What We Talk About When We Talk About Killer Robots: the Prospects of an Autonomous Weapons Treaty

• Publication year: 2023
• Citation: Vol. 51 No. 2

What We Talk About When We Talk About Killer Robots: The Prospects of an Autonomous Weapons Treaty

[Page 501]

Collin M. Douglas^*

Table of Contents

I. Introduction........................................................................................502

II. Background.........................................................................................504

A. Use and Regulation of Autonomous Weapons...................504
B. Calls for an Autonomous Weapons Treaty........................508
C. Nuclear Weapons...............................................................513
D. Anti-Personnel Mines........................................................517
E. Blinding Laser Weapons....................................................520

III. Analysis..............................................................................................522

A. Factors Which Lead to Regulation....................................522

i. Public Image...............................................................524
ii. Dual Use.....................................................................526
iii. Strategic Value............................................................527
iv. Deployment.................................................................529
v. Proliferation Risk........................................................530
vi. Framing....................................................................... 532

B. What Sets Autonomous Weapons Apart and How This Affects Regulation.............................................535

IV. Conclusion.........................................................................................537

[Page 502]

I. Introduction

Artificial intelligence is rapidly advancing and being used in a variety of industries for a variety of purposes.¹ Rightly or wrongly, artificial intelligence has been deployed in sectors as diverse as finance, health care, criminal justice, and transportation.² As with most new technologies, it was only a matter of time until the development of artificial intelligence turned toward the bat-tlefield.³

The development of autonomous weapons led many to warn of the potential dangers of unleashing weapons controlled by artificial intelligence, also known as autonomous weapon systems (AWS).⁴ Most notably, this development led to the creation of the Campaign to Stop Killer Robots (CKR), which intends to do just that. CKR advocates for an international treaty banning the use and development of AWS.⁵ The organization touts the support of more than 200 non-governmental organizations and academic partners, thousands of artificial intelligence experts, and a majority of the public for a treaty banning the use of AWS.⁶

It would be easy to dismiss attempts to regulate AWS as unlikely simply because powerful states do not want to regulate these systems.⁷ While this

[Page 503]

lack of desire to regulate AWS use may partly explain why a regulatory framework has not come to be, there are subtler, more nuanced forces preventing its creation. A treaty regulating or banning autonomous weapons is unlikely to be created in the near future because of the specific characteristics of autonomous weapons and their contrast with other weapons which have been regulated, how these weapons are perceived by the general public, and how activists talk about them.

This Note will argue that a treaty regulating autonomous weapons systems is unlikely to come into being in the near future, and if one does, it will most likely follow the regulatory path of nuclear weapons by starting with non-proliferation. This Note will compare the conditions which brought about other similar and notable weapons regulations or weapons ban treaties with the current conditions surrounding, and the perception of, autonomous weapons. The analysis will also focus on specific characteristics of AWS and how these are distinguished from successfully regulated technologies. This Note will further discuss how AWS are less likely to be regulated because the subjective factors affecting their perceptions among the general public, activists, and policymakers indicate that there is a high value on their development and use but a low perceived value of their negative characteristics.

The conditions surrounding regulated weapons that will be examined are: (i) whether there is a well-publicized use of that weapon that creates a striking image for the public, (ii) whether there is a legitimate or valuable dual use for the weapon, (iii) the military or strategic value placed on the weapon, (iv) the length of time and extent to which the weapon was deployed before its regulation or ban, (v) the likelihood that the weapon would be used by non-state actors or rogue states, and (vi) the framing and messaging that those supporting a regulation or ban invoke in their campaigns.

This Note will examine factors leading up to and playing a part in the regulation or banning of weapons by the Convention on the Prohibition of the Use, Stockpiling, Production and Transfer of Anti-Personnel Mines and on their Destruction;⁸ the Nuclear Nonproliferation Treaty⁹ (and other nuclear weapons treaties generally); and Protocol IV to the Convention on Certain Conventional Weapons, otherwise known as the Protocol on Blinding Laser Weapons.¹⁰ This Note will then argue that specific characteristics of AWS

[Page 504]

complicate their road to regulation by examining and comparing the characteristics of AWS with the specific factors which led to the regulation or banning of other weapons, including their inherent features and how the threats posed by these weapons were discussed and perceived leading up to their regulation. By considering how the weapons are discussed by activists, members of the military, politicians, and other powerful and elite groups, this Note will examine the effects on how the weapons are perceived, and therefore, how they are regulated. The second part of this argument focuses on how autonomous weapons match up against those characteristics which led to the regulation or banning of other weapons and explores explanations of why AWS are less likely to be regulated. This comparative analysis provides insight into how to approach regulating AWS specifically, but also how to approach regulating weapons more broadly.

Part II will examine the details of what autonomous weapons are and how they are deployed. Part II will also discuss other treaties regulating or banning weapons, how those treaties came about, how the actors supporting those treaties framed them, and what societal ill the treaty was aimed at. Part Ill will compare those previous treaties and the weapons they regulate with the proposed autonomous weapons treaty and AWS. Part Ill will also argue that the factors that led to the regulation of nuclear weapons, anti-personnel mines, and blinding laser weapons do not favor the creation of an autonomous weapons treaty. Part IV will conclude by discussing how a change in the unsettled nature of autonomous weapon systems could affect the likelihood that they are regulated by a treaty.

II. Background

A. Use and Regulation of Autonomous Weapons

Autonomous weapons have not yet been widely deployed on the battlefield, but interest in their use and development is rapidly increasing. Increased interest and attention in AWS on the part of militaries is mirrored in the public discourse, where regulation is now a central discussion point.

An important distinction for understanding autonomous weapons is that they are not actually weapons. AWS are systems for controlling weapons. To illustrate this point, imagine a soldier holding a rifle. This image is typical of what someone would associate with the military, and the rifle in the soldier's hand is typical of the image someone would associate with the word weapon. The rifle and the soldier holding the rifle are a weapon and that weapon's operating system, respectively. The soldier is the system that operates the rifle. To automate this system, the rifle does not change, the soldier does. Automating that system turns the soldier into a computer program with the ability to hold, aim, and shoot that weapon without human intervention. This is why

[Page 505]

AWS are not technically weapons, but because they are human-less machines that operate weapons, they are themselves referred to as weapons.

The lack of a widely agreed-upon definition of autonomous weapons complicates their discussion.¹¹ The United Nations Institute for Disarmament Research compiled the different ways that states and organizations have proposed to define AWS, which can be broken down into three categories: a technology-centered approach, a human-centered approach, and a function-centered approach.¹²

The technological approach focuses on a detailed, technical definition of autonomous weapons.¹³ This is the type of definition that is typical to arms control treaties, one which deals in "aspects such as technical specifications, range, payload, and intended operating environment."¹⁴ The most apparent problem with this definition is the fact that "[a]utonomy is a characteristic, not a thing in and of itself. . . . It could be applied to different parts of any system. . . . You might have an adjustable object with an autonomous mode, automatic mode and human-operated mode. It will be difficult to capture the variety of characteristics."¹⁵ The concept of autonomy could be applied to any number of systems, lethal or non-lethal, for any number of purposes, at any level of sophistication. A technical definition of autonomy would be so convoluted as to be unworkable.

The human-centered approach to defining autonomous weapons systems is the most intuitive. This definition is human-centered in that it focuses on what a human is not doing, or in other words, what level of control the human has given over to the machine. The language associated with this definition, whether a human is "in, on, or out of the loop,"¹⁶ conveys whether...