2026 ASEE Annual Conference & Exposition

Three years after the release of ChatGPT, debates on the use of generative AI in higher education continue. Educators are still seeking guidance and examples of how to integrate these tools into courses to intentionally prepare students for an AI-enabled future. In Electrical Engineering, some connections with AI are more obvious–where students are expected to use these tools to code and optimize systems, and more curricula are beginning to teach students how these systems work–but even these uses will require students to develop more complex skills to appropriately leverage these tools. In particular, critical thinking as an engineering skill (e.g., articulating and challenging assumptions in problem solving, designing experiments, and collaborating with other engineers)–skills that map to multiple ABET Student Outcomes–will become more important than ever, as AI easily generates correct-sounding false arguments.

To prepare students in how they may use AI in their work, while helping them build critical thinking skills in a disciplinary context, we have created a “novice” chatbot for Solid State Devices and Materials, a required core course in the electrical engineering major that most students take in their junior year. The chatbot is introduced to students late in the course and is used generally as a resource students can use to practice course concepts, and specifically in conjunction with targeted exercises to help students practice evaluating AI output.

In our presentation, we will present a customized retrieval-augmented generator (RAG) chatbot built on ChatGPT for Solid State Devices and Materials, and how we tuned it to perform worse than the base LLM. We will share how we tested to assure the chatbot’s course performance was appropriately degraded to a “novice” (targeting ~75% correct or a C level performance), and what learning activities leveraged the chatbot target students’ critical thinking skills (primarily critiquing proposed AI-generated solutions). In particular, we targeted the following concepts, in which students frequently make mistakes: temperature dependance; drift; mobility; diffusion of carriers; and metal-semiconductor junctions and contacts. This is opposite to how many others have used chatbots to support student learning (e.g., automated TAs). As far as the authors are aware, this application of an intentionally errant AI-chatbot has only been applied to a chemical engineering course–a course that the underlying LLM likely has much less trending data than Solid State Devices and Materials, and a far more visual-analysis driven course–and no equivalent examples of an errant AI chatbot have been reported in electrical engineering. As such, our project will test whether the strategies employed to create a novice chatbot for chemical engineering still hold for a more conceptual, text-driven course. By sharing our experience, we hope to demonstrate that approaches to create “novice” chatbots are transferable, and that use of intentionally trained chatbots can help students understand how they may more critically interact with AI in their future engineering practice.