A brief look at reducing the efficiency of genocides

In 2012, while working on a humanitarian base in Haiti, I found a copy of Dr. James Orbinki’s book An Imperfect Offering on our community bookshelf. As a doctor, humanitarian, and a past president of MSF, he shares deep insights through the telling of often heartbreaking stories. This week I decided to revisit his book. While there are many pages that force anyone with empathy to give pause, one line stopped me and has stuck with me.

“The genocide in Rwanda was the most efficient genocide of the twentieth century...”

The word “efficient” stopped me. I use it all of the time. I read about, advise, and optimize for it all of the time. I have never used it as a descriptor in such a shameful context. It makes complete sense, however. One of the great tragedies of genocides is the pure “efficiency” of it all. Considering this, I began wondering if there is an opportunity to reduce the brutal efficiency.

I have summarized my thoughts here and shared my full process below. I would be grateful for feedback and discourse. My quick exploration went roughly as follows:

  • UN definition of genocide

  • A “back of the envelope” calculation of how “efficient” the world’s modern genocides had been

  • Identify the steps of genocide

  • Identify opportunities to interfere

  • Use two “case studies” or “vignettes” to:

    • consider a process for intervening

    • estimate the possible impact

    • consider the cost of an intervention

  • Note the weaknesses of this post

My current thoughts following this < 50-hour exploration of the topic:

  • The shortest events tend to be the most “efficient.” With two exceptions, the top ten most efficient genocidal events lasted less than 9 months. I speculate that these condensed events likely have the highest ROI potential for any intervention. They also likely suffer from the slow bureaucracy of international agencies.

  • Of the entities attempting to identify the causes and stages of genocides, I found Dr. Gregory Stanton’s “Ten Stages of Genocide” to capture the consensus and be the most distinctly identifiable. (His organization also has a genocide watch and warning function.)

  • During the Rwandan genocide, the US military passed on the opportunity to save lives at the cost of ~$40,800 per life saved (completely ignoring all of the DALY’s, spillover effects, and other suffering). That is ~2.25x the “Cost-effectiveness of counterfactual government health spending” estimate used by GiveWell.

  • Speculative opportunities for civilian intervention:

    • Communication Interference:

      • In a future situation of increased immediate violence incited and organized by a dominant media outlet (similar to the Rwandan genocide), communications interference could prove an effective and efficient intervention. In the age of social media and cell phones, the costs could be lower and the opportunities to intervene could be greater.

      • Likely illegal (though not as illegal as genocide, or even a single murder), it could be possible to jam a radio signal with limited technology. (And at a lower cost than the 1994 rate of $8500 per flight hour for a US military signal-jamming plane.)

      • Another option would be for a DMI-style emergency response arm to offer a counter-voice. (Cited as an option by the US military during the Rwandan genocide.)

      • Facebook’s role in the Myanmar Genocide of the country’s mostly Muslim Rohingya minority group instigated considerations into the platform’s role in shutting down violent and targeted rhetoric (credit to Aaron Gertler for sharing this insight). Is there a role for a ML/​AI system to improve the detection of the fake accounts used by Myanmar? Is there value to an organized troll/​disinformation scheme to disrupt their attempts to organize and instigate on the platform? From an operations perspective, it could be an impactful career to work with these platforms to have a playbook in place to respond to these events.

    • Activist Investing:

      • While the likelihood of having an impact is uncertain, there may be opportunities to use infrastructure interests (aligned interests) to lobby state actors to either intervene or reduce the conflict.

      • The cost of the intervention is tied to the investment. So it could be negative (positive economic outcome) or up to the value invested (company fails). With relatively “safe” investments in well-established companies, the risk could be mitigated.

If you are interested in further exploring my full thought process, I have shared it below:

(1) Defining Genocide:

Genocide is defined as...

“… any of the following acts committed with intent to destroy, in whole or in part, a national, ethnical, racial or religious group, as such:

(a) Killing members of the group;

(b) Causing serious bodily or mental harm to members of the group;

(c) Deliberately inflicting on the group conditions of life calculated to bring about its physical destruction in whole or in part;

(d) Imposing measures intended to prevent births within the group;

(e) Forcibly transferring children of the group to another group.”

