The Paris AI Action Summit was another expensive, flashy failure of multilateral global governance. It signaled the end of a short-lived attempt by liberal democracies to, as the Action Summit website put it, “shape an effective and inclusive framework of international governance for AI.” The stimulus for these efforts came from the AI panic of 2023, in which machine learning applications were suddenly perceived as a “new” and “dangerous” technology that required global collective action.
Yet America Firster JD Vance was sent to the summit to announce the U.S. pull back from the collective effort. Instead of effective and inclusive international governance, he said: “The United States of America is the leader in AI, and our administration plans to keep it that way.” So AI is now a symbol of our national prowess. The US is singing the machine intelligence version of “drill, baby drill.”
Whether or not the US is leading AI tech development, its new turn is creating a trend toward defections from the collective effort to regulate AI safety. In the new AI unilateralist model, it’s every neo-mercantilist state for itself. It’s being framed as a “race” in which the winner takes all (even though it’s not). The Paris Summit made the breakup of the wannabe global governors of AI public and explicit. Here’s a good report about it.
Even before Trump’s election, there were signs that nations were moving from an “AI is an existential threat to humanity” policy to an “AI is a strategic technology that states must race to control for their own benefit.” The latter approach implies an industrial policy. States must have their own chip industries, data brokers, software developers, model designer-trainers. If these things come from other countries they are dangerous. More: states must lead the world in AI, for if they fall behind they will become the prey of the nations who master it. Or so we are told.
Leadership or Decoupling?
These political wind shifts create some weird cross-currents, however. Who is the true MAGA: Vance in Paris saying “build [AI], baby, build” or Senator Josh Hawley of Missouri, who’s saying, Regulate, Regulate, Regulate? Hawley has introduced the “Decoupling America’s AI Capabilities from China Act.” It would ban software downloads, restrict research and investment on AI, and reduce competition. MAGA Republicans it seems, also want to regulate because…China.
State of the Net (SOTN)
Meanwhile, back in Washington, the Internet Education Foundation held its State of the Net conference on the same day. This was a smaller-scale, U.S.-based conference that facilitates the exchange of ideas and acquaintances in a policy community centered on Washington. The same shift in attitudes that showed up in Paris was evident there, too. The AI doomers and the federal regulatory agencies following in their wake had lost their voice. A younger crowd of staffers and policy wonks were realizing that machine learning applications can do interesting things and do not threaten us all with immediate extinction. People were charmed, really, by the entry of the upstart DeepSeek onto the scene, although the usual crew of national security wonks are trying, somewhat lamely, to portray it as another national security threat. The new administration seems to be more interested in promoting technological leadership than in responding to fears of its risks with ex ante regulation. Congresspeople and agency heads seemed more relaxed about it
I was on a panel at SOTN about the geopolitics of the Internet. This panel reflected two positions about cybersecurity, China, globalization, and trade. There were three fearful protectionists (Hogg, Schwartz and Lewis), and one confident liberal (me). My presence on the panel was itself a good sign. A pro-trade, pro-peace, pro-market, pro-technology position is usually not represented at all in DC policy making circles. The “bipartisan consensus” about China has not allowed this position into the discourse. The Trump disruptions, however, with its potentially transactional approach to China, does. The panel became a way to challenge the prevailing wisdom in Washington about tech trade and China.
Threat claims with no threat model
On the panel, I criticized what I called the “missing threat model.” The operating assumption in Washington now is that any form of digitized or internet-enabled connection with China is a national security threat. These claims are made without demonstrating what vulnerability is going to be exploited, at what scale, for how long, to what effect. No need to demonstrate how an attack on a single home WiFi router might progress to the level of a threat to our collective security. You just need to say “it comes from China.” I also noted that none of the most damaging Chinese attacks (OPM, Salt Typhoon) leveraged Chinese apps or equipment. These calls for banning apps, for domestic production of chips and clouds, for walling off and localizing data, for cutting off cable and telecom interconnections, for censoring media, all have a common source: they come from exaggerating or overestimating the national security vulnerabilities created by trade in digital goods and services.
When are we going to realize how destructive the missing threat model is? All of the ICT we use, all of its rapid progress for the past 30-40 years, is a product of the globalization of production and consumption of digital goods. The internet itself was one of the primary drivers of globalization. To be against digital trade with Chinese people and firms is to be against the Internet.
Digital trade and security
The Russian invasion of Ukraine showed that you can’t take over a country using cyber means exclusively. For 8-10 years prior to the 2022 invasion, Russia struck at elements of Ukraine’s civilian infrastructure almost at will. But it could not do anything politically useful with that. It had to send tanks, missiles and soldiers. From Ukraine we have also learned that a globally integrated infrastructure increases resilience. Using a transnational cloud company, Microsoft, Ukraine’s government transferred critical data out of its own territory in case of a major compromise or total loss of its territory. Globalized LEOS services provided a lifeline of internet access. Finally, by watching Ukraine buy and deploy Chinese drones to be used against Russia we learned that even military capabilities can be enhanced by trade with an imputed adversary. We also learned that American-made drones built not in a competitive market, but commissioned by the Defense Department, while fully compliant with DoD regulations and software reviews, do not perform well on the battlefield. Maybe a bit more military-civil fusion is in order?
Hawley’s bill should keep us alert to the downside of the MAGA reset in digital policy, but for now, some space is being created for a more open debate about the relationship between national security and digital trade.
The post So Much for Global Governance of AI… appeared first on Internet Governance Project.
Source: Internet Governance Forum