The United States is preparing for its artificial intelligence dialogue with China, and our officials are under tremendous pressure from many sincere people to cooperate on A.I. risk with our foremost strategic competitor. Because American and Chinese companies are building the most powerful models, these people argue that an agreement, even a treaty, between the superpowers is humanity’s last backstop against potential catastrophe.
美国正筹备与中国的人工智能对话，而许多怀着良好初衷的人正在向我国官员施加巨大的压力，要求在人工智能风险领域与我们首要的战略竞争对手展开合作。由于美中两国企业正在构建最强大的模型，这些人认为，超级大国之间达成协议甚至条约是人类防范潜在灾难的最后防线。
I get the logic. But as the former State Department envoy who co-led the first U.S.-China A.I. dialogue, I fear the coming talks will not save us. If the past is prologue, China will have little interest in anything beyond gaining a competitive advantage.
我理解这种逻辑。但是，作为曾共同主导首轮美中人工智能对话的前国务院特使，我担心即将举行的会谈无法拯救我们。如果历史可以作为参考，那么除了获取竞争优势，中国对其他任何事情都不会太感兴趣。
The agenda of the May 2024 meeting was crystal clear to both sides because our presidents had agreed to it: Focus on the risks of advanced A.I. systems and the potential for safety cooperation. Yet Chinese officials steered talks away from the potential risks of A.I. systems for hours. They even demurred from explaining their domestic A.I. regulations — despite having regulators in the room.
2024年5月那次会议的议程对双方而言都极为明确，因为两国元首已对此达成一致：重点关注高级人工智能系统的风险及安全合作的潜力。然而，在长达数小时的时间里，中方官员一直回避人工智能系统的潜在风险问题。他们甚至不愿解释其国内的人工智能法规——尽管现场就有监管官员。
Instead, China chided us for our concerns about A.I. risks. They portrayed our safety approach as a barrier to unlocking the technology’s opportunity and a way to stop poorer countries from accessing A.I. tools. They unfurled proposals for subdialogues that would have most likely resulted in endless procedural diplomacy.
中方反而因我们对人工智能风险的担忧而指责我们。他们将我们的安全策略描绘成阻碍释放技术机遇的屏障，以及阻止较贫穷国家获取人工智能工具的手段。他们还抛出了关于举行子对话的提议，这极有可能引发无休止的程序性外交。
A moment outside a hotel conference room sticks in my mind. A Chinese counterpart turned to me with a puzzled expression and asked why we actually wanted to talk about A.I. risk. We took the mandate from our presidents seriously. They saw only the opportunity to jockey for strategic advantage.
酒店会议室外的一个瞬间令我记忆犹新。一位中方对等官员面带困惑地转向我，问我们究竟为何想谈论人工智能风险。我们认真履行了两国元首的授权。他们只看到了争夺战略优势的机会。
In this month's meetings, I expect China to again downplay safety risks. It could dangle the possibility of safety cooperation in exchange for concessions such as the softening of chip controls. I think it will continue to portray itself as an altruistic savior dedicated to ensuring that the developing world gains A.I. access. Meanwhile, it will be likely to keep doing everything it can to get ahead — like generating propaganda in the hopes of inflaming opposition to data centers and encouraging Chinese companies to continue exploiting U.S. models to improve their systems.
在本月的会议上，我预计中国将再次淡化安全风险。它可能会以安全合作的可能性为筹码，换取诸如放宽芯片管制等让步。我认为，中国将继续把自己塑造成一个无私的救世主，致力于确保发展中国家获得人工智能的使用权。与此同时，它可能会继续想方设法争取领先——例如制造宣传，试图煽动对数据中心的反对情绪，并鼓励中国企业继续利用美国模型来改进自身系统。
Maybe China will toss its old playbook. Even President Xi is talking about risk and loss of control in A.I. Vivid new incidents show that A.I. systems can slip the leash of human control. Fear may converge with interest to produce coordinated action. Or at least a new understanding.
也许中国会摒弃旧的套路。甚至中国国家主席习近平也在谈论人工智能的风险和失控问题。一些新近发生的生动事件表明，人工智能系统可能会挣脱人类的控制。恐惧可能会与利益交汇，从而促成协调一致的行动，或者至少促成一种新的共识。
But if Beijing now takes the risks of A.I. seriously, the message has not reached its domestic companies. The Chinese government touts the global adoption of its open models, but Chinese models do not currently match U.S. models in safety performance that matters — thwarting access to the most dangerous capabilities. Its companies rarely disclose these risks and say almost nothing about signs that their models, too, might act recklessly beyond human control.
但是，如果说北京现在正认真对待人工智能的风险，那么这一信息并未传达到其国内企业。中国政府宣扬其开放性模型在全球被采用，但中国模型目前在关键的安全表现（即阻止获取最危险的能力）上无法与美国模型相提并论。中国企业很少披露这些风险，对其模型也可能超出人类控制而鲁莽行事的迹象几乎闭口不谈。
Complicating this already muddy situation for U.S. officials are the A.I. scientists, politicians, Wall Street and technology heavy hitters demanding an immediate A.I. alliance to thwart rogue A.I. I worry China will take advantage of these individuals’ sincerity, especially if diplomacy fails to produce a comprehensive safety agreement. Beijing could feign seriousness, float a hollow offer, then pin blame for its failure on the United States. The result could spark internal debates that weaken the core bipartisan consensus that the United States must lead in all facets of A.I., making it easier for China to forge ahead in the global technology competition.
对于美国官员来说，人工智能科学家、政界人士、华尔街和科技界重量级人物纷纷要求立即建立人工智能联盟，以防范失控的人工智能，这使原本就错综复杂的局面变得更加棘手。我担心中国会利用这些人的真诚，特别是如果外交努力未能促成一项全面的安全协议。北京可能会假装态度严肃，抛出一个空洞的提议，然后将其失败的责任推给美国。其结果可能会引发内部辩论，削弱“美国必须在人工智能各个领域保持领先”这一核心的两党共识，从而使中国更容易在全球技术竞争中占据上风。
What’s more, no durable diplomatic agreement can be reached while domestic policy is under construction. Companies must first start with technical and oversight solutions, which can be turned into policies by the government and only then translated into international agreement. Rather than press American officials for a diplomatic solution that cannot be realized, safety advocates with influential voices should push developers in both countries to raise standards, invest real money in research to build safer A.I. and increase transparency and reporting. And they should continue unofficial conversations with Chinese business leaders and former officials — a step that could encourage the government to act.
更重要的是，在国内政策仍在制定中的情况下，无法达成持久的外交协议。企业必须首先从技术和监管解决方案着手，这些方案可由政府转化为政策，随后才能转化为国际协议。具有影响力的安全倡导者与其向美国官员施压以寻求无法实现的外交解决方案，不如推动两国开发者提高标准，投入真金白银研发更安全的人工智能，并提升透明度与报告水平。此外，他们还应继续与中国商业领袖及前官员进行非正式对话——这一举措可能会促使政府采取行动。
That’s not to say these summits can’t accomplish anything. The United States can and should test China’s commitment to addressing A.I. risk and hold its developers to the same transparency and safety performance benchmarks that American developers follow. Future talks could verify that progress has been made.
这并不是说这些峰会毫无建树。美国能够且应当检验中国应对人工智能风险的承诺，并要求其开发者遵守与美国开发者相同的透明度及安全表现基准。未来的会谈可以验证是否已取得进展。
If both sides meet in earnest, the very best I could envision is a nonbinding commitment in which each side would agree to put pressure on its companies to publish and improve safety benchmarking for cybersecurity and biological risks, and to disclose (rather than bury) reports of concerning incidents. Both sides could also affirm their existing shared commitments on A.I. safety, including the 2024 agreement to maintain human control over nuclear weapons decisions. This will be a multiyear endeavor; countries should begin by affirming they are players in a longer arc of progress.
如果双方诚意会晤，我能设想的最佳情况是达成一项不具约束力的承诺，即双方同意向各自企业施压，要求其发布并完善针对网络安全和生物风险的安全基准，并披露（而非掩盖）引发担忧的事件报告。双方还可以重申现有的人工智能安全共同承诺，包括2024年达成的关于维持人类对核武器决策控制的协议。这将是一项历时多年的努力；各国应首先确认自身是这一长远进程的参与者。
I realize this sounds underwhelming for those who see catastrophe ahead. But considering the rigidity of great power diplomacy and the immaturity of domestic policy, this would be an achievement against which to measure whatever comes next.
我意识到，对于那些预见未来会发生灾难的人而言，这听起来或许有些平淡无奇。但是，考虑到大国外交的刻板性以及国内政策的不成熟，这将是一项成就，足以为衡量后续任何进展提供标尺。
I am confident U.S. officials understand the stakes, as well as the tactics of their counterparts. And I hope that their efforts to find a balance will be supported by our A.I. safety community. At home, the core work must continue to build a path that all countries can sign on to.
我深信美国官员了解其中的利害关系，也清楚对手的策略。我希望他们为寻求平衡所做的努力能得到我们的人工智能安全社区的支持。在国内，核心工作必须继续推进，以开辟一条所有国家都能认同的道路。