Examining Self-Regulated Learning Processes in Design Tasks with Large Language Models

Focusing on human-AI collaboration process, where humans and AI work together toward shared goals, has implications for designing AI-enabled learning tasks.We explore selfregulated learning (SRL) in two tasks leveraging generative AI (Large Language Models; LLMs): (1) co-write a story with GPT-4 and (2) brainstorm ideas for design sketches with GPT-3.5.We develop a codebook that accounts for SRL phases (forethought, performance, selfreflection) and fine-grained cognitive and socioemotional interactions with LLMs.We conduct Ordered Network Analysis to compare SRL processes between participants scoring higher and lower in design artifacts.The higher-scoring participants more frequently utilized metacognitive strategies like monitoring and self-instruction to maintain focus in performing tasks, in conjunction with cognitive interactions to prompt and evaluate LLMs' outputs.We discuss the design of learning systems and tasks to promote evaluative strategies in working with LLMs.