Enhancing Undergraduate Learning in Physics: The Effects of AI-Assisted and AI-Guided Instruction on Knowledge, Self-Efficacy, and Satisfaction

<p>Generative artificial intelligence (AI) tools such as large language models are increasingly used for self-directed learning, yet there is limited experimental evidence on whether they improve learning outcomes and whether theory-informed guidance adds value beyond AI use alone. Prior literature suggests that effective learning depends not only on access to information, but also on how instruction supports attention, working memory, long-term memory, and cognitive load, while also fostering motivational processes such as self-efficacy. Individual differences such as Need for Cognition may also shape how learners engage with cognitively demanding instructional tools such as AI.</p> <p>The present study examined how different forms of instructional support affect undergraduate learning in an introductory physics self-study task focused on electricity and circuits. Undergraduate participants with no prior college physics experience were randomly assigned to one of four conditions: video instruction, AI-only instruction, AI with theory-based guidance, or a no-instruction control. Using a pretest-posttest mixed design, the study assessed knowledge, physics self-efficacy, and learner satisfaction, while also examining Need for Cognition and coded AI engagement behaviors within the AI conditions.</p> <p>Results showed that participants in the instructional conditions demonstrated clear improvements in knowledge and self-efficacy relative to the control condition, and all instructional conditions produced substantially higher satisfaction than the control group. However, the three instructional formats did not differ meaningfully from one another on the primary learning outcomes, indicating that AI-based instruction and video instruction were similarly effective for short-term learning in this controlled setting. Within the AI conditions, the theory-based guidance changed aspects of learner engagement, including greater use of guidance-aligned and feedback-seeking prompts, but did not yield better immediate knowledge, self-efficacy, or satisfaction outcomes than AI alone. Need for Cognition was also examined as an individual-difference variable, but it was not significantly associated with the primary outcomes.</p> <p>Overall, the findings suggest that AI can function as an effective short-term instructional tool for novice learners in a self-study environment, producing outcomes comparable to a traditional video-based format. At the same time, brief prompting guidance may influence how learners interact with AI without necessarily improving immediate performance. These findings have implications for the design of AI-supported instruction in higher education and suggest that future research should examine longer-term retention, transfer, and the conditions under which additional scaffolding is most beneficial.</p>

Name thesis:degree_name

Doctor of Philosophy in Human Factors

Level thesis:degree_level

Dissertation - Open Access

Discipline thesis:degree_discipline

Human Factors and Behavioral Neurobiology

Year

2026

Author dc:creator

• Woods, Stephen

• Artificial Intelligence (AI)
• Human Factors
• Learning
• Knowledge
• Training
• Human Factors Psychology

Repository record dc:identifier

https://commons.erau.edu/edt/972

OAI identifier oai:identifier

oai:commons.erau.edu:edt-2014