A software program called “Annie” uses machine learning to place refugees in cities where they are most likely to be welcomed and find success.
PITTSBURGH—Half a world away from the refugee camp in Uganda where he lived for a dozen years, Baudjo Njabu tells me about his first winter in the United States.
“The biggest challenge is the cold,” he said in Swahili, speaking through an interpreter. We’re sitting on dining chairs in his sparsely furnished living room. Outside, snow covers the grass on the other side of the glass patio doors that lead to the back of the townhouse he is renting in western Pittsburgh. Njabu recounts how his children missed school recently because the bus was delayed and they couldn’t bear the frigid temperatures. His daughter and two sons sit with their mother on a leather couch nearby, half-listening to his replies, distracted by their cellphones and an old Western playing on the television.
All of this has been a major adjustment. Since arriving here in November, Njabu, who is 58 but looks far younger, says he has felt welcomed: Aid workers have helped him rent a place to live, figure out his utility bills, and navigate public transit. His neighbors, fellow refugees among them, are friendly. His children are in school, and he and his wife found jobs in a food-processing facility and at a commercial laundry, respectively, soon after arriving.
About a four-hour drive away, in Silver Spring, Maryland, Karen Monken sits hunched over her laptop, projecting her screen onto a whiteboard. Monken, an associate director at HIAS, a migrant-assistance charity, tells me Njabu and his family were specifically placed in Pittsburgh “because of the high employment probability forecasted by Annie.”
She was referring not to a person, but to a software program. Named for Annie Moore, the Irishwoman who was the first person to pass through Ellis Island, the New York outpost that served as the gateway for millions of immigrants to America, Annie is at the core of an ambitious experiment, one that, were it deployed more widely, could transform how refugees are allocated and treated around the world.
So while Njabu’s decision to settle in Pittsburgh might seem like happenstance, it has less to do with serendipity and more with technology.
For nearly 70 years, the process of interviewing, allocating, and accepting refugees has gone largely unchanged. In 1951, 145 countries came together in Geneva, Switzerland, to sign the Refugee Convention, the pact that defines who is a refugee, what refugees’ rights are, and what legal obligations states have to protect them.
This process was born of the idealism of the postwar years—an attempt to make certain that those fleeing war or persecution could find safety so that horrific moments in history, such as the Holocaust, didn’t recur. The pact may have been far from perfect, but in successive years, it was a lifeline to Afghans, Bosnians, Kurds, and others displaced by conflict.
The world is a much different place now, though. The rise of populism has brought with it a concomitant hostility toward immigrants in general and refugees in particular. Last October, a gunman who had previously posted anti-Semitic messages online against HIAS killed 11 worshippers in a Pittsburgh synagogue. Many of the policy arguments over resettlement have shifted focus from humanitarian relief to security threats and cost. The Trump administration has drastically cut the number of refugees the United States accepts, and large parts of Europe are following suit.
If it works, Annie could change that dynamic. Developed at Worcester Polytechnic Institute in Massachusetts, Lund University in Sweden, and the University of Oxford in Britain, the software uses what’s known as a matching algorithm to allocate refugees with no ties to the United States to their new homes. (Refugees with ties to the United States are resettled in places where they have family or community support; software isn’t involved in the process.)
Annie’s algorithm is based on a machine learning model in which a computer is fed huge piles of data from past placements, so that the program can refine its future recommendations. The system examines a series of variables—physical ailments, age, levels of education and languages spoken, for example—related to each refugee case. In other words, the software uses previous outcomes and current constraints to recommend where a refugee is most likely to succeed. Every city where HIAS has an office or an affiliate is given a score for each refugee. The higher the score, the better the match.
This is a drastic departure from how refugees are typically resettled. Each week, HIAS and the eight other agencies that allocate refugees in the United States make their decisions based largely on local capacity, with limited emphasis on individual characteristics or needs.
“One of the questions we kept asking is, Why are we not making decisions based on what’s best for the refugee?” Mike Mitchell, HIAS’s associate vice president, told me. “Why are we not using data to inform this decision making?”
For instance, if a refugee speaks only a particular language, the agency usually ensures that they are settled in a community where others speak that language. But that typically does not take into account the refugees’ own skills, and requires the people making decisions on placements to juggle discrete bits of information—about the refugee, about the constraints her family might have, about the location where she will be resettled.
The software can hold and sift through thousands of data points and can make decisions almost instantly. Monken told me that what previously took half a day now takes about an hour. This difference is a meaningful one. Less than 5 percent of the 1.2 million refugees who were in need of resettlement last year were actually placed, according to the UN refugee agency. Increasing the capacity of agencies in terms of the number of cases they could handle wouldn’t just make things move faster; it could forever change the lives of a vast number of people around the world.
The idea underpinning Annie and other efforts like it isn’t new. Alvin Roth and Lloyd Shapley were jointly awarded the Nobel Prize in Economics in 2012 for their work, conducted decades apart, looking into what is known as the stable-marriage problem to determine how best to pair couples so both people are happy with their partner. In groups with an equal number of men and women, each person ranks members of the opposite sex in order of preference. The goal is to match couples so neither partner would rather be with anyone else in the group—the so-called stable marriage.
Solving the problem means trying to determine how to match individuals to the best possible outcomes. For example, each year, medical students in the United States are asked to rank the hospitals they’d like to be residents in; an applicant is placed in a hospital she picked if the facility also ranked the applicant highly on its own list. The same logic is also used to pair kidney donors with recipients.
This research has not yet been used on a large scale when it comes to refugee resettlement, and for now, Annie does not take refugees’ preferences into account, instead focusing on employment outcomes, but it could have profound consequences in the United States, and worldwide. Switzerland is testing its own version of Annie, though in its case the algorithm was developed by programmers at Stanford University. The Swiss are studying the outcomes over several years of 2,000 refugees, half placed using the algorithm, half placed without it. Sweden is considering using Annie, too.
The software itself is in its infancy right now. For one, it is lacking in data: HIAS has been using Annie since last summer and has placed about 250 people via the software so far. There’s no exact number on how many refugees Annie must place in order to measure the program’s success. Instead, the software’s efficacy will be measured over several years and through the economic outcomes of the cases that go through the algorithm. Back-testing using data from previous years has yielded promising results, but the real outcomes will take a long time to discover. (Acquiring more data will be its own challenge: The Trump administration’s policy of reducing the number of refugees resettled in the United States means that last year the country accepted fewer refugees, just 22,491, than at any other point since President Jimmy Carter signed the Refugee Act of 1980.)
There is concern that, as Annie and similar tools improve, an algorithm will take over a critical task—placing refugees—that a human is now performing. Officials at HIAS and the programmers who developed the software told me they were aware of those fears. Their solution: Annie will only ever make suggestions; Monken and her colleagues at HIAS make the final decision.
Monken told me that while she accepts three out of four of Annie’s recommendations, she does occasionally overrule the software’s suggestion. In one such case, I met a gay Ugandan refugee in Philadelphia whom Annie had originally placed in another city. Monken knew, however, that Philadelphia had a community of LGBTQ-friendly Ugandans, and assigned him there instead.
His experience might have differed from Njabu’s, but at the heart of these humanitarian endeavors lies a simple calculus: how to efficiently distribute refugees in order to provide the best possible outcome for them, as well as the communities that welcome them and the countries that accept them.
“We’re on the cusp of a real opportunity here to do good using these technologies,” Andrew Trapp, an associate professor at Worcester Polytechnic Institute who heads the team that developed Annie, told me. “These are really challenging problems, and we have a great opportunity to improve society using technology.”
Njabu’s path from the Democratic Republic of Congo to Pittsburgh was as long as it was unlikely.
He can’t return to the DRC, which he and millions of others fled because of a long-running conflict. He took his family to Uganda, which has one of the most progressive refugee policies in the world and which now plays host to hundreds of thousands of people fleeing conflicts in the Great Lakes region in Africa. Life there, Njabu says, was good enough for him to run a barbershop and a business selling vegetables. Two years ago, when it became clear he could not return to the DRC, he applied for refugee status in the United States, and finally last winter, after rigorous vetting and health screening by the U.S. government, Njabu, his wife, and three of their children (he has six more in Uganda, including adults with their own children) found themselves here. He had no idea about the role Annie played in his journey.
They did not know anyone, but have made a new life quickly. His wife works at a commercial laundry, and he got a job at a facility that makes food products such as nutrition bars (a place where, as he described it, “they make something like biscuits, but [with] chocolate”). There are challenges, chief among them his inability, for now, to communicate in English; there is also little time or money for leisure and entertainment.
Still, he told me, his life had improved. I asked Njabu about his hopes for his new life and for his children. He thought for a moment and said he wants stability and peace. “I want them to excel,” he said.
The early stages of the experiment with Annie means it won’t be known for some time what success will look like for Njabu and his family. But it could usher in a new era in refugee resettlement, one in which research and technology play a greater role in placing vulnerable populations in parts of the country, and the world, where they are most likely to succeed.
“It’s not very different from what HIAS did in the very early years in the docks of Ellis Island when you had people actually helping people coming off the docks to get jobs,” Mitchell, the HIAS associate vice president, said. “This is now using data to improve the lives of those new refugees who are coming.”