Asia-Pacific Journal Series – The De Jure Necessity and De Facto Improbability of Establishing a New Military AI Treaty

Editors’ note: This post appears as part of a series highlighting articles in a recent volume of the Asia-Pacific Journal of International Law, on which presentations were made at a workshop hosted by National University of Singapore’s Centre for International Law. Introductory post by Samuel White, Danielle Yeow, and Jonathan Kwik can be found here.

The deployment of artificial intelligence (AI) in the military domain is becoming reality. The regulation of military AI is no longer a potential consideration but rather a pertinent contemporary issue. A foremost concern is how military AI should be regulated.

Some believe it is unnecessary to establish a new treaty for military AI; others disagree. The same division also exists among States, affecting the next steps in international cooperation. This post builds upon our previously published article and aims to investigate the necessity and probability of establishing a new treaty for military AI and then evaluate the value of the treaty-making process.

The De Jure Necessity of Establishing a New Treaty

Establishing a new treaty for military AI is de jure necessary. Although the International Court of Justice’s Nuclear Weapons advisory opinion could support the applicability of existing international law to military AI, the real problem is the insufficiency of existing law. Specifically, the current international legal framework struggles to adequately address problems such as moral dilemmas and accountability gaps.

Regarding the moral dilemma, the International Committee of the Red Cross contends that it is morally unacceptable for machines to determine who lives or dies. While this is morally debatable, there are currently no positive rules explicitly prohibiting or restricting the use of military AI solely on moral grounds. Concerning the accountability gap, this issue remains controversial in the literature and complicates the practical allocation of responsibility.

Interpreting existing international law might resolve the questions raised above, but fragmented interpretations could undermine legal certainty, which plays a significant role in international legal systems. Without a new treaty, States lack the legal certainty requisite as a precondition for legal compliance. On this basis, establishing a new treaty for military AI is de jure necessary.

The De Facto Improbability of Establishing a New Treaty

Although establishing a new treaty is de jure necessary, it is almost de facto improbable. Among treaty-making platforms, the Convention on Certain Conventional Weapons (CCW) mechanism is the most influential. However, concluding a new instrument under the CCW requires consensus. Because several States have objected, establishing a widely participated-in treaty on military AI under the CCW mechanism is highly improbable.

Human Rights Watch has proposed pursuing a new treaty for military AI through an alternative mechanism beyond the CCW. However, major military powers that oppose a new treaty via the CCW might also refuse to participate in a treaty negotiated under an alternative mechanism. This would substantially affect the treaty’s actual effectiveness, rendering a treaty that successfully regulates military AI de facto improbable.

Several factors could explain this improbability. First, unilateral regulation could directly affect the distribution of military power and military advantage, thereby undermining effective regulation. Second, for some States, interpreting existing ambiguous international law according to their preferences would better align with their interests, as it could allow them to avoid actual regulation. Furthermore, the intensification of geopolitical rivalry has shaped an environment that incentivizes the deployment of AI weaponry rather than its restriction.

Under the influence of the aforementioned factors, the dynamic in military AI regulation is one of “majority follows minority.” While about 130 States support a binding treaty, the opposition of a small minority of States renders it improbable that such a treaty will conclude, or renders any concluded treaty incapable of having substantive legal force.

The Treaty Process

Despite the improbability of establishing a treaty to effectively regulate military AI, the pursuit of such an instrument is not futile. The value of the treaty-making process remains even if the treaty itself does not manifest.

First, the treaty-making process could serve as an incubator for the formation of customary international law (CIL). The treaty-making process would facilitate supporting States articulating their legal positions on military AI, thereby generating State practice with opinio juris. Simultaneously, this process could pressure minority opposing States to articulate their legal positions, reducing the ambiguity that they intentionally maintain and facilitating the generation and crystallization of CIL.

Second, the newly established treaty text could serve as substantive evidence influencing the interpretation of existing international law. Practice regarding the new treaty could constitute supplementary means of treaty interpretation under Article 32 of the Vienna Convention on the Law of Treaties. To clarify ambiguous or obscure provisions in the Geneva Conventions, the interpreters could refer to the new proposed treaty on military AI.

Third, the international community could also impose constraints through reputational costs on States not party to the treaty drafting process. In theory, if the State pays more reputational cost when it is against a treaty than it gains, then the State is more likely to act consistently with the treaty. Notably, such reputational costs could arise even while the treaty remains in draft form. This effect reinforces how the negotiation process itself pressures objecting States to articulate their positions and thereby contributes to CIL formation. That said, reputational costs do not directly ensure that States fully control the use of military AI. Instead, they indirectly prompt States to make choices aligned with international expectations by increasing or decreasing these costs.

Concluding Thoughts

Existing law is insufficient to regulate military AI. However, a minority of objecting States will make it difficult to negotiate a required consensus to create a military AI treaty under the CCW. That doesn’t mean the treaty-making process itself is not worthwhile. Rather, the process itself, and the resulting CIL and reputational costs, might effectively regulate both treaty participants and non-participants within the evolving regulatory framework for military AI in the future.

***

Zhixiong Huang is a Hongyi Chair Professor at Wuhan University School of Law and Director of the Wuhan University Institute for Cyber Governance.

Haokun Yu is a PhD Candidate in International Law at Wuhan University School of Law and a Visiting Researcher at Georgetown University Law Center.

The views expressed are those of the authors, and do not necessarily reflect the official position of the United States Military Academy, Department of the Army, or Department of Defense. 

Articles of War is a forum for professionals to share opinions and cultivate ideas. Articles of War does not screen articles to fit a particular editorial agenda, nor endorse or advocate material that is published. Authorship does not indicate affiliation with Articles of War, the Lieber Institute, or the United States Military Academy West Point.

Photo credit: