More than getting the answer: How students engage with generative AI in technical problem solving

Abstract

Generative artificial intelligence (GenAI) can produce complete solutions and explanations for quantitative STEM problems, raising questions about the forms of cognitive engagement students enact when using these tools. Although prior research has examined AI use, perceptions, and performance, less is known about how students engage with GenAI during structured technical problem solving. This study examined 38 undergraduate students in an operations management course who first solved a linear programming problem manually and then used GenAI on the same task. Student reflections were analyzed using deductive thematic analysis guided by the ICAP framework, with available AI interaction records used to validate classifications and develop illustrative cases. Three engagement patterns were documented. Passive engagement involved receiving AI-generated output without using the interaction to compare, verify, reinterpret, or extend understanding. Active engagement involved using AI to check or confirm previously completed reasoning, whereas Constructive engagement involved seeking explanation or conceptual interpretation, particularly around sensitivity analysis, shadow prices, and binding constraints. No reflection or available interaction record provided sufficient evidence of Interactive engagement, including in cases involving multiple conversational turns. The findings show that students can position GenAI differently within the same technical task, from receiving answers to verifying existing reasoning or seeking conceptual explanation. These distinctions suggest that STEM educators should attend not only to whether students use GenAI, but also to the forms of engagement that instructional tasks and expectations invite.