One of the more depressing elements in the narrative at the CCWUN Experts’ Meeting this week has been the argument, repeated by a number of autonomous weapons proponents both in plenary and discussion, that an advantage of such weapons is the following: unlike human soldiers, they would never commit rape.
This is but a new twist on a broader argument most prominently made by Georgia Tech Professor Ronald Arkin in his book Governing Lethal Behavior in Autonomous Robots, but shared by many AWS proponents. The argument is that autonomous weapons might not just be good for national security, they would be good for human security, because they would do better than human soldiers at complying with war law. This will allegedly occur through a combination of situational judgment, ability to distinguish civilians from combatants, and good old-fashioned restraint. So the argument goes, a machine would never get tired, or go berserk when its buddy died, or make lethal mistakes due to anger, jealousy, or malice.
Certainly a robot warrior would never rape.
Below I will first address a number of problems with the rape argument, and then show how these problems cast doubt on the broader (and I would say over-optimistic) prediction of machine just-warrior-hood.
Let’s start with rape. First of all, what definition are these delegates using that leads them to think that rape could not be committed by an autonomous machine? Most likely, I figure, one that locates the act of and propensity for rape in biological male sexual impulses. But here is the actual international definition of rape, which is codified in the 1998 Rome Statute of the International Criminal Court:
“The perpetrator invaded the body of a person by conduct resulting in penetration, however slight, of any part of the body of the victim or of the perpetrator with a sexual organ, or of the anal or genital opening of the victim with any object or any other part of the body. The invasion was committed by force, or by threat of force or coercion, such as that caused by fear of violence, duress, detention, psychological oppression or abuse of power, against such person or another person, or by taking advantage of a coercive environment, or the invasion was committed against a person incapable of giving genuine consent.”
If I can imagine a machine inflicting lethal violence on a human being through projectile, I have no problem whatsoever imagining a machine forcibly penetrating the body of a male or female human being in ways consistent with this definition. Of course, a machine programmed to kill according to the algorithm Professor Ron Arkin has in mind probably wouldn’t do that as a matter of course. But who is to say that a machine couldn’t be programmed to rape instead of kill if some nefarious dictator had his hands on the most advanced model as a result of a CCW-legitimated robotic arms race and decided to use these as tools of terror instead of human security?
Underlying these techno-optimists’ thinking is an important fallacy: they assume that war rape is a crime committed opportunistically by soldiers, often untrained and lawless rebel groups, rather than ordered by state. Yet this is one of many “myths” of wartime sexual violence identified in a recent United States Institute of Peace Report:
Several recent studies have found that state armed groups are far more likely than rebel groups to be reported as perpetrators of rape and other sexual violence. The recent PRIO study of African conflicts found that, between 2000 and 2009, armed state actors were more likely to be reported as perpetrators than either rebel groups or pro-government militias. Between 1989 and 1999, state forces and other armed actors were equally likely to be reported as perpetrators.
If conflict-related rape is commonly perpetrated by state militaries, then the assumption may be that these crimes are the opportunistic behavior of bad apples rather than strategic policy: therefore the key problem for states is preventing soldiers from raping out of emotion or lust and that by creating emotionless, lustless soldiers one will solve the problem. But if so that’s also an error: history shows states often order rape. During World War II, Korean women were forcibly enslaved and raped as part of Japan’s “comfort women” policy. During the genocidal violence that accompanied Bangladesh’s secession from Pakistan, tens of thousands of women were raped on orders of the West Pakistan military. In Bosnia-Herzegovina, male soldiers were ordered to rape on pain of castration and death. In Peru and Guatemala, a metanalysis by Michelle Lieby shows a majority of rapes were carried out by the state, many systematically.
These are not isolated anecdotes. Harvard University Professor Dara Cohen has collected quantitative data on rape in civil wars cross-nationally between 1980-2009. She found that of the 86 major civil wars during this period, 67 conflicts have reports of state perpetrators of rape. According to Cohen, her dataset also show that 85% of those conflicts include reports of rape in detention, an act which implicates states – the same states that would have autonomous weapons in their arsenals if proponents have their way. In an email exchange this evening with Professor Cohen on the idea that autonomous weapons could “never rape” she replied:
“What I can say is that the vast majority of reported wartime rape is perpetrated by states, and a lot of that seems to be in the context of detention and torture. If sexualized torture is a common tool of states for interrogation and punishment during periods of war, then the “human-ness” of the torturer is moot. A world with robot sexual torturers is actually a pretty scary one: robots also lack morality, so one could imagine an extremely brutal machine that might have the ability to perform horrible tortures that most humans would find repugnant or unbearable.”
The reason that none of this occurs to delegates could be good old fashioned gender non-awareness: how many male experts in international disarmament law are also experts in conflict-related sexual violence?* But it could also be a function of the undue techno-optimism that motivates not just the rape argument but all arguments about AI humanitarianism. Simply put, when AWS proponents make the argument that AWS would commit fewer war crimes than human soldiers, they seem to mean fewer war crimes than the average human soldier whose state wants her not to commit war crimes.** They overlook the number of war crimes that occur because human soldiers fail to disobey unlawful orders from the state.
It stands to reason that robotic soldiers would be even less likely to disobey orders to kill noncombatants, torture, rape, etc – in fact if they are a product of their programming it is plausible they would be almost certain to carry out unlawful orders to commit war crimes, genocide, or crimes against humanity, with no conscience, empathy or independent moral judgment to get in the way. Indeed this is a key counter-argument made by Human Rights Watch in its new report out today rebutting proponents’ claims.
“Due to their lack of emotion, fully autonomous weapons could be the perfect tools for leaders who seek to oppress their own people or to attack civilians in enemy countries. Even the most hardened troops can eventually turn on their leader if ordered to fire on their own people or to commit war crimes. An abusive leader who can resort to fully autonomous weapons would be free of the fear that armed forces would resist being deployed against certain targets.”
Of course both proponents and opponents of AWS are trading in hypotheticals. But in a counter-factual argument the best predictor of human practice in future is the past.
In short the techno-humanitarian argument rests on a logical paradox: humans are less likely than robots to comply with the law, but it is these untrustworthy humans who would be the principals in any scenario where robotic agents were unleashed to carry out the law. So even if Ron Arkin is right that a humanitarian government could create the perfect humanitarian robotic soldier,*** it is equally true that a dictator could create the perfect tool of repression. Robots could be programmed as just warriors, and they could as easily be programmed to kill, to restrain, to torture and yes to rape. Saying otherwise blithely not only miscommunicates what rape is and how it occurs, it wantonly mis-states what we do know about the determinants of war crimes. And perhaps most of all it encourages us to make policy based on a naively utopian expectation, one unsupported by any empirical evidence and dramatically at odds with the history of organized warfare.
*Indeed this is another point made to me in hallways by female participants who were offended by the rape references: it’s not as these men are champions of sexual violence reduction across the board, they said, or knew anything about or were involved in any way in gender violence initiatives. Rather, it seemed to many women present that rape victims’ bodies were simply being referenced as a tool of political argument to justify weapons development. This is particularly egregious given the lack of gender representation on the panels, but that’s a subject for another post.
**I am not even sure that this argument holds up well, but that is the subject of a different post.
***I find this argument highly suspect as well, but again that is the subject of a different post. Stay tuned!