sides agreed to establish a China-U.S. AI government dialogue. The first dialogue successfully took place in Geneva, Switzerland, on May 14, 2024, sending a positive signal to the outside world about China-U.S. cooperation to strengthen global AI governance.

The other, more pessimistic view argues that China-U.S. negotiations surrounding AI are merely for performance and don't touch on the substance of AI governance. The cooperation between the two countries remains at a crossroads. With Trump's return to the White House, the Republicans will continue their previous practices, suppressing China in areas of advanced technology like chips and semiconductors. China will also adopt retaliatory measures in response, making China-U.S. cooperation on AI governance potentially minimal. After taking office, Trump also announced that he would appoint an AI czar, mainly responsible for AI policy, national security, and industrial development, emphasizing comprehensive competition with China in the AI field.

In reality, both perspectives may be overly pessimistic or optimistic. On one hand, suppressing China's rise in AI has become a bipartisan consensus in the U.S., but China has accumulated practical experience and strategic preparedness. The Biden administration aggressively contained China's AI industry, exemplified by the $53 billion "Chip Act" to bolster the U.S. semiconductor sector. In contrast, the Trump administration has begun focusing on revitalizing the U.S. semiconductor industry through high tariffs on foreign AI chip manufacturers rather than domestic subsidies. This approach could drive companies like TSMC and Samsung to shift toward China's larger, more open market. Additionally, Trump's stance on Taiwan, accusing it of "stealing" U.S. semiconductor business and demanding more security payments, may strain U.S.-Taiwan semiconductor relations, benefiting China given Taiwan's critical role in advanced AI chip production. On the other hand, optimism about China-U.S. AI governance is unwarranted. Both sides recognize that controlling AI technology means reshaping the global order, and losing the strategic window in AI development would mean losing competitive dominance. Thus, tensions in the AI field will remain comprehensive and intense.

With Trump's return to the White House, the trend of the two countries vying for dominance in emerging AI technologies has intensified and will continue to do so, and there will be some new issues and challenges in China-U.S. AI governance. In combination with the "Prisoner's Dilemma” theory, the dilemma of China-U.S. AI governance mainly manifests as a second-order dilemma (see Figure 1).

Figure 1: The "Prisoner's Dilemma" of China-U.S. AI Governance

The first-order dilemma is that the U.S. AI policy has shifted from "hitting the brakes" to "stepping on the gas," prioritizing national interests at the potential expense of China and others. This reflects a first-order dilemma in game theory, where individual decisions to maximize self-interest can harm collective interests. While the Biden administration emphasized AI governance and regulation, Trump's approach focuses on deregulation and rapid AI development. Trump has already rescinded Biden's 2023 executive order on AI, which aimed to address risks like privacy and security, arguing that overregulation stifles innovation and disadvantages U.S. competitiveness.

In contrast, China's AI policy has consistently prioritized development over governance, adopting a "development first, governance later" approach. This shift from "asynchronous and heterotonic" competition under Biden to "synchronous and homotonic" competition under Trump means both nations may engage in direct, intense competition in AI sectors like autonomous driving and IoT. Consequently, both countries may downplay AI governance, adopting a laissez-faire strategy that fuels innovation but risks neglecting critical regulatory concerns, leading to heightened AI rivalry and reduced investment in governance.

The second-order dilemma arises as competition and mutual distrust between China and the United States hinder cooperative AI governance. Drawing from the prisoner's dilemma theory, this scenario highlights how individuals, while seeking to maximize their own interests, must also account for the reactions and strategies of others. This dynamic can create a cyclical dilemma: even if both parties recognize that cooperation would benefit the collective, the lack of trust and interaction prevents effective collaboration, ultimately harming global interests. AI risks are pervasive and affect all nations, necessitating global cooperation for effective governance. However, if China and the U.S. remain trapped in a governance dilemma fueled by distrust, it could destabilize the global AI system. Despite the internet and intelligent technologies promoting openness and inclusiveness, "camp confrontation" and Cold War mentalities risk spilling into the AI domain. This could result in the emergence of "parallel worlds," as some scholars suggest, where the U.S. and China lead separate, isolated AI systems. Such fragmentation would hinder scientific progress and economic growth, ultimately leading to a lose-lose outcome, akin to the adage, "There are no winners in a gunfight."

To address the second-order dilemma in China-U.S. AI governance, both nations must establish a transactional "strategic stability dialogue" mechanism. First, they need to identify shared interests in AI governance, such as economic revitalization, political achievements, and long-term strategic goals. Past failures, like the cybersecurity dialogue, highlight the importance of mutual understanding and avoiding self-centered negotiations. Second, a contract spirit of mutual respect and fair trade is essential. Both countries, despite fierce competition, share complementary strengths—China excels in AI applications, while the U.S. leads in research and development. This symbiotic relationship can foster cooperation rather than zero-sum competition. Third, they should maintain "dislocated competition" rather than resonating competition, leveraging their respective advantages while ensuring transparency in negotiations. For instance, China leads in AI patent applications, while the U.S. dominates in AI models. Finally, dialogue mechanisms must remain open, even amid bilateral tensions. Drawing lessons from Cold War-era U.S.-Soviet nuclear dialogues, sustained communication can build understanding and eventually lead to agreements, even without immediate trust. Small, incremental steps can pave the way for larger collaborations in AI governance.