您现在的位置: VOA >> VOA双语新闻 >> 双语新闻 >> 正文


更新时间:2018/2/4 11:53:22 来源:本站原创 作者:佚名

The algorithm that could help tackle the refugee crisis

There are a total of 65.6 million forcibly displaced people in the world today, fleeing conflict, persecution and corruption, according to the United Nations Refugee Agency – an all-time record.

根据联合国难民署(United Nations Refugee Agency)的数据,全世界目前因为冲突、迫害和腐败等各种原因而背井离乡的难民总数约为6,560万人,创下历史最高纪录。

In the countries where many displaced people dream of starting new, successful lives such as Australia, the US, the UK and Germany, refugee resettlement is a complicated, contentious issue.


Political pushback aside, there are huge obstacles that jam the system. Many nations currently place incoming refugees haphazardly, dispatching them to disparate regions based purely on whether these areas have enough space to accommodate extra people. But there’s no guarantee that these areas will provide enough jobs for the new arrivals – and being unemployed can be a huge barrier for a refugee to find jobs and afford to live in their new homeland.


But a team of researchers at Stanford University, ETH Zurich and Dartmouth College has come up with a system they believe can vastly improve the job prospects of newly settled refugees.

但斯坦福大学(Stanford University)的一个研究团队开发了一套系统,他们认为可以帮助新安置的难民极大地改善就业前景。

Outlined in a paper published in the journal Science today, the team have created a data-driven algorithm that learns how to allocate displaced people to where they are much more likely to find jobs. It hasn’t been tested in the real world yet, but the researchers believe it could boost the likelihood of employment for each family by up to 70%.


Currently, a bureaucrat will use a spreadsheet to assign families to locations based on capacity constraints, says Jens Hainmueller, one of the researchers at Stanford’s Immigration Policy Lab. “There’s a bed in Minnesota, you go to Minnesota. There’s no purposeful matching.”

斯坦福移民政策实验室研究员延斯·海恩穆勒(Jens Hainmueller)表示,政府目前根据容量限制,用电子表格来决定难民家庭前往何处。"明尼苏达(Minnesota)有一张床,你就去明尼苏达。但并没有形成有目的的匹配。"

If a resettlement agency could analyse an immigrant’s demographic profile and send them to a town, city or region where they’d be more likely to find a job, they would be a better chance to succeed.The team analysed figures from two developed countries: The US (using data from more than 30,000 refugees aged 18-64, who arrived between 2011 to 2016) and Switzerland (more than 20,000 refugees from 1999 to 2013). The algorithm was built on the likelihood that individual refugees found employment in their host country.


First, the team looked at the refugees’ demographic data: education, age, gender and English fluency. From there, they looked for “synergies” between these characteristics and regions with high employment rates for people with those specific characteristics.


Then they found trends: if, for example, certain African refugees spoke French, they’d obviously find work more easily in French-speaking Swiss cantons (regions) than in German-speaking ones.


And voila – using the algorithm, a resettlement agency could analyse an immigrant’s demographic profile and use available data to place them where they’d be a great deal more likely to succeed.


“If there is a meatpacking plant that employs young male refugees, and there’s a demand for that, that algorithm would pick that up,” says Hainmueller.


A simple way of thinking about it, the researchers say, is to use the example of two young Afghani men with the same education level and age who are sent to two different locations in their new country. One finds work in place A, the other doesn’t in place B. The team’s machine-learning algorithm learns from that, and next time, if a third person arrives with a similar background, the programme will then know to send him to place A if possible.


To be sure, every situation and each individual is different. The team acknowledges that a human official might sometimes have to override a placement match. In that way, like a lot of AI, it complements humans rather than replace them.


“The machine learning techniques we are using are extremely flexible,” says Kirk Bansak, another member of the team. “They are able to discover and find [patterns] in really complex data.”

"我们正在使用的机器学习技术非常灵活。"该团队另外一名成员科克·班萨克(Kirk Bansak)说,"他们可以在十分复杂的数据中寻找和发现各种模式。"

For example, had the algorithm been implemented in the US between 2011 and 2016, the researchers believe the average employment rate could have risen from 34 to 48% (a 41% increase). In Switzerland, it might have been boosted from 15% employed to 26%.


“What we see is that refugees are much more likely to find work, they learn the language more quickly, they integrate more quickly, and are also not going to end up taking a lot of resources in terms of health benefits. They are economically integrated, pay taxes and make contributions to society,” Hainmueller says.


Of course, more research is needed, but the team is working with governments and organisations to set up pilot programmes to test the algorithm’s effectiveness in the real world. Eventually, they hope for places like the US and Switzerland to use the algorithm (the code for which is available to organisations for free, the university says) when matching refugees to their new homes. The Swiss government has publicly expressed interest, the team says, and they're also in talks two resettlement agencies in the US to implement the system.


If implemented, the Stanford researchers hope their algorithm can strengthen the workforce and revitalise a local economy – it could potentially help nations handle a thorny political issue.


“We have the historical data anyway,” Hainmueller says. “We may as well learn from it.”