The grant will be used to research human-AI teaming through social media in emergency response during the COVID-19 pandemic.
BYU IT and Cybersecurity assistant professor Amanda Hughes has been awarded a competitive $24,498 National Science Foundation (NSF) RAPID grant. Her project, titled “Human-AI Teaming for Big Data Analytics to Enhance Response to the COVID-19 Pandemic,” is in collaboration with the University of Texas at Austin and George Mason University .
The goal of the project is to better understand the process of real-time decision making by human volunteers while they rapidly convert social media data into structured codes that machines can understand. That data is used to help emergency responders in disaster relief. The use of human analysis and input combined with AI technology is what is referred to as “human-AI teaming”. Researchers will also use the knowledge gained in the study to improve human-AI teaming to aid in advancing disaster relief and emergency response.
Hughes is the principal investigator for BYU’s portion of the project. Her team is primarily responsible for increasing understanding of human decision making and coding that information for the AI systems.
“We recognize that in order to make these AI systems more effective, we need to have good data that goes into those systems,” said Hughes. In order to understand that process they are working with volunteers from the Montgomery County Community Emergency Response Team (CERT) in Maryland and Steve Peterson, community coordinator for the project.
Her team has conducted nearly 60 interviews so far for the project. Interviews are split into two rounds. The first round consists of a participant sharing their screen via Zoom for one hour and assigning values to individual social media posts while being observed by two interviewers. Value designations included risk, prevention, positive sentiment, negative sentiment and irrelevant. Value options were chosen by the team’s emergency responder contact who deemed them as information that would be helpful to emergency responders.
The second round of interviews are conducted in a similar format to the first, with a slightly different task. Volunteers are shown tweets that have already been assigned values from an AI system that part of the team has been working on. Volunteers are then asked to verify the label on the tweet or correct it if they feel it is incorrect. Collaborators at the University of Texas at Austin, led by professor Keri Stephens, are also working to conduct and analyze these interviews.
Information found throughout the interview and analysis phases will be used to improve the “Citizen Helper” software, developed by Hemant Purohit, collaborator and assistant professor at George Mason University. Citizen Helper is a program used to translate messages from social media into effective and actionable data that assists emergency responders.
During crisis events like the COVID-19 pandemic, emergency responders and decision-makers need timely and accurate information to address the uncertain and quickly evolving conditions. Social media can provide this type of information, but actionable information to support decision making—such as data about the exposure of COVID-19 and misinformation—is buried in the large volume of available social media data. This project leverages the strengths of humans and AI to design systems like Citizen Helper that can filter relevant information to assist decision-makers and emergency responders.
While the immediate goals of the project are related to COVID-19, the implications could be much larger, expanding to hurricane relief, earthquake relief, fire relief and more. With advancements in understanding and further development of the Citizen Helper software, the group hopes to make this technology accessible, user-friendly and effective in producing actionable data through volunteer-machine teaming.
Speaking on her experience studying social media use in emergency response, Hughes pointed out that one obstacle is that smaller jurisdictions don’t have the resources to dedicate to this area. “We feel like a viable way to overcome that is to introduce this idea of people volunteering to do this type of work. It’s not hard work, but coordinating and doing it well is not easy, either,” said Hughes.
“If we can provide systems and AI that can help in this process, then I think we can make a real difference in how emergency responders are able to gather that information and respond to it.”
Students interested in getting involved with this project can contact
for more information.