Should the United States slow down artificial intelligence development to address possible existential risks to humanity? The Trump administration prefers a “let it rip” approach to winning the AI race. But a growing chorus of American tech titans want to slow down AI development so that risk prevention measures can catch up with the technology. Anthropic’s Dario Amodei, for example, argued that “we must slow the pace at which we improve the capabilities of AI models.” He calls for “building AI at a balanced rate that aims to ensure its safety while still achieving its benefits and grappling with important geopolitical dilemmas.” OpenAI’s Sam Altman posted on X, “I agree with Dario that we need to pace the frontier.” Should the United States slow down artificial intelligence development to address possible existential risks to humanity? The Trump administration prefers a “let it rip” approach to winning the AI race. But a growing chorus of American tech titans want to slow down AI development so that risk prevention measures can catch up with the technology. Anthropic’s Dario Amodei, for example, argued that “we must slow the pace at which we improve the capabilities of AI models.” He calls for “building AI at a balanced rate that aims to ensure its safety while still achieving its benefits and grappling with important geopolitical dilemmas.” OpenAI’s Sam Altman posted on X, “I agree with Dario that we need to pace the frontier.” However, the case for greater controls contains a major weakness: It rests on coordination with China. Amodei’s three-point plan for pacing AI, for example, will require “cooperation with China, the autocratic country with by far the most advanced AI capabilities.” Altman agreed. He said, “Even if just the U.S. and China could agree on some shared standards and testing for development of this technology, I think that’d be a wonderful accomplishment that the two of them can deliver.” Jacob Coxon, who made international news when he resigned from Anthropic in September, said, “A race with China could be very bad and that we should be taking pretty rapid actions for international coordination.” But coordinating with China is much easier said than done. Chinese President Xi Jinping is not interested in negotiating controls on AI development, and even if he were, he could not be trusted to abide by them. Believing that we can work with Beijing is a tech-world fantasy rooted in a shallow understanding of geopolitics. To be fair, Amodei wrote, “We must not be naïve here: the geopolitical stakes are so high that there will likely be stark limits on what can be achieved.” But he and other technologists do not seem to fully grasp the depth of the challenge. Washington cannot afford, therefore, to unilaterally restrain itself as Beijing races ahead. Just like with Cold War-era nuclear arms control agreements, any AI restraints must be mutually binding and verifiable. But Beijing is not interested in negotiating binding restraints. Instead, Xi has criticized Western attempts to control AI. China’s AI diplomacy instead trumpets the technology’s potential for economic development, especially for the global south. In response to Amodei’s proposals for coordinated controls, Chinese Foreign Ministry spokesperson Guo Jiakun stated that “fearmongering, confrontation, and vicious competition will only disrupt the process of global AI governance.” Moreover, China and other dictatorships might even have rational incentives to embrace—not control—AI applications that court catastrophic risk. They trust algorithms more than their own people. Humans do not always reliably carry out orders, but computers do. I’ve talked to experts who say that China’s military officers, for example, are worryingly optimistic about the benefits and controllability of AI-enabled weapons systems. Further, even if Washington and Beijing could agree on mutual constraints, Beijing is unlikely to abide by them. Dictatorships in general—and China, in particular—tend to cheat on international agreements. During the Cold War, the Soviet Union regularly cheated on nuclear arms control agreements while dictatorships like Iran, Iraq, Libya, Syria, and North Korea have systematically violated their commitments under the Nuclear Nonproliferation Treaty. Turning to the case of China, Beijing joined the World Trade Organization only to systematically violate its principles by stealing intellectual property from the West. It used its position in the World Health Organization to obstruct an effective investigation into the origins of COVID-19. China makes a mockery of the United Nations Human Rights Council as it engages in crimes against humanity in Xinjiang. Xi promised former U.S. President Barack Obama in the Rose Garden that he would not militarize islands in the South China Sea, and then he turned around and did just that. The list goes on. China’s compliance with a new AI regime is unlikely to be any better. How would new AI limitations be effectively verified and monitored? A new international agreement to control AI would likely mean that Western democracies restrain themselves as Communist China cheats covertly and races ahead. U.S. and allied diplomats can certainly work among themselves to conceive reasonable standards and to pursue talks with their Chinese counterparts. The United Nations can set up a new agency for AI coordination. But everyone should be realistic that these efforts will likely be empty talk shops unable to make real progress. In the absence of an effective pathway for cooperation with Beijing, Washington’s best path is to push forward and win the AI race. To be sure, AI presents possible existential threats, from the AI-enabled development of advanced bio and cyber weapons to lethal autonomous weapons systems and even automated nuclear command and control. In principle, it would be desirable to have binding international guardrails to protect against such catastrophic risks. But there are competing demands. Washington must also win the AI race. The United States has enjoyed technological leadership for so long—at least since the time of Thomas Edison—that it almost takes it for granted. The United States’ technological edge has provided enormous economic, military, soft power, and geopolitical advantages. Now, China wants to claim these advantages for itself. Losing the AI race to China would be a disaster. AI superiority would allow China to dominate the battlefields and boardrooms of the future. As Trump has rightly argued, “If we don’t win AI, we’re going to be put in a very bad position.” Losing the AI race would also have enormous consequences for Western values. What would it mean for democracy and human rights, for example, if the world is dependent on a dictatorship for 21st-century technology? China could use the technology for espionage, sabotage, supply chain coercion, and censorship. Could we even use Chinese-built AI tools to search for Tiananmen Square in the future? What else would be off limits? With the stakes this high, the United States cannot afford to compete with one hand tied behind its back. By securing AI leadership, Washington will be in a much stronger position to force China to play its game and negotiate future AI guardrails from a position of strength. As Trump said, “We’re leading now over China by a lot, and everyone else, and we’re going to keep it that way. I’m not going to stifle growth of something that will be bigger than the industrial revolution.” It is unfortunate that Washington does not have a more reliable negotiating partner in Beijing. But given that reality, the least bad feasible path forward for now is to let it rip.
Cooperating With China on AI Control Is a Fantasy
Full Article
Original Source
Read the full article at Foreignpolicy →KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.