Millions of people are forcibly displaced annually due to conflict, persecution, or natural disasters, and they seek asylum in neighboring countries. The refugee crisis triggered by the war in Ukraine is the latest high-profile example, but there is a trend of global events not restricted to a particular country or continent that can happen for many different reasons. These global events exacerbate the humanitarian burden placed on countries giving asylum.
Most asylum-seekers and refugees don’t pose a threat to public safety, but when they arrive in vast numbers they can put considerable strain on border security processing and longer-term refugee and immigration management and tracking. Small numbers of criminals and terrorists could hide among this population and must be effectively identified and isolated for public safety – in some cases, with urgency. When there is no data or only partial data available, this makes the challenge of making decisions acute.
Border authorities need to act quickly and effectively to decide who should be allowed entry, and how and where asylum-seekers and refugees should be routed for temporary sanctuary or longer-term resettlement. Above all, fairness and accountability must be consistently maintained at all levels of the process – for the destitute, the countries that accept them, and international governing bodies tasked with maintaining humanitarian standards and equitable responsibility-sharing among countries.
Data-driven decisions take some of the guesswork and bias out of immigration management processes and help to promote overall fairness. Machine learning (ML) and data fusion technologies form the bedrock of this approach, with their combined ability to leverage big data and advanced analytics to manage the risk posed by potential asylum-seekers and refugees. This includes closer scrutiny of criminal and terrorist elements, with more efficient threat assessment and risk scoring – leading to improved prioritization.
However, there are foundational data challenges inherent to immigration management processes. The processing of asylum-seekers’ and refugees’ information and applications is typically done manually, and relies heavily on their own claims. Many arrive at the border with no credentials.
Compounding this issue, poor data quality used for training AI algorithms can produce equally poor outcomes. A simple example: An individual may cross the same border several times, but incomplete record-taking could result in the person being recognized in databases as different people.
Verbal statements given at the border – under great stress, in some cases – can leave a lot of data gaps. This in itself is a big challenge for border and immigration analysts – one that typically requires additional manual investigation.
Faulty face recognition is another potential impediment to effective and efficient immigration processing. Research has shown that the accuracy rates of facial recognition algorithms are particularly low in the case of minorities, women, and children. A 2018 MIT study of three commercial facial-recognition systems found they had error rates of up to 34 percent for dark-skinned women – nearly 49 times that for white men. When it comes to identity verification, a systemic failure to accurately distinguish faces in this way presents obvious problems.
Lastly, asylum-seeker and refugee data privacy must be maintained according to strict guidelines where possible. For vulnerable populations, the oversharing of sensitive data can introduce several risks that can further magnify power asymmetries with immigration officials. ML algorithms should ultimately aid in decision intelligence for migration analysts seeking the most informed, efficient ways to assess risk while prioritizing human rights.
ML AND DATA FUSION BENEFITS
At the solution level, these constraints pose many technical challenges, with broader implications for algorithmic accountability and fairness. Effective border security and immigration management requires a decision intelligence framework that is both fair to asylum-seekers and refugees, and provides a rapid, informed assessment of their threat risk for countries of first asylum/host countries. ML and data fusion play important roles here.
ML can add speed and agility to the workflow, beginning at the border processing stage – it helps ensure that well-intentioned asylum-seekers and refugees aren’t held in limbo. This can be achieved with faster ML-guided identity checks and data analysis for asylum and visa applications.
Post border-processing, ML can aid in decision intelligence affecting how and where asylum-seekers and refugees are routed for sanctuary and/or resettlement. Improved tracking is an important part of the equation, for the purpose of quickly locating and notifying asylum-seekers and refugees on occasions when their origin country’s status has changed, and it’s safe/legal to return.
ML and data fusion are also invaluable for risk assessment and risk scoring, to help identify asylum-seekers and refugees who pose potential security risks. The ability to harness and correlate numerous, disparate data sources ultimately provides a fuller, multidimensional understanding of threats and bad actors while helping to eliminate ‘blind spots.’
ML has already demonstrated effectiveness in immigration management applications. For example, the German Federal Office for Migration and Refugees has tested automatic face and dialect recognition, name transliteration and analysis of mobile data devices for identity verification to support decision intelligence in the asylum determination process. Meanwhile, UK police forces use predictive policing algorithms such as PredPol, Sweden has used migration algorithms to forecast future migration flows, and Switzerland has tested an algorithm to improve refugee integration (source).
Improved decision intelligence for border security and immigration management in turn helps streamline asylum-seeker and refugee processing and can help with the enforcement of essential humanitarian protections. ML and data fusion enable government and border control entities to accelerate border processing workflows and make better, faster decisions regarding security threats.
Over the longer term, ML and data fusion technologies can be used to help better situate asylum-seekers and refugees according to their needs and ability to contribute. They can also predict waves of migration and help countries collaborate and prepare for them. Lastly, they can improve data sharing among countries, in part to democratize this process globally and to improve fairness and accountability across the board.
The views expressed here are the writer’s and are not necessarily endorsed by Homeland Security Today, which welcomes a broad range of viewpoints in support of securing our homeland. To submit a piece for consideration, email [email protected].