— Convention on the Prevention and Punishment of the Crime of Genocide, Article 2

(2) Estimating Efficiency

I’ve created a Google Spreadsheet to create a rough calculation of the efficiency of genocides, as measured by the number of Individuals Killed Per Day. My calculation should be considered a very rough “back of the envelope” calculation. I started with the 70 entries found on Wikipedia’s List of genocides by death toll. I reduced the data set to 48 entries by removing incomplete entries and all entries before 1894. My intention is that this trimming will reduce problematic errors and improve the modern contextual relevance, all while keeping the data set as large as reasonable.

Process for calculations:

  • Estimate the total duration of the genocides in days (end day included)

  • Divided the low and high estimates of fatalities by the number of days

  • Calculate a simple mean: (L+H)/​2

  • Reorder by mean

Once ordered by the mean, by my rough calculations, it appears that Dr. Orbiniski was mistaken. Tragically, four other genocidal events could be categorized as more “efficient” killing occurrences. Worth noting though, one of those events was inextricably tied to the events in Rwanda.

(3) Stages of a Genocide

I then moved on to what factors can identify a genocidal event. I found the UN and other organizations to offer unproductively vague identification factors. In my personal opinion, Dr. Gregory Stanton: “Ten Stages of Genocide” to be the most identifiable.

It is worth going on a brief tangent to highlight his tools, “Genocide Watch,” an interactive international map that identifies countries engaged in the Ten Stages, and Genocide Alerts, which does what it sounds like it does.

His ten stages include:

  1. Classification

  2. Symbolization

  3. Discrimination

  4. Dehumanization

  5. Organization

  6. Polarization

  7. Preparation

  8. Persecution

  9. Extermination

  10. Denial

Many organizations work to reduce the tensions (stages 1-6) that can spark genocidal events. There are organizations that lobby to recognize events as genocide and ensure that they are taught historically as such. I have not found, in my limited search, any organization beyond the UN that provides operational responses to the Persecution and Extermination stages.

(4) Opportunities to Intervene (vignettes)

Rwanda: Radio Disruption

When considering interventions, I was immediately curious about the Rwandan Genocide. The violence was primarily limited to 100 days and the radio was a significant coordinator and instigator. Throughout the conflict, there were international calls to stop the broadcasts. So influential were the radio broadcasts, that in 2003, three “media leaders” were convicted of “ genocide, incitement to genocide, conspiracy, and crimes against humanity, extermination, and persecution” by the UN International Criminal Tribunal. I was even able to find a 2014 Harvard study that sought to quantify the effects of the RTLM radio broadcasts on the violence. It states that “approximately 10% of the overall violence, can be attributed to the station.” It also estimates that “approximately 6.5 percent (28,000 persons) of the individual violence and 29 percent (22,000 persons) of the collective violence can be attributed to the broadcasts.”(2012 Working Paper PDF)

One research article casts doubt on the proxy used in the Harvard study.

Beyond spending months cultivating toxic and divisive rhetoric that contributed to the culture of violence, the radio station took direct steps in organizing the killings (i.e. - broadcasting names and addresses of individuals). If that result is true, and (falsely) assuming a uniform distribution of violence, then 50,000 − 107,100 lives could have been spared if the radio had been prevented from broadcasting. In Foreign Affairs, Alan Kuperman proposed that President Clinton “could not have known that a nationwide genocide was underway” until about two weeks (14%) into the killing. An action then still could have had considerable impact.

There were multiple reasons that the Americans failed to act, but concerning the radio station, it may be easiest to cite The Atlantic analysis directly;

“The country best equipped to prevent the genocide planners from broadcasting murderous instructions directly to the population was the United States. Marley offered three possibilities. The United States could destroy the antenna. It could transmit “counter-broadcasts” urging perpetrators to stop the genocide. Or it could jam the hate radio station’s broadcasts. This could have been done from an airborne platform such as the Air Force’s Commando Solo airplane. Anthony Lake raised the matter with Secretary of Defense William Perry at the end of April. Pentagon officials considered all the proposals non-starters. On May 5 Frank Wisner, the undersecretary of defense for policy, prepared a memo for Sandy Berger, then the deputy national-security adviser. Wisner’s memo testifies to the unwillingness of the U.S. government to make even financial sacrifices to diminish the killing.

“We have looked at options to stop the broadcasts within the Pentagon, discussed them interagency and concluded jamming is an ineffective and expensive mechanism that will not accomplish the objective the NSC Advisor seeks.

International legal conventions complicate airborne or ground based jamming and the mountainous terrain reduces the effectiveness of either option. Commando Solo, an Air National Guard asset, is the only suitable DOD jamming platform. It costs approximately $8500 per flight hour and requires a semi-secure area of operations due to its vulnerability and limited self-protection.

