Engineering Students' Prompting Strategies with ChatGPT to Solve Engineering Design Tasks
Main Article Content
Abstract
Record ID: 263
Award(s): Excellence in Undergraduate Research Mentorship; Excellence in Research Communication
Program Affiliation: Undergraduate Research Co-op Fellowship
Presentation Type: Poster
Abstract: This research project delves into students' prompting strategies when utilizing generative AI in the context of addressing engineering design problems. Amidst the growing adoption of generative AI platforms like ChatGPT, there is a growing need to explore how to effectively integrate them into educational settings and instruct students on appropriate usage. The current literature primarily focuses on students' usage preferences of generative AI, scant attention has been paid to analyzing their actual prompting strategies. Conducted within an introductory engineering course at a large Midwest university, our study involved administering a brainstorming assignment for a semi-autonomous robot project to teams of 3-4 students. The assignment involved beginning with a structured brainstorming session, followed by a conversation with ChatGPT. Students were tasked with planning at least 5 prompts for the AI system, engaging in dialogue with ChatGPT for at least 30 minutes, and subsequently integrating the new insights with the ideas from the traditional brainstorming session. Our research question centers on understanding the prompting strategies employed by students when interfacing with ChatGPT for engineering design problem-solving. We collected completed assignments from consenting teams (n = 100 teams, 396 prompts) and analyzed them using an inductive qualitative coding approach to identify recurring patterns in their interactions. Preliminary findings reveal that students predominantly seek information akin to conventional search engine queries, with a subset directly soliciting design solutions or asking the system to make evaluative comparisons. Notably, a proportion of prompts instructed ChatGPT to produce the "best" solution instead of generating multiple possible solutions.