“Can Speech Kill?” was the title question to last week’s fascinating show with Lynne Tirrell. The obvious answer, it seems, should be: yes, but not directly.

Unless magic exists, uttering a few words won’t by itself cause someone to drop dead. But there are two kinds of case where talking can obviously cause death and even be counted as murderous. First—as one caller pointed out—is that of giving an order. If someone orders a subordinate to murder and the subordinate does, then the superior has committed murder as well. Second—which wasn’t discussed on the show—one can help commit murder by knowingly giving information to a person intent on killing. If you want to kill Jones and if I, knowing this, tell you Jones’s whereabouts, my words have helped you kill. In both kinds of case, words kill, but not directly; rather, they communicate information that incites or enables one human being (using whatever violent tools) to kill another.

But Tirrell and our two hosts, Ken Taylor and Josh Landy, weren’t focused on the obvious cases. They were focused on hate speech, Tirrell’s area of active research, which is what made the show philosophically interesting. If one person engages in hate speech against another—using racial slurs or de-humanizing language such as “cockroaches” or “rats”—can that language be counted as killing or contributing to killing other people?

Unlike in the cases of direct order or knowingly informing a murderer, it’s not obvious here that the answer is yes. But a case that it is, as discussed on the show, can be built by considering the role of propagandistic hate speech in genocides, such as in Nazi Germany or Rwanda in 1994. Der Stürmer, for example, was a widely-read anti-Semitic publication that ran in Germany starting in 1923 and going through the end of WWII; its pages were full of slanders, caricatures, and calls for extermination of Jews, which arguably facilitated the holocaust. In Rwanda, the most prominent hate speech outlet during the genocide was the private radio station Radio Television Libre des Milles Collines (RTLM), which regularly called for a “final war” to “exterminate the cockroaches”—that is, the ethnic Tutsis. And as Philosophy Talk contributor David Livingstone Smith would point out, such dehumanizing language removes psychological resistance to, and even motivates, mass killing.

Staunch free speech advocates, however, balk at such arguments. While granting that direct orders to kill or deliberately informing a known murderer constitute illegal acts (the “obvious” cases), a free speech activist would maintain that, in the propaganda/hate speech cases, it’s the killer who kills and not the talker. After all, in cases of hate speech, the physical killer still has to make a choice and is under no compulsion from the propagandist, however horrible the speech may be.

Is either argument right?

My goal here is to point out an uncomfortable truth. It seems to me that there exists—both conceptually and in actuality—a messy continuum between cases where a speaker is obviously a murderer in virtue of using certain words and cases where a speaker, though saying bad things, can’t be assigned blame for anyone’s death. The messiness of this continuum, furthermore, fuels the contentiousness of the issue by making it hard to find a stable or clear solution.

Here’s the continuum. (This version of it is fictionalized for the sake of clarity, but it wouldn’t be terribly hard, I think, to find real cases that correspond to various points on the continuum.)

First, consider a man with robot that wields a knife. Suppose the robot is programmed with speech recognition technology. The robot recognizes imperative sentences of the form “Kill [NAME]” and responds by seeking the person whose name was used and stabbing them to death. Clearly, if the man knowingly utters “Kill Frank Jones” and the robot goes and stabs Frank Jones to death, then the man has committed murder by speaking. The words were lethal.

Second, consider a woman with a similar robot, which also has a knife and speech recognition software. But this robot doesn’t just recognize imperatives. This one also recognizes declarative sentences of the form “[NAME] is a cockroach.” This robot responds then in the same way as the first; that is, it goes and stabs the named person to death. Obviously, if the woman knowingly utters “Frank Jones is a cockroach” in the presence of the robot and Frank Jones gets stabbed to death, the woman has committed murder.

Third, consider a man who has a robot like the woman’s—except with one difference: this time the robot is probabilistic. Saying “Frank Jones is a cockroach” in the presence of this robot won’t guarantee that the robot goes and stabs Frank Jones; rather, it puts the robot in a state that has a 35% chance of transitioning into the state where it sets out to stab Frank Jones to death. Here, we can still see clearly that if the man knowingly utters “Frank Jones is a cockroach” in the presence of the robot and Frank Jones is then killed by the robot, the man has murdered Frank Jones by speaking those words. But what if the man says the words around the robot and it doesn’t transition into the murderous state? Or what if the chances were far lower than 35%—say they were 0.1%?

Fourth, consider a woman who has a bunch of knife-wielding, speech-recognizing robots that don’t just understand individual names; they also process ethnicity names. Each robot, on processing the sentence “[ETHNICITY NAME]s are cockroaches” transitions into a state where it is somewhat more likely to go out and kill a random person of that ethnicity. But let’s add that the woman is only guessing at what the probability change is and has no idea which particular person or people would be killed. Still, I think it’s obvious that the woman has murdered by speech, if she utters, say, “Tutsis are cockroaches” in the presence of the robots and they go out and kill one or more Tutsis.