I believe it would be wiser to use air to assist in Rwanda in the [food] relief effort …”

The plane would have needed to remain in Rwandan airspace while it waited for radio transmissions to begin. “First we would have had to figure out whether it made sense to use Commando Solo,” Wisner recalls. “Then we had to get it from where it was already and be sure it could be moved. Then we would have needed flight clearance from all the countries nearby. And then we would need the political go-ahead. By the time we got all this, weeks would have passed. And it was not going to solve the fundamental problem...”

Just working with the $8500 per flight hour, that number works out to be $204,000 per day, assuming the radio station played non-stop. Given that the conflict lasted 100 days, and 6.5% of the individual violence (ignoring spill-over effects) could be attributed to the broadcasts, if you could disrupt 1 day (1%) of transmissions, you could plausibly reduce 6.5% of the individual violence by 1%. Using the mean, that equals ~5 lives saved per day of interrupted transmission. Saving 5 lives for $205,000 works out to $40,800 per life saved (~10x the current cost of life saved through the Helen Keller Initiative ($4,016) per GiveWell estimates).

(In response to radio jamming being brought up yet again at a meeting, one Pentagon official responded; “Pru, radios don’t kill people. People kill people!”)

There are certainly weaknesses and limitations to this thought exercise. For one, the killings were not uniformly distributed. Secondly, the radio station began broadcasting in July 1993 and the serious killing did not begin until April 2004, allowing the station to “plant seeds” of divisive rhetoric, especially among young men. Additionally, it is unclear what response (i.e. backlash), if any, there would have been to attempts to stop the broadcasts.

Somalia: Investment & Corporate Lobbying

Between May 1988 and March 1989 there was a state-sponsored genocide of 50,000-100,000+ Isaaqs, a clan in the north. Notably, there was considerable American influence in the area. Throughout the 1980s US oil companies were finding deposits in Somalia. While they remained undeveloped the US established naval and military bases in the country, overseeing the Persian Gulf and the Red Sea. The Times reported that around two-thirds of Somalia was allocated for oil exploration, specifically by four US oil companies (Conoco, Amoco, Chevron, and Phillips) In 1988, the GDP per capita of Somalia was ~$148 (current US$ WB). The investment by the oil companies and the US military was a significant financial support to the country.

In August 1988, Amnesty International began running stories of the killings that would turn into a genocide. At the time (August 15, 1988), an investor could have purchased shares in Chevron for $11.25. Over the remainder of the genocide, the stock had an annual return rate of ~50%. An investor (or group of investors) could have purchased enough shares to advocate for the company to speak out to the relevant parties (investment activism). Further exploration of the strategy and efficacy of investment activism would be highly valuable in such a strategy. It is likely highly uncertain what effect, if any, this would have had. Would the oil companies have been influenced by activist investors (to complain/​protest/​call on the US/​divest from the region)? Would the applicable government(s) have been influenced by any action taken by the oil companies? Even if the impact was effectively zero, the financial costs would effectively be negative (a positive economic outcome). This case study looks at what holding the investment would look like 25 years later (though I am confused about the cited share price).

(A more modern context: Aaron Gertler referred me to Facebook’s role in the Myanmar genocide. Since October 09, 2016, when the genocide is cited to have begun, Facebook’s stock has had an ARR of 22.19% and a $10,000 investment would have more than doubled to $22,625.70. On Nov. 08, Aung San Suu Kyi’s ruling party won re-election.)

Weaknesses of this post:

  • The killings were not uniformly distributed during the windows analyzed. It is possible that smaller windows would have considerably higher “efficiency” rates.

  • My curating of the dates and numbers was based on quick research with little follow-up. There are likely flaws.

  • I have not yet made a calculation that considers the proportion of the targeted population.

  • My considerations completely neglect the impact on those who survived with injury, trauma, or other loss.

  • The mean calculation is unweighted.

  • I am not fully confident in my calculation of the value of the investment considered.

  • My 1% calculation is certainly over-simplified to provide a perspective in a complicated situation.

[I would like to thank Aaron Gertler for his encouragement at EA conferences to write on the Forum, including taking the time to host workshops, share resources, and offer his personal assistance. It is likely that I would not have written this, my first post, without his nudging (at least not in the same time frame). Your time and effort influenced behavior change. Also, thanks to the kind folks over on the FB Effective Altruism Editing and Review page who shared their thoughts (specifically those who messaged me privately).]