Simret Gebreegziabher, a doctoral student in the Department of Computer Science and Engineering at the University of Notre Dame, has received an IBM PhD Fellowship Award for the 2024-2025 academic year.

The prestigious award recognizes Gebreegziabher’s work to date within the realm of responsible Human-Artificial Intelligence (AI) interaction and matches her with IBM mentors as she continues research to develop frameworks that democratize AI.

“Humans inherently possess diverse values, intentions, and goals,” explains Gebreegziabher. “This diversity shapes how individuals interact with AI models and what they want from the models. I’m interested in how differences in values, intentions, and goals challenge human-AI collaboration, particularly when the interaction involves uncertainty or ambiguity.

“Specifically, my research helps us to better understand what the model is learning—both how it falls short and how users might customize it.”

Gebreegziabher points to “hate speech” as an instructive example of the ambiguity that is her research focus. An AI model learns patterns that differentiate “hate speech” from user-labeled data—a task made difficult by humans’ subjective perception and the lack of a universal definition of “hate speech.” Gebreegziabher’s research asks: How can we best define the boundaries of this category during the model’s training as well as allow users to appeal or challenge the model?

“A certain model might be ‘overfitted’ to specific words, resulting in it flagging benign content as harmful simply because it contains those terms, while potentially overlooking context or nuance,” Gebreegziahber explains.

“When users can discern and refine the model’s classification boundaries while reflecting on their own understanding of the concepts, AI can be of most benefit to the most users.”

Gebreegziabher is the lead author of three peer-reviewed papers and has been a primary student contributor to the “Evaluation, Metrics, and Benchmarks of Large Language Models (LLMs)” project at the ND-IBM Technology Ethics Lab, a key component of the Notre Dame Institute for Ethics and the Common Good.

Her adviser, Toby Jia-Jun Li, is assistant professor in the department of computer science and engineering and director of the newly founded Human-Centered Responsible AI (HRAI) Lab at the Lucy Family Institute for Data & Society.

“Simret’s work is vital in ensuring that AI systems not only perform tasks but also reflect complex human ethics and intentions, addressing a critical gap in AI,” said Li. “Such research is crucial as AI becomes pervasive in society, where misalignments can perpetuate biases and societal harm.”

IBM has awarded Gebreegziabher a $40,000 stipend for the fellowship year.

“I am deeply grateful to my advisor, and collaborators for their invaluable support and guidance,” said Gebreegziabher. “This recognition motivates me to continue advancing research that bridges AI with human-centered values.”