Machine Learning for Peace: Digital Tools for Civic Action
Details on Core Activities
Background
In December 2021, USAID launched the Machine Learning for Peace (ML4P) website to provide digital tools that help US government agencies and their local partners in civil society and media by monitoring and anticipating major political events and influence from geopolitical competitors that may undermine democratic resilience.
ML4P uses recent advances in big data and machine learning to provide actionable data at an unprecedented scale and frequency across 60 countries. MLP builds continuously on a massive repository of more than 100 million news articles capturing more than 12 years of daily coverage from more than 350 online newspapers publishing in more than 35 languages. Using the best available methods and intensive quality control, MLP scrapes the full publication history of a curated sample of high-quality local sources for each country.
Applying cutting edge tools from machine learning, ML4P then identifies the events and locations being reported in each article and uses predictive analytics to detect historical patterns and forecast likely changes in political conditions and foreign influence up to six months into the future. ML4P’s digital tools provide access to monthly data tracking 20 events bearing on civic space, including censorship, corruption, and legal changes, and 22 events indicative of foreign influence by geopolitical competitors, ranging from security engagements to domestic interference. Importantly, these data are updated every 90 days, ensuring their timeliness for studying current events in addition to longer-term patterns and trends.
Digital Tools
ML4P data and analytics are made publicly available through five interactive data dashboards (Civic Space and Foreign Influence) designed to satisfy two needs common across local and international civic actors, policymakers and donors intent on promoting democratic resilience:
First, the interactive data dashboard provides users with the ability to document recent events and trends with hard data, allowing for more effective communication about local conditions and priorities to donors, policymakers, and the public. This includes data on events for which other data sources do not exist, including things like censorship, corruption, and legal changes by national governments in aid-receiving countries and events like diplomatic activity and policy influence between the US’s geopolitical competitors and aid-receiving countries.
Second, the forecasting dashboards provide advanced warning when major events are likely to occur over the next six months, intended to increase the capacity for strategic planning across US government agencies and their local partners. Recently, ML4P expanded forecasting efforts to include influence by the US’s geopolitical competitors on aid-receiving countries.
Between June 1, 2023 and March 1, 2024, the ML4P website received more than 12,000 page views from 3,100 unique visitors spread across 116 different countries, our data dashboards average 155 hours of active usage time per month, and the project has received more than 500 sign-ups to our monthly mailing list.
High-Level Reporting
Our Civic Space Early Warning System (pictured below) summarizes the results from hundreds of forecasting models into one simple visual representation of what our models are predicting across countries. By visiting the interactive data dashboard, you can learn about the types of events we can forecast accurately for each country and whether our models are predicting important increases in activity around any of those events.
For each country, we provide brief quarterly updates that summarize recent events detected by our data, analyze the accuracy of our previous forecasts, and discuss any major civic events our models are predicting for the next six months. See a recent update on South Africa below for a recent example.
To assess the overall track record of our predictions, we reviewed all predictions since December 2022. For each prediction, we assessed whether the data from future updates aligned with the prediction that we reported several months prior. More than 70% of our models’ predictions successfully provided advanced warning of a major event that transpired over the following months.
Importantly, ML4P recently expanded forecasting efforts to include influence by the US’s geopolitical competitors on aid-receiving countries. In the near future, these forecasts will be integrated into the CSEWS dashboard and geopolitical events detected in our data will be described in our quartly update reports for each country.
Core Activities
ML4P is currently seeking core support in order to continue its basic operations. These tasks include:
Data Collection: ML4P will continue to update the current roster of 60 countries on a quarterly to capture the most recent civic space and foreign influence events. New countries are added at a pace of one per quarter and are selected in consultation with USAID/DRG. ML4P regularly explores opportunities to share this underlying data with other researchers and organizations, integrate new technologies into our system, and increase the utility of the data to track and forecast new types of events.
Dashboard Maintenance: ML4P will continue to maintain the MLP dashboards with up-to-date data. This includes ongoing efforts to improve the dashboards to increase their utility for our target audience.
Report Production: ML4P will continue to produce Quarterly Forecasting Reports for each country in the ML4P roster. The reports include short descriptions of: a) major events detected, along with summaries of key dynamics in each country within the database since the last update; b) the performance of the previous quarter’s forecast for each country relative to what occurred over the last quarter; c) a description of any significant increases in activity across our civic space indicators forecasted for the coming quarter.
Reporting on Foreign Influence: In addition to these core activities, ML4P is planning to expand our Quarterly Forecasting Reports by integrating event detection and forecasting for Chinese and Russian influence events. Currently, our monthly reports focus on the civic space events. For each country, future reports will include a summary of any major foreign influence events detected over the last quarter and an assessment of whether our models are predicting an increase in foreign influence over the next six months.
In addition to these core activities, MLP is eager to provide Mission-specific analytic support through buy-ins from interested Missions. This support could include in-depth briefings and analysis, expansion of the analysis into sub-regional areas, augmentation of data processing to detect and forecast new events, such as election violence or local environmental conflicts, and policy-driven research on political events and foreign influence, including the use of ML4P data to evaluate the impact of USAID programming. MLP has worked with a variety of USAID missions to leverage MLP data for commissioned research projects. Recent projects include measuring the impact of legislation restricting freedom of the press on the publication practices of media outlets and legislation restricting free speech on citizen’s online behavior in Tanzania, the use of captured media to deflect coverage of corruption cases in Bukele’s El Salvador, the REMEDIOS evaluation, which seeks to measure the impact of USAID programming on anti-corruption reporting in Central America, and ATARI, which produced new insights into how America’s geopolitical competitors have adapted their approaches to foreign influence over the last 12 years.
Contact
USAID Missions interested in learning more about MLP or the potential for an MLP buy-in through SEGA can contact the research team directly or the COR for SEGA.
- MLP Research Team: mlpeace-devlab@sas.upenn.edu
