Artificial intelligence’s destructive potential has resulted in a flurry of recent governance activity. Mere days before the U.K. hosted its AI Safety Summit from Nov. 1-2, the Biden administration announced the executive order on “Safe, Secure, and Trustworthy Artificial Intelligence.” Though the summit in the U.K. set out to focus on catastrophic risks of AI, the U.S. efforts have focused on more concrete issues such as its military uses.
While at the summit, U.S. Vice President Kamala Harris announced several new initiatives that comprised the executive order, such as the new AI Safety Institute. But crucially, Harris also announced that 31 nations had joined the Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy. The declaration was first announced at the Summit on Responsible Artificial Intelligence in the Military Domain held in February in the Netherlands.
The update on the declaration signals U.S. commitment to this effort and brings attention to the signatory list that includes U.S. allies such as Canada, Australia and France. Notably missing, though, are Russia and China. The likelihood of either state joining a U.S.-led effort in the current geopolitical climate is exceedingly low.
Russia was not welcome at either summit and is unlikely — with its ongoing invasion of Ukraine — to be brought into these discussions. Even if it was invited, Russia would not likely sign even voluntary documents, as it does not wish to see any regulation, binding or nonbinding, on emerging technologies.
China did attend the summits and sign onto two non-legally binding instruments: the “REAIM 2023 Call to Action”; and the “Bletchley Declaration” agreed to at the U.K. summit. While important for further dialogue, this obscures a greater obstacle that China poses on AI regulation generally — and specifically on military applications of AI.
While China is unlikely to be as obstructionist as Russia has been in multilateral discussions on autonomous weapons, it is clear it will only agree to nonbinding instruments and those that are on its terms. This means that it is unlikely to join the U.S. Political Declaration on Responsible Military Use of AI and Autonomy due to its strategic interests in relation to the technology and the broader geopolitical competition with the U.S.
This was evident when it came to the recent vote on the first-ever resolution on autonomous weapons at the First Committee of the General Assembly, which generally notes that states recognize the urgency to address growing autonomy in weapon systems and to hold more talks. While 164 voted to approve the resolution, China abstained.
China’s abstention highlights that it will try to shape any outcome, including delaying efforts, until the terms are favorable to its ambitions to achieve military AI supremacy.
Only two states voted against the resolution: Russia and India.
Neither vote is surprising given both Russia and India have pushed back against more significant regulatory steps at the U.N. Convention on Certain Conventional Weapons, or CCW. Indeed, that forum has largely stalled due to the treatment of consensus as unanimity as well as resistance by Russia and India.
Is the pushback by China, Russia and India insurmountable?
Over the years of discussions on autonomous weapons at the CCW, it has become evident that allies talking to allies does not address the challenge of more adversarial states or states that would be adversaries, primarily for the U.S. and its allies.
China joining some of these discussions should be welcomed. However, there should be no illusion that the presence of China or its signing of nonbinding measures is indicative of its willingness to commit to hard laws.
Now, this may not appear to be an issue, as neither the U.S. nor its allies are too interested in hard laws on military AI. Even the expected “landmark agreement” — reached on the margins of the Asia-Pacific Economic Cooperation summit between the U.S. and China, apparently banning the use of AI in weapons, drones and nuclear command and control — is more than anything going to feature voluntary and aspirational measures.
However, voluntary agreements and exchange of information are much easier to do with allies. When crisis scenarios among more adversarial states arise — and they are likely to as more states deploy AI and more autonomous systems in battlespaces — it will be important to have clarity on what is permissible, communication channels open, and clear rules guiding uses of AI and autonomy. It is likely sooner rather than later that states will realize the benefit of some legally binding instruments as well.
The political declaration and the first-ever U.N. resolution on autonomous weapons are important steps forward, as is the expected bilateral agreement between the U.S. and China. But more governance, including hard laws and complementary processes on military AI and autonomous weapons, is needed. This will require a degree of skilled diplomacy to engage not just allies but potential adversaries, and to craft legal agreements. Only then will the risks that come with military AI, such as errors and conflict escalation, be truly addressed.
Branka Marijan is a senior researcher at Project Ploughshares, specializing in the military and security implications of emerging technologies. She is also a contributor to the Centre for International Governance Innovation think tank.