The Educational Impact of Biased AI Coding Assistants: Usage Patterns and Student Performance in a Course on AI-Agent Social Simulation

As an extension of the project "Biased AI improves human decision-making but reduces trust," this study examines whether embedding differently biased perspectives along the dimension of novelty-reliability in a coding assistant (a Claude Code agent) affects students' productivity and innovativity in weekly homework and research-question development over a 3-month course on AI Agents for Social Science and Society. This study will provide fresh evidence on (a) how students' engagement with a coding AI assistant relates to productivity and innovativity in iterative homework and research-question development in an AI Agents for Social Science and Society course, and (b) whether introducing task-relevant biases (novelty-biased, reliability-biased, and a balanced neutral policy) to coding assistant(s) is associated with systematic differences in students' weekly outputs and working processes.