Artificial intelligence as a crisis helper: software that predicts refugee movements

Icon: enlarge

Agadez in Niger: Migrants prepare to cross the desert

What would cost even an expert hours, "Foresight" delivers in seconds. The forecasting software of the Danish Refugee Council (DRC) evaluates information from more than 120 data sources, analyzes patterns and then calculates, for example, how the corona crisis can affect displacement and displacement.

Accordingly, as a result of the pandemic in the countries of Burkina Faso, Niger, Mali and Nigeria, a total of around one million people could be forced to leave their homes - the misery in the conflict-ridden region is therefore forecast to worsen again.

Software such as "Foresight" should make decision-making processes in humanitarian aid and corresponding operations more efficient in the future. "The tool helps us to predict what will happen more quickly, so that we can plan better and intervene earlier in a humanitarian crisis," says Charlotte Slente, Secretary General of the DRC. Fast reactions are also often cheaper.

Icon: enlarge

Displaced women in Mali prepare food. The corona crisis will exacerbate the plight of many people

Experts from the refugee aid organization have already tried in the past to make predictions about flight and displacement based on data and empirical values.

The algorithm should now help to predict even more precisely when a scenario will occur and to better analyze the effects of various factors such as the economic situation, conflict, climate or governance. Slente calls this the "DNA of displacement".

Icon: enlarge

"Foresight" can simulate future scenarios - and helps experts from the humanitarian field with planning

The algorithm is not a silver bullet, however: "The software doesn't know what will happen tomorrow or next week," says Slente. "The forecasts refer to the next one to three years - and the model cannot say anything about whether people will flee within the country or whether they will decide to cross national borders."

First pilot projects

The Danish Refugee Council wants to make its forecasting software available to other organizations via a platform in the future. There is great interest in such tools in the humanitarian field, but the use of data analysis and forecasting software is still in the experimental phase.

The data expert Miguel Luengo Oroz attributes the fact that it took longer for the trend to take hold in crisis management and development aid due to a lack of technical knowledge, but also to limited capacities in humanitarian organizations. "Development experts and data experts usually do not speak the same language, they do not have a common vocabulary, come from different contexts and often cannot coordinate their goals," wrote Luengo Oroz, Chief Data Scientist of "UN Global Pulse"in an essay.

"UN Global Pulse" is one of the United Nations' initiatives, which is why it is doing pioneering work at the interface between development aid and technology: Among other things, it researches the potential of big data and artificial intelligence and supports other UN agencies in implementing projects. During the pandemic, governments around the world are particularly interested in data-based models that can estimate the spread of Covid 19 infections and the effect of planned strategies and possible side effects.

Weather, tweets and goat prices

The innovation team of the UN refugee agency UNHCR has already developed various experimental approaches to collect information on future migration movements.

For example, data analysts used open source data to determine future weather conditions for boat trips across the Mediterranean and observed Facebook posts by smugglers in order to obtain information on price developments, frequently used routes and the crowds at junctions. They also used algorithms to analyze social networks such as Twitter for keywords and negative expressions of emotions in order to record the moods and narratives of migrants.

Sometimes rather unusual-looking data such as rising prices for water cans and falling goat prices point to an impending movement, as in Somalia: "When people started getting rid of 50 or 60 goats at a time, prices were depressed and we had an indication that that people are planning to flee, "said Andrew Harper from the UN refugee agency UNHCR.

Icon: enlarge

Sometimes goats can also be indicators of intent to flee

The backgrounds for flight and displacement are diverse, which also makes the development of forecasting software a challenge. "It is a complex interplay of many different factors, and you cannot explain the phenomenon by focusing on just one aspect," says Alexander Kjærum, the tech expert at the Danish Refugee Council (DRC) who co-developed the Foresight software Has. He was surprised by the extent to which, in addition to conflicts, governance and human rights violations have an impact on displacement.

