“America is not prepared to defend or compete in the AI era.” The introduction to the final report of the National Security Commission on Artificial Intelligence released on March 2 does not mince words. The United States, in its telling, needs to start taking artificial intelligence seriously at a national level, lest it be overtaken by China and exposed to AI-driven threats from malicious actors of all types.
The report itself covers a wide range of topics related to AI, but its focus (unsurprisingly, given that the commission was created through the 2019 National Defense Authorization Act) is principally on national security implications. There is one particular aspect of the report which is worth a closer examination: its call, in chapter 3, for the U.S. to bring its global military allies and partners together around a common set of AI standards for interoperability and common operational doctrine for the development and use of AI-driven emerging technologies.
The report recommends that the U.S. draw its disparate allies and partners together to retain its AI technological edge over China and Russia. Specifically, it calls for creating an “Atlantic-Pacific Security Technology Partnership,” drawing together the Quad, NATO, the Five Eyes nations and the members of the AI Partnership for Defense, along with other unspecified allies and partners.
But — speaking as a lapsed think tanker — making sweeping policy recommendations is easy. Operationalizing them is less so.
On its face, it makes sense to pursue standardization among military allies. The U.S. has been doing so since the Second World War, when it became the “Arsenal of Democracy” and shipped everything from rations to fighter aircraft to allies across the world fighting very different types of wars. That experience informed American post-war alliances and spoke to the crucial importance of simplifying production and supply chains, not to mention creating uniform performance and training standards to make military personnel capable of fighting together efficiently and effectively. As a result, U.S.-led standards are prevalent across the militaries of the developed world. As one example, NATO (largely at the urging of the U.S.) standardized rifle and machine-gun ammunition around the 5.56 mm and 7.62 mm rounds. Those remain the standard calibers for not only NATO member states but countries around the world aligned with the U.S. It is, after all, much cheaper and easier to buy into an existing standard than to create a new one.
But software is not a rifle bullet. Artificially intelligent software, in particular, is inherently adaptive and flexible. And American allies have hugely divergent laws and public views of surveillance and the legitimate role of technology in daily life. To even agree on a relatively basic set of standards between nations as differently oriented as Germany, Japan, India, Israel and Singapore — just to name a few — will be an enormously complex and politically fraught endeavor.
That political element should not be underrated as an obstacle to the proposed partnership. The report identifies China as the prime competitor and potential threat; there is little mention of potential cooperation on AI with the PRC. Russia, too, is listed as an adversary, primarily around the idea of AI-enabled disinformation campaigns. But many of the countries listed as necessary partners in the Partnership do not share the ascendant American geostrategic outlook. Will India, with its fairly close political and military ties with Russia, want to be part of an effort directed in part at limiting Russian influence? Will Israel, which maintains strong political ties with Russia and a degree of industrial and commercial exchange with China? Will EU nations, whose approach to China has caused significant consternation in Washington at times, want to sign on to an effort that is cast in explicitly national security terms — especially given Washington’s recent and questionable use of national security justifications in its approach to trade policy?
Furthermore, there is the question of whether the putative members of this body will be interested in working with each other. The American strategy in the Cold War was to build a hub and spokes model of alliances: the U.S. would support European countries through NATO, and separately support a set of allies in the Asia-Pacific region, all in the name of containing Communism. The success of the strategy did not rely on bringing Asia and Europe together. But the U.S. position now is far less dominant than it was in the late 1940s, when the U.S. was the only feasible non-Soviet backer for European and Asian nations still struggling to rebuild from the shattering damage of the Second World War. Absent some clearer and more present danger, the U.S. needs to offer more meaningful incentives to rebuild and reform its alliance structures around a technological challenge — and it is not clear from this report what those might be.
None of this is to suggest that the potential threats posed by artificial intelligence aren’t real or present or growing. But to suggest that the U.S. circa 2021 can simply rally a sizeable fraction of the world around its agenda betrays a certain naivety about the geopolitical order into which AI systems are going to be deployed.