How to predict conflict before it’s too late
This article was initially published on The Diplomatic Pouch, a blog from Georgetown University’s Institute for the Study of Diplomacy
The rise of artificial intelligence (AI) has proven both promising and perilous. AI is already commonly used in voice-powered personal assistants in our homes, in facial recognition software for our phones, and in recommender algorithms for platforms like Netflix that provide watch suggestions. Increasingly, business and private sector initiatives are using AI, and it is even making its way into contracts for government services. At the same time, AI is also being used to advance military technology and to surveil and identify ethnic minority populations, sounding alarm bells for ethicists and human rights advocates. These recent advancements in AI, while impressive, are only be the beginning of future applications. Conflict prediction and humanitarian responses are two areas where we should apply AI for good.
Currently, humanitarian assistance often arrives well after a developing crisis turns into a full-fledged catastrophe. Aid workers are only dispatched once events have reached a certain threshold, and resources are distributed based on need and capacity. Artificial intelligence already plays some role in aid and response; during natural disasters, responders may use AI to examine satellite imagery and drones to geotag areas of relief, collect data, and estimate funding requirements. Relief agencies use AI to increase the efficiency of procurement processes and to help responders on the ground communicate with beneficiaries as needed. But what if data-driven AI could be used to predict humanitarian crises before they even began?
Institutions like Uppsala University, the Simon-Skjodt Center for the Prevention of Genocide, the Peace Research Institute Oslo, Dartmouth College, and the Armed Conflict Location & Event Data Project (ACLED) have long worked on conflict early warning. Many of these projects have also created successful models that can predict conflict trends within the parameters of their data collection techniques.
These models identify independent and dependent variables as precedents of modern conflict, which then define the data they collect; experts in conflict analysis have identified many of these variables as economic factors (inflation, sudden GDP shifts, high economic inequality, corruption), political indicators (populist or nationalist sentiment, presence of armed groups, freedom of press, military personnel), and social indicators (high poverty rates, youth unemployment, ethnic tension), among others. Accuracy from these traditional models can range from 50 to 75 percent.
However, AI can be harnessed to push accuracy even further. Today, the vast majority of recent AI advancements and applications refer to a category of algorithms known as machine learning. In some categories of machine learning, algorithms can classify objects, make predictions, or discern patterns in data. In the scope of an AI Early Warning System (EWS), the machine learning system could make accurate forecasts based upon available datasets that are the most relevant for each region of interest. These datasets would use the three types of indicators — economic, political, and social — and could be tailored more specifically to each country or region of the world.
AI models are also self-learning. This means that the system would be able to recognize patterns within conflict indicator datasets that would allow it to increase its accuracy in every calculation. By understanding the relationships between enormous catalogs of data, the AI EWS could function better than any other traditional model has done before. Additionally, given that some of the biggest indicators of conflict have now shifted to the point where they can be tracked on social media, AI sentiment analysis on real-time social media and natural language is one of the most important pieces of conflict prediction that does not exist in a comprehensive EWS today. Besides data collection and analysis, AI systems can also work faster than humans without fatigue and process vastly more amounts of data when forecasting, predicting, and providing analysis. These characteristics could allow an AI EWS to make predictions faster than any existing conflict forecast model without sacrificing accuracy or time. To be successful in this scope, both the right indicators and a carefully trained ML system would be needed to accurately forecast violence.
In many conflict-affected regions, the use of an AI EWS might have drawn attention to these regions before the consequences were too dire. The Arab Spring, the Venezuela crisis, the Colombian civil war, the Rwandan Genocide, and violence in the Sahel are all areas that could have benefited from early intervention relevant to the crisis at hand. It is here that an AI system could also shine. Early warning on the most prevalent indicators of impending instability, could enable intervention and mediation techniques to mitigate the larger issues while focusing on long-term peacemaking. If corruption is the main cause of tension in an area, anti-corruption techniques could be used to lessen the possibility of conflict; the same could be applied to youth unemployment or democratization.
However, there are times where we have seen hesitation from policymakers to take action and invest personnel and funding, especially if the consequences of inaction are ambiguous or uncertain. To incentivize action, a secondary model could also be created to show the consequences of non-intervention; if a model knows the demographics of a country likely to erupt in conflict, it could also demonstrate the consequences if 50–75 percent of that population migrated to bordering countries or were internally displaced, information that may have proven valuable when gauging the decision to intervene in the Syrian Civil War. However, the availability of data here is key. Without accurate, steady data streams, the AI EWS would not be able to consistently do its work.
In practice, this kind of AI conflict early warning system could be a tool for the U.S. Department of State and other diplomatic players to predict migration flows before they occur, identify regional conflicts as they begin to brew, and pursue better-informed diplomacy with both allies and counterparts around the world. If this tool is structured using open-source data and is made available to all parties, then NGOs, corporations, and foreign governments could also use the tool for their own interests in conflict analysis. Such a model could also save significant sums in foreign aid and intervention as conflicts are dealt with before they rise to an unmanageable scale.
We already see some success in integrating AI into conflict early warning as demonstrated by the private company Peloria, which is the only institution that incorporates AI into its predictive work, claiming 84 percent accuracy. As the nature of conflict continues to evolve, we must get better at identifying and mitigating conflict early and effectively. Incorporating artificial intelligence is the next step in that process.
Ishanee Chanda graduated in May 2021 with a Master of Science in Foreign Service degree from Georgetown University, with a focus on refugee and humanitarian emergencies. She is particularly interested in refugee and migration issues, resettlement practices, the protection of human rights, and the rise of right-wing nationalism across the globe.