Fifth, replace the robots from the last example with people who undergo an increase in likelihood of killing by way of experiencing increased motivation to kill. These (potentially) murderous listeners still have their own goals and motivations, but these motivations are pushed and pulled by a given speaker’s sentences of the form “[ETHNICITY NAME]s are cockroaches.”

Sixth, replace the individual speaker with a large group of people who are saying “[ETHNICITY NAME]s are cockroaches,” no single one of whom is sufficient to increase the likelihood that other people go out and kill, but where the existence of a group of voices is sufficient to increase other people’s motivations to kill. In such a case, can we say that the speech of any one person in this group of voices was lethal?  

This sixth spot on the continuum, arguably, is a simplified approximation of the situations for most of the propagandists and hate speech mongers in Nazi Germany and in 1994 Rwanda.

So we might at this point reason as follows. Since the hate speech in Nazi Germany and Rwanda occupies the sixth spot on the continuum, and since the sixth spot on the continuum bears obvious structural parallels to positions 1-5, where speech was more clearly murderous, the propagandists in Nazi Germany and in Rwanda also count as having murdered by their words.

I actually think that argument (or a filled-in version of it) is more or less right, and for that reason I agree with Tirrell. But we also start to see why this is an intellectually uncomfortable territory, for we could easily continue along the continuum to a point where, though someone is speaking in a deplorable fashion, that person can’t be said to have murdered by words.

Somewhere farther along the continuum, say the nth spot, no one increases the likelihood that anyone is killed by uttering an ethnic slur, since no one in the society at that time is remotely murderous out of ethnic hatred. However, the following counterfactual is true at spot n: if there were a chorus of many people uttering such slurs, then the likelihood of someone’s being killed for their ethnicity would go up. Now say (without the antecedent of that counterfactual being true) someone in situation n utters a slur about an ethnicity. Two things to me at this point seem true. First, it’s still bad; slurs are ugly and demeaning, even if no one is at risk of being killed. Second, such speech at spot n is not lethal. Why not? Because in context, there is no chance of its causing someone to die. (Furthermore, labeling such speech at position n “murderous” would dilute the category of interest [lethal speech] to the point where it’s uninformative.)

So the pendulum swings the other way. Someone might just as well trace a line from the nth spot to the sixth spot and say, “See, because the sixth spot resembles n in various ways, the speech occurring in spot 6 should also not count as lethal.” I myself would reject such an argument, and in arguing for such a rejection, I’d point to morally relevant differences between spot n and the sixth spot.

But engaging that argument is not the point of this blog. The point is to show why this is a difficult problem—to highlight the complexity of the problem space. We could just as well extend the continuum out in multiple directions along multiple dimensions. And that would further show that there are no clean dividing lines between “obvious” cases and the non-obvious ones. Nor is this multi-dimensional continuum merely hypothetical. Note, for example, that the radio voices on RTLM didn’t just call Tutsis “cockroaches.” They also said things such as “they must be exterminated” (which resembles an order) and gave information as to the whereabouts of Tutsis in ways that enabled the executors of the genocide to go and find them (which constitutes knowingly informing a murderer). We could just as well construct continua from those cases to other bad but non-lethal ones—and then back to direct orders and targeted lethal informing. These continua also won’t have bright lines.

These theoretical difficulties bleed over into practical difficulties. Free speech and preservation of life are both, in their own ways, sacred values. And sacred values seem to demand clear, bright lines in order to protect and preserve them. The existence of the sort of continua I’ve pointed out here guarantees that it will never be obvious where the lines around the sacred values should be drawn; there will always be an element of the arbitrary. That’s our practical and intellectual predicament.

My own approach, if I could choose, would be similar to the approach to speech taken in present day Germany. In Germany, speech generally is free, but hate speech (such as denying the holocaust) and other speech inciting ethnic violence are against the law. This approach recognizes the lethal danger of contained in some forms of speech, while doing very little that would inhibit the free flow of information, which is crucial to democracy. I find it ironic yet strangely illuminating that this reasonably effective approach to the problem (as much as could be hoped) should have been pioneered in a country in which speech, historically, proved to have its most lethal powers. I hope we never stop learning from that example. Speech can kill, even if not directly.

Relevant links:

Last week’s show: https://www.philosophytalk.org/shows/can-speech-kill

Lynne Tirrell: https://philosophy.uconn.edu/2017/07/25/welcome-lynne-tirrell/

David Livingstone Smith’s book, Less Than Human: Why We Demean, Enslave and Exterminate Others: https://www.davidlivingstonesmith.com/project-09

BBC reporting on Rwandan radio propaganda: http://news.bbc.co.uk/2/hi/africa/3257748.stm

On Der Stürmer: https://en.wikipedia.org/wiki/Der_Stürmer

On Germany’s current law against hate speech: https://en.wikipedia.org/wiki/Volksverhetzung

On recent development in German hate speech legislation: https://www.nytimes.com/2017/06/30/business/germany-facebook-google-twitter.html