Machine Learning for Peace: website forecasts civic spaces

Machine Learning for Peace graphic 1

Civic spaces can serve as a barometer of political health.  Policymakers and researchers need accurate analysis of the civic spaces around the globe in order to know what locations need help. Devlab@Duke has launched a new interactive online tool to address shrinking civic space and growing authoritarianism around the world. This tool will put incredibly fine-grained data and highly accurate forecasts of civic space events into the hands of policymakers, practitioners, and researchers working to defend democracy.

This event debuted The Machine Learning for Peace (MLP) project, which combines recent advances in natural language processing, recurrent massive web scraping of more than 100 international, regional, and local sources of online news, high-frequency economic data, and machine learning forecasting models to produce cutting-edge research that can guide policy and practice. Each month, we scrape thousands of new articles from these sources to add to this enormous corpus of digital news. Each article is classified according to 19 events that capture changing civic space events and 22 events that capture the efforts of authoritarian regimes to influence developing countries. To date, MLP has scraped and processed more than 68 million news articles across more than 20 languages and 30 countries.

In order to predict future trends in civic space activity, DevLab’s models identify the patterns in our historical data that were associated with past changes. The researchers then combine knowledge of these historical patterns with data on what is happening right now to predict what is likely to happen in future months. In producing these forecasts, DevLab utilizes models that prioritize interpretability so that users can identify which movements in the data are influencing the output of each forecasting model.

With civic space shifting rapidly around the world, faster and more potent technological interventions, like this one, will be critical for bolstering local civil societies. The MLP project website hosts detailed descriptions of this unprecedented data repository, the methodologies and procedures used to maintain the dataset and produce forecasts, easily digestible visualizations of data and forecasts, and research products associated with the project. In January 2021, MLP will also launch a quarterly online survey of local civil society experts to further incorporate insights from local voices into our analyses.

MLP is supported by USAID’s Center for Democracy, Human Rights, and Governance and the Enabling and Protecting Civic Space (EPCS) Illuminating New Solutions and Programmatic Innovations for Resilient Spaces (INSPIRES), a seven-member consortium led by Internews. Through this activity and many others, INSPIRES is increasing the knowledge and capacity of citizens, organizations, the media, and donors, to quickly respond to growing restrictions on democratic freedoms of association, assembly, and expression. The launch event took place on December 15, 2021 as part of the Open Government Partnership Global Summit and included representatives from the United States Agency for International Development and the International Center for Not-for-Profit Law.