On July 28, the New York Times ran a story with the headline “U.S. Says Russia Tested Cruise Missile, Violating Treaty.” Surprisingly, this story about an arms control treaty from the Cold War — a treaty that I helped negotiate — offers some important lessons for the future regulation of autonomous weapons systems.
The treaty that the Russians allegedly violated is the Treaty Between the United States of America and the Union of Soviet Socialist Republics on the Elimination of Their Intermediate-Range and Shorter-Range Missiles (the INF Treaty), signed in December 1987. Under the treaty, “neither Party shall . . . produce or flight-test any intermediate-range missiles or . . . produce, flight-test or launch any shorter-range missiles.” INF Treaty, art. VI. Thus, if the missile that the Russians flight-tested was an “intermediate-range missile” or a “shorter-range missile,” they violated the treaty.
The treaty covers two types of missiles: ground-launched ballistic missiles (GLBMs) and ground-launched cruise missiles (GLCMs). According to the treaty, “the term ‘intermediate-range missile’ means a GLBM or GLCM having a range capability in excess of 1000 kilometers but not in excess of 5500 kilometers.” Similarly, “the term ‘shorter-range missile’ means a GLBM or GLCM having a range capability equal to or in excess of 500 kilometers but not in excess of 1000 kilometers.” INF Treaty, art. I.
It is very straightforward to determine the range capability of a GLBM. Article VII says: “The range capability of a GLBM . . . shall be considered to be the maximum range to which it has been tested.” This rule makes sense. U.S. officials can easily determine the maximum range to which a Russian GLBM has been tested, and vice-verse. Moreover, military officials would not want to use a GLBM in combat to attack a target at a range much further than the range to which the missile had been tested. To have confidence that a GLBM will work properly, they need to test the missile at something close to its maximum range. Thus, the treaty rules minimize the likelihood of any serious dispute about the range capability of a GLBM.
However, the Russians reportedly flight-tested a GLCM, not a GLBM. The rule for determining the range capability of a GLCM is quite different. Article VII says: “The range capability of a GLCM . . . shall be considered to be the maximum distance which can be covered by the missile in its standard design mode flying until fuel exhaustion, determined by projecting its flight path onto the earth’s sphere from the point of launch to the point of impact.” As a former member of the U.S. delegation that negotiated the treaty, I can attest to the fact that we knew when we drafted the treaty that this language was potentially problematic. It is difficult for the United States to determine the “maximum distance which can be covered” by a Russian GLCM in its “standard design mode,” especially if the Russians are unwilling to share detailed design information. The Russians have the same problem in reverse.
So, why not specify in the treaty that “the range capability of a GLCM” is “the maximum range to which it has been tested”? The answer is that engineers do not need to test GLCMs at their maximum range to be confident that they will operate effectively at their maximum range. Military officials can have high confidence that a GLCM will perform effectively even if it has been tested at only half the maximum range. Therefore, the dispute between the United States and Russia about the alleged treaty violation hinges, to some extent, on the “range capability” of the GLCM that the Russians flight-tested. They have better information than we do. But they are (not surprisingly) disinclined to share that information.
There are lessons here for the future of arms control. Autonomous weapons, known colloquially as “drones” or “killer robots,” may well be the weapons of the future. There are many varieties of autonomous weapons, but some of them have much in common with GLCMs. In particular, it is likely that engineers and military officials will be able to have high confidence in the performance of some autonomous weapon systems without having to test them at their maximum range. This fact poses a problem for any future effort to conclude an arms control agreement to regulate autonomous weapons.
One way to address this problem would be to craft an agreement that covers all autonomous weapons of a designated class with a range capability greater than zero. A zero limit makes it much easier to monitor compliance. However, range capability is a useful criterion for distinguishing between less threatening and more threatening weapons. An agreement that sets a threshold of zero might simply be unacceptable because it would ban weapons that states want to be able to use.
A second option would be to require fairly detailed information exchange. Suppose that a prospective treaty on autonomous weapons banned a specified class of weapons with a range capability greater than 1000 kilometers, for example. A treaty provision mandating detailed exchange of information about weapon systems could help minimize disputes about the range capability of new weapons. On the other hand, the barriers to information exchange are significant. Military officials might have legitimate concerns that sharing too much information about weapons not prohibited by the treaty would undermine the military effectiveness of those weapons. But some information exchange about non-prohibited weapons is essential for State A to be confident that State B is not testing prohibited weapons under the guise of testing non-prohibited weapons.
There is no easy solution to this problem. One thing is certain, though. If states want to develop a workable arms control regime for autonomous weapons, they will need to think hard about how to monitor compliance with treaty rules that use range capability as a criterion for distinguishing between prohibited and permitted weapons.