When the pilot project started two and a half years ago, it was unclear whether the software would be able to achieve meaningful results - the development was a constant process of "trial and error". Today Kjærum is confident: "We are still in the development phase and are constantly working to refine and improve the software, but the results so far are encouraging and show a high level of accuracy," he says.

Low error rate

To test "Foresight", the team fed the software with data on Myanmar and Afghanistan and had them simulate scenarios for periods from the past. The forecasts were then compared with reality. According to Kjærum, the algorithm made very precise predictions with an error rate of eight to ten percent.

Icon: enlarge

Sudden events such as the mass displacement of the Rohingya from Myanmar pose a challenge to the algorithm

The self-learning algorithm analyzes historical data from the past 20 to 25 years for its predictions in order to identify patterns and to draw conclusions for the future from them: the more complete statistics from the past are available, the more precise the forecasts. The software primarily accesses national data from organizations such as the United Nations, the World Bank and governments - on conflicts and human rights violations, economic development or socio-economic factors. The disadvantage of this approach: Since the software focuses on national rather than local data, causes such as locally limited conflicts are less difficult to evaluate because the data density is lower.

Fifteen key indicators are particularly relevant for the algorithm to calculate the risk of forced displacement - including unemployment, corruption, public services, food security, political murders, civilian victims in conflicts, human rights violations, natural disasters and social inequality. The software not only calculates the number of displaced persons, users can also explore the relationship between the various indicators in a causality model.

Icon: enlarge

The software reveals relationships between a large number of factors

The algorithm can also simulate various scenarios. If experts want to estimate the effects of the corona crisis, for example, they can simulate what would happen if the economy in a country collapsed by a few percentage points, the ability to govern falls or the intensity of the conflict increases. The software then creates new risk forecasts for forced evictions.

"But it is difficult for the software to predict very sudden or unprecedented events, such as the Rohingya conflict in 2017 and the mass displacement," says Kjærum. Myanmar soldiers murdered thousands of people, destroyed villages and raped women and children of the Muslim minority in a brutal military offensive. More than 700,000 people had to flee.

Further country simulations are to be made available in the coming months. In the future, the code should also be available online for other users so that organizations can adapt it to their needs and add further data.

"Forecasts are not a precise science and should not be the only basis for making decisions"

Charlotte Slente, Secretary General of the Danish Refugee Council

Charlotte Slente, Secretary General of the Danish Refugee Council (DRC) is also aware of the software's limits: "Forecasts are not a precise science and should not be the only basis for decision-making," she says. The experiences and assessments of crisis workers and development experts will not be replaced by an algorithm in the future either.

This contribution is part of the Global Society project

What is the Global Society project? Up arrow Down arrow

Under the heading Global Society, reporters from Asia, Africa, Latin America and Europe report on injustices in a globalized world, socio-political challenges and sustainable development. The reports, analyzes, photo series, videos and podcasts appear in the international section of SPIEGEL. The project is long-term and will be supported by the Bill & Melinda Gates Foundation (BMGF) for three years.

A detailed FAQ with questions and answers about the project can be found here.

What does the funding look like in concrete terms? Up arrow Down arrow

The Bill & Melinda Gates Foundation (BMGF) is supporting the project for three years with a total of around 2.3 million euros.

Is the journalistic content independent of the foundation? Up arrow Down arrow

Yes. The editorial content is created without any influence from the Gates Foundation.

Do other media have similar projects? Up arrow Down arrow

Yes. Big European media like "The Guardian" and "El País" have set up similar sections on their news sites with "Global Development" and "Planeta Futuro" with the support of the Gates Foundation.

Have there already been similar projects at SPIEGEL? Up arrow Down arrow

In the past few years, SPIEGEL has already implemented two projects with the European Journalism Center (EJC) and the support of the Bill & Melinda Gates Foundation: The "Expedition The Day After Tomorrow" on global sustainability goals and the journalistic refugee project "The New Arrivals", as part of this several award-winning multimedia reports on the topics of migration and flight have been produced.

Where can I find all publications on Global Society? Up arrow Down arrow

The pieces can be found at SPIEGEL on the topic Global Society.