Student Perceived Trust in Math problem Solving in a Human-AI Collaboration : A Comparison Study to understand trust and interpersonal trust dynamics.

Abstract

This master's thesis addresses the critical and underexplored area of student-perceived trust in mathematical problem-solving within human-AI collaboration, particularly with the rapid integration of ChatGPT into educational contexts. While advanced large language models like ChatGPT are rapidly integrating into educational contexts, there remains a significant gap in understanding the complex dynamics of student trust in AI-generated content and its subsequent impact on interpersonal trust when AI-assisted work is shared. The research aims to identify key factors influencing students' trust in ChatGPT's mathematical solutions. It further seeks to understand how the correctness of ChatGPT-generated solutions impacts both human-AI (H-AI) trust, and human-human using AI (H-HAI) interpersonal trust, and to examine the relationship between specific types of errors (conceptual vs. calculation) and student trust in ChatGPT's mathematical competency. Additionally, the research explores how perceived transparency, or lack thereof, regarding ChatGPT usage influences (H-HAI) interpersonal trust concerning a student's mathematical abilities. A quantitative approach was applied, utilizing a 45-item Likert-scale questionnaire administered to students with prior experience using ChatGPT in the University of Turku. Students evaluated ChatGPT-generated math solutions and assessed trust implications across both human-AI and human-human mediated by ChatGPT scenarios. The findings illuminate the complex interplay between correct and erroneous solutions (conceptual vs. calculation errors), the effect of perceived transparency or its absence, and the role of output quality in calibrating trust.