Machine Learning as an Emerging Social Justice Tool

"How we talk about refugees matters."

by Wendy Norris, Ph.D.

illustration of a digitized human face

My jaw dropped when I read the story:

“Since the start of this crisis we have witnessed a hyperinflation in the language around refugees: they are invariably ‘waves’ or ‘floods’, and they are ‘streaming’ into Europe. [UK Prime Minister] David Cameron went so far as to describe these people as a ‘swarm’, as if they were dangerous insects.”

I cried. My mind reeled. I couldn’t make sense of how anyone could use such dehumanizing language to portray desperate people fleeing their war-torn countries. Worse yet, the article appeared in the Guardian just three weeks after the world was horrified by a viral photo of the tiny, lifeless body of Alan Kurdi, a 2-year-old boy washed ashore on a Turkish beach after the inflatable boat carrying his family and other Syrian War refugees capsized in the Mediterranean Sea.

How we talk about refugees matters. As political discourse to effect social change. As legal precedent to right wrongs. And, more importantly, as an indelible marker of our own humanity toward the most vulnerable among us.

When I launched research exploring the refugee crisis in Europe in 2015, global attention was turned to the humanitarian catastrophes arising from violent conflicts in Syria, Iraq, Afghanistan, and across the Sahel region of West-Central Africa. Seven years later, a new large-scale refugee crisis caused by the Russian invasion of Ukraine makes this study ever more urgent.

One of the most challenging, pervasive, and heartbreaking social justice problems in the world is the plight of refugees. Today, of the 84 million people worldwide who have been displaced from their homes, one in three is considered a refugee, an international legal designation for the 27 million people who have left their countries and cannot return because of a well-founded fear of persecution.

The causes of their displacement are complex and include a wide range of social and political reasons — war, armed ethnic conflicts, human rights abuses, or violence based on their religious beliefs, nationality, orientation, and/or membership in particular social groups.

Reporting news stories about refugees is extraordinarily challenging because of the variety of causes of the exodus and the trauma many people have suffered in their own country and during their escape. I should know. I am a former political investigative reporter and editor who covered social concerns for more than 10 years. Metaphors are a common journalistic technique to help break down complicated issues and to help “paint a picture” that advances the storytelling about a topic unfamiliar to the reader.

But metaphors can shape understanding. In a preliminary study of hundreds of news articles, an undergraduate research assistant and I discovered that refugees were frequently dehumanized — not just in the heat of political debate but also in everyday depictions in mainstream newspapers across the world. Using natural disaster metaphors such as flood, avalanche, and swarm to describe the scope and scale of an unfathomable humanitarian crisis risks shaping public opinion about the refugees themselves as uncontrollable torrents, destructive events, and terrifying infestations. Our pilot study revealed that this emotionally laden language may communicate unintended media messages that can have a profound effect on public support for basic human rights, refugee protection, and resettlement efforts.

Wondering if there are patterns in what kinds of news organizations use these phrases — and when, where, and how — prompted us to turn to advanced computer science techniques as an emerging social justice tool.

In our research as part of Nazareth’s new Institute for Technology, AI and Society, we will use machine learning techniques to find relationships in massive datasets that are too large to study through traditional humanities-based research methods. Applying computational models to detect subtle patterns in the data can help us better identify and document hidden biases in depictions of refugees.

Since our initial study of a few hundred stories, we have collected more than 12,000 full-text mainstream news articles and 7.5 million Twitter posts about refugees published in 2015 during a 12-week period punctuated by the tragic death of Alan Kurdi and the suffering of too many others during the height of the exodus to Europe. Next, undergraduate student researchers will update the study with new data from the unfolding crisis in Ukraine. Using machine learning in tandem with other qualitative and critical approaches for understanding media and language, puts Nazareth at the forefront of exciting interdisciplinary collaborations between the humanities and computer science.

Wendy Norris is an assistant professor of computer science in the Mathematics Department.

Ethics Meets AI at Nazareth

Learn more about undergraduate majors and minors in the College’s interdisciplinary Institute for Technology, Artificial Intelligence, and Society (ITAS):

Ethics Meets AI at Nazareth ≫