Where and how do people use ChatGPT in disaster scenarios? How well do large language models perform in answering real-time questions about emergent events? How can we design empathetic robot dogs to effectively support individuals with autism?
A group of 9 PhD fellows from 6 disciplines spent their summer applying trustworthy AI to address real-world problems just like these.
The fellows participated in a summer research incubator as part of the GW Co-Design of Trustworthy AI in Systems (DTAIS) program, a National Science Foundation Research Traineeship (NRT) that supports PhD students conducting research at the intersection of artificial intelligence and human work systems.
DTAIS fellows presented their research at a recent event organized by the GW Trustworthy AI Initiative, where they engaged with peers from both academia and industry.
Sarah Starr, a DTAIS fellow and PhD student in Engineering Management and Systems Engineering, participated in this summer's incubator and presented her work on human oversight of agentic AI at the showcase. The experience helped her turn an initial research question into a working human-subjects pilot study with preliminary results and jump-start the next phase of her dissertation research.
"The incubator was invaluable for the interdisciplinary feedback I got at every stage of designing and running my first human-subjects study end-to-end,” Starr began.
“Before I had a single participant session, peers and faculty pressure-tested my experimental design and worked through the user interface. Beyond catching problems I might otherwise have missed, their recommendations materially improved the study's flow and design, and I incorporated many of them directly into the final protocol,” she said.
“Once I had initial data, the same group helped me think through the analysis and how to best interpret the results. The research showcase was a great opportunity to share my preliminary findings and connect with researchers across the TAI community on the emerging challenges of agentic systems," she concluded.
GW TAI Program Director, Michael Lyons, remarked that the showcase highlighted the benefits of researching AI and trustworthiness through an interdisciplinary lens.
“The feedback and collaboration stemming from the summer incubator demonstrate the capabilities of our fellows and also the increased impact of their research when they are able to collaborate in a truly interdisciplinary manner,” Lyons began.
“The showcase spotlights their hard work on these research sprints to GW students and faculty wishing to learn more about trustworthy AI, as well as external guests and alumni who see the value that this research has on our wider community.”