Co-intelligent classrooms: Developing custom AI agents for responsible AI in education

As Artificial Intelligence plays a growing role in education, students too often turn to it as a substitute for independent work rather than as a tool to support their learning. In our study, we developed custom AI agents that students in several courses could interact with as learning tools for specific topics. Topics included undergraduate physics, signal processing and filtering, and the physics of sound. Our agents were developed using curated, course-specific datasets and following ethical guidelines for educational AI use. We compared how agents made with different AI platforms (ChatGPT, Google Gemini, and Claude.ai) performed across three criteria: creator usability, accuracy of responses, and effectiveness of responsible AI guardrails. Student evaluators assessed each AI agent's strengths and weaknesses through logged interactions with a custom interface. Our process offers a model that could be replicated by educators across multiple disciplines who are interested in responsible ways AI could support student learning in their courses.