When many people think about artificial intelligence, they picture fast code and smart machines. However, Yin-Chun Lu wants the focus to return to people. Lu, a second-year Ph.D. student in NC State’s Industrial and Systems Engineering Department, studies how humans form expectations. Recently, her paper was selected as one of 12 finalists out of 62 submissions to the Student Research Competition at CHI 2026. As a result, she will travel to Barcelona to present her single-authored research.
Lu studies how people make quick judgments in uncertain situations. For example, if someone stands by a road, you might assume they will stay there. According to Lu, that instinct reflects a continuation bias. Furthermore, she argues that such bias shapes the data used to train AI systems. “My research focuses on how to measure non-biased high-level cognitive judgments, specifically how people form expectations in uncertain situations,” Lu said. In other words, she studies what people think will happen next. Consequently, her work looks beyond visible behavior and into human thought.
Much AI training data comes from crowdsourcing. In these systems, many people label images or predict outcomes. Lu calls the shared pattern in these judgments “collective expectation.” “I use the term ‘collective expectation’ to describe the aggregate pattern that emerges from these shared judgments,” she said. However, most datasets capture only movement or actions. “Most AI datasets capture observable behavior such as movement or trajectories,” Lu said. “However, expectations are also critical because they influence how humans make decisions.” Therefore, she tests whether task framing can reduce bias in crowd labels and improve AI training data.
Meanwhile, Renran Tian, an assistant professor in the Industrial and Systems Engineering Department, believes the project addresses a major challenge in AI. “As AI becomes prevalent, it is critical to ensure the smooth integration of AI-enabled systems into human lives,” Tian said. In addition, he noted that AI must reflect human thinking patterns. “One major challenge in such ubiquitous interactions is ensuring that AI can imitate the high-level cognition underlying how humans make decisions and communicate with others,” Tian said. Therefore, he said Lu’s approach offers a scalable way to benchmark human higher-level cognition.
Because graduate SRC papers must be single-authored, Lu led the project independently. “Preparing a single-authored submission pushed me to take full ownership of the research and drive the project forward from start to finish,” Lu said. However, she also values collaboration. “I realized that independent research does not mean working alone,” Lu said. She thanked Tian for guidance and feedback, and her lab mates for support.
In April, Lu will present her work in person in Barcelona. To prepare, she is refining her visuals and delivery. “To help people quickly grasp the key message, I am refining the visual layout of my poster so that the figures can tell the story on their own,” she said. Ultimately, Lu believes AI must stay grounded in human values. “People talk about AI constantly, but I believe the conversation ultimately has to return to humans because AI is built on human goals, human data and human decisions,” Lu said.
This post was originally published in the Edward P. Fitts Department of Industrial and Systems Engineering.
