May 20, 2026
In her last semester at Pitt, Michelle Star (SCI ’26) had her human-computer interaction (HCI) research paper accepted into the highly competitive Association for Computing Machinery (ACM) Student Research Competition.
Star was one of only six international undergraduate students to get their research paper accepted into the yearly Computer-Human Interaction (CHI) competition. Due to her acceptance into the prestigious competition, Star also attended the annual ACM CHI conference.
Star said being accepted into the competition and conference demonstrated to her the importance of her work, as CHI is the “gold standard” in HCI research.
“Being accepted... meant my work was rigorously reviewed and recognized as making a meaningful contribution to the field,” Star said. “It also showed me that the HCI community recognizes the critical need to audit AI systems for demographic bias, and that my focus on multi-turn evaluation addresses a real and urgent gap in how we currently test AI.”
Star’s submission to the competition, her paper titled “Evaluating Supportive LLM Behavior Over Multiple Turns across Demographics,” detailed her research into algorithmic bias in conversational AI. Star explained that her research asked whether AI is responding equally to all people who turn to it for support.
“I studied whether large language models, the technology behind AI chatbots, provide fair and consistent emotional support across different demographic groups,” Star said. “People are increasingly turning to AI chatbots for emotional support because they offer a fast, accessible, and low-barrier option. My research began with a gap I saw in the current literature on evaluating LLMs for emotional and mental health support.”
Many studies on this same subject only looked at one message and one response from AI chatbots. Star explained her research investigates longer, more complex conversations in order to see if the AI’s behavior changes over time.
“Real-world emotional support conversations with AI are multi-turn, with conversations unfolding back and forth over time. People do not present their problems all at once, wrapped up in a neat package,” Star said. “Instead, they open up gradually, revealing context and emotional details piece by piece.”
Star said she went about this research by creating conversations with an LLM to see if the AI changed or maintained its supportive behavior.
“I simulated conversations in which a user revealed pieces of support-seeking narratives over a conversation, while a supportive assistant LLM responded,” Star said. “I evaluated the model’s responses using sentiment analysis and an empathy classifier to measure how well the model matched emotional tone and whether it maintained supportive behavior across turns.”
The results of this work showed that the AI responded differently to various demographics, specifically, there was a statistical difference between the AI’s support of mothers and fathers, where the AI matched the emotional tone of mothers 80.6% of the time, while only matching 45.7% of the time for fathers. Star also showed that the AI could become less supportive as the conversation went on. Star said these results are important because they lead to a deeper understanding of how AI systems function in a realistic situation.
“These findings matter because LLMs are increasingly being used in emotionally sensitive contexts,” Star said. “If certain groups are quietly receiving less aligned or less consistent support, that raises a serious fairness concern.”
The 2026 ACM CHI conference, held in Barcelona, Spain, between April 13-17, gave Star an opportunity to travel internationally, attend the conference, and present her research to other attendees.
Star's experience listening, conversing, and learning about other research and work being done at the conference was eye-opening.
“The days are nonstop packed with papers, workshops, and meetups, all back-to-back,” Star said. “Though, some of my best conversations and connections happened in the in-between moments: on the green outside the venue, at coffee breaks, or lingering after sessions ended.”
The conference allowed Star to meet the people behind the research in a field she has a deep passion for. Star said being accepted into the student research competition and attending the conference was the highlight of her undergraduate journey.
“The conference opened my mind to potential new research avenues I may want to pursue in the future and left me feeling excited about what comes next,” Star said. “More than anything, it gave me a sense of clarity and reminded me what all of this work is for.”
Star recently graduated from Pitt, earning her Bachelor of Philosophy in Data Science and minor in computer science, and attended Pitt’s honors commencement at the end of April. Her experience with research and the CHI Student Research Competition exemplifies her growth as a researcher and showcases the passion that will propel her into her life at graduate school.