Students use AI to aid United Nations in Ukrainian refugee response

Two RIT students are using artificial intelligence (AI) to better support refugees and reshape how humanitarian organizations make data-driven decisions.

Natalie Crowell and Olivia Croteau—who are both third-year humanities, computing, and design majors—are developing an AI tool that analyzes publicly available social media data, specifically from chat groups where refugees discuss needs related to housing, food, and other resources. For humanitarian organizations, this synthesized data can provide invaluable insights, at a low cost.

The students were hired to work on the project as a co-op. The tool that they create could help the United Nations Migration Agency, International Organization for Migration (IOM), improve its response to the ongoing refugee crisis in Poland. According to the United Nations, about 1 million Ukrainian refugees are now in neighboring Poland.

“The goal is to make AI tools as accessible as possible to the people running humanitarian aid,” said Croteau, who is a double major studying new media interactive development.

“We think that having a cheap way to collect and interact with data could be foundational to organizations like this,” added Crowell. “Especially at a time when so much funding is being cut.”

How it started

Crowell and Croteau met in a first-year Python coding class at RIT. They’ve been friends ever since.

While Crowell was taking a course last fall for her geographic information systems immersion, she learned about new research from her professor, Brian Tomaszewski.

Tomaszewski, a professor in RIT’s School of Interactive Games and Media, spent time researching and teaching in Poland as part of a 2023 Fulbright U.S. Scholar Award. He used large-language models to study the situation of forced displacement of Ukrainian refugees and sought feedback from IOM staff to publish a paper.

The professor was looking for students to continue work based on his research. When Crowell told Croteau about this, they brainstormed an idea, wrote the entire proposal for their tool in one morning, and presented it to Tomaszewski.

“Natalie and Olivia are just another example of the amazing students we have here, “said Tomaszewski. “They are directly on the front line of recent upheavals with the foreign assistance given by the United States, which in some sense is frustrating, but also a good learning experience at the beginning of their careers considering humanitarian work.”

How it works

The current prototype extracts discussions by Ukrainian refugees on the popular open-source messaging app Telegram.

Each day, thousands of people will contribute to group discussions about aid. While humanitarian workers do comb through Telegram messages, it can be difficult to read them all.

Crowell and Croteau are designing a tool that uses AI to assign discussions into categories that IOM has already identified. The students are taking a natural language processing (NLP) vector approach to transform sentences into a series of numbers, which then get categorized based on similarity. Some of these categories include mobility, access to education, accommodations to housing, and protection.

The tool includes a fine-tuned NLP model that the students are training to detect locational references. These references are then geocoded with the latitude and longitude. The developers are also implementing AI to analyze photos and videos and translate messages into other languages.

“Humanitarian aid workers typically have to deliver reports, with graphs and visual data,” said Croteau, who is from Ballston Lake, N.Y. “Having interactive data is a great way to get the message across.”

The tool will include a dashboard with AI-created visualizations. Using geocoding, the tool can map data to show where resources are most needed.

Additionally, the students will add confidence scores to the dashboard. It can also generate an AI synopsis of the biggest trends and most common questions of the day.

“What’s great is that this data is coming directly from the people and what the people need,” said Crowell, who is from Pasadena, Calif. “As humanitarian researchers, listing the source of data is essential.”

How it helps

The RIT student researchers are working with Harley Emery, data and research officer with IOM Poland, to fine-tune their tool. Emery shared her personal thoughts on the project, which do not reflect the views of the organization.

Emery explained that humanitarian organizations rely heavily on data to identify the locations and needs of vulnerable migrants and refugees. Without timely, accurate data, it can be difficult to identify target populations for humanitarian interventions, she said.

“While IOM has many of its own methods for gathering and analyzing data on displaced people, I have yet to see a ready-to-use tool that can analyze public social media data on the fly, as this project aims to,” said Emery. “I see a tool like Natalie and Olivia’s being used as a way to rapidly analyze and summarize public discourse online—at low-cost and requiring very little time—to gain initial insights which could then be explored and verified further as needed.”

While the political environment and recent funding cuts pose challenges, the students remain passionate about their work. They see a scalability factor with their project. This low-cost, high-impact tool could be applied to other global crises.

Croteau and Crowell plan to write a paper about their project for the IEEE Global Humanitarian Conference.

“It’s invigorating to use AI to make meaningful impact,” said Croteau. “This is about more than just tech—it’s about being part of the solution.”