09/12/13

Tech challenge develops algorithms to predict atrocities

Mass Grave_Albert Gonzalez Farran_Unamid
Copyright: Albert Gonzalez Farran/Unamid

Speed read

  • The algorithms emerged from a global tech contest to prevent atrocities
  • Their predictions are partly based on data on past violent events
  • The models could help governments and charities identify risks and intervene

Send to a friend

The details you provide on this page will not be used to send unsolicited email, and will not be sold to a 3rd party. See privacy policy.

Mathematical models that use existing socio-political data to predict mass atrocities could soon inform governments and NGOs on how and where to take preventative action.

The models emerged from one strand of the Tech Challenge for Atrocity Prevention, a competition run by the US Agency for International Development (USAID) and NGO Humanity United. The winners were announced last month (18 November) and will now work with the organiser to further develop and pilot their innovations.

The five winners from different countries who won between US$1,000 and US$12,000, were among nearly 100 entrants who developed algorithms to predict when and where mass atrocities are likely to happen.

Around 1.5 billion people live in countries affected by conflict, sometimes including atrocities such as genocides, mass rape and ethnic cleansing, according to the World Bank’s World Development Report 2011. Many of these countries are in the developing world.

“Algorithms are vital to the prediction and prevention of atrocities.”

David Mace, California Institute of Technology

The competition organisers hope the new algorithms could help governments and human rights organisations identify at-risk regions, potentially allowing them to intervene before mass atrocities happen.

The competition started from the premise that certain social and political measurements are linked to increased likelihood of atrocities. Yet because such factors interact in complex ways, organisations working to prevent atrocities lack a reliable method of predicting when and where they might happen next.

The algorithms use sociopolitical indicators and data on past atrocities as their inputs. The data was drawn from archives such as the Global Database of Events, Language and Tone, a data set that encodes more than 200 million globally newsworthy events, recording cultural information such as the people involved, their location and any religious connections.


Spotting patterns

“Algorithms are vital to the prediction and prevention of atrocities as computers have the unique capability of seeing patterns in large data sets that humans often miss,” says David Mace of the California Institute of Technology, United States. Mace’s entry won him an ‘ideation’ award, a subcategory of the competition that rewarded unique new concepts for algorithms.

Mace’s algorithm — as yet only at the design stage — views the world as a mosaic of harmful and neutral regions tenuously interacting with each other. It models the flow of tension between countries to predict where relations are stressed and violence is likely to occur.

The Model Challenge, which was the last of five sub-challenges within the overarching Tech Challenge for Atrocity Prevention launched in 2012, “successfully drew in new minds and thinking as well as added incentive to develop solutions to challenging problems like atrocity prevention”, according to Mace.

Other strands of the overall challenge asked entrants to develop new ways of collecting social data from hard-to-access locations, among other things.

Daniel Sullivan, director of policy and government relations at NGO United to End Genocide, says: “Algorithms that can help identify mass atrocities can help to focus the attention and resources of groups like ours who seek to ensure early attention and action that can halt or prevent atrocities.”

But Sullivan says further development will be needed to improve algorithms. “There are so many variables involved with different cases of mass atrocities that it is unrealistic to think that a new effort will get everything correct immediately,” he says.

Citing the example of the post-election violence in Kenya that began in 2007 and killed more than 1,000 people, Sullivan says that the US Institute of Peace’s Genocide Prevention Task Force “put together a list of countries at highest risk of mass atrocities based upon variables known to be associated with such atrocities — yet Kenya was not on the list when the violence ensued”.

With all sub-challenges now complete, USAID and Humanity United say they will work with the winners to develop and pilot new tools for atrocity prevention.

Link to the winners of the Model Challenge