By Uma Ravat (University of California at Santa Barbara)
Information
Believing AI is here to stay, I set out to incorporate it into my courses in ways that preserve the foundational learning students cannot afford to skip, build the AI literacy they will need in the workforce, and open up possibilities for deeper engagement. The core design challenge is how to integrate AI in ways that deepen student learning rather than allow students to shortcut it. Recent work in the statistics education community reflects this tension. Efforts to design AI tools that encourage students to attempt problems before receiving feedback (Çetinkaya-Rundel, 2025) signal a shared concern about preserving authentic learning in an age of AI-generated answers. This poster contributes to that conversation by documenting three course designs with evidence of what worked, what did not, and what remains to be learned. Each intervention embeds AI with a distinct role and a common design principle: AI should scaffold independent thinking, not substitute for it. In PSTAT 120B (Mathematical Statistics Theory), I deployed GauchoBot, a course-specific chatbot designed to support learning outside office hours by asking guiding questions rather than providing direct answers. Students completed weekly homework reflection surveys tracking help-seeking behaviors (independent attempts, human help, AI use), along with pre-, mid-, and post-course surveys measuring confidence, motivation, and AI attitudes. These measures capture self-efficacy, motivation, and help-seeking behaviors linked to student success (Spencer et al., 2023; Bandura, 1997; Deci & Ryan, 2000). In PSTAT 194CS (Computational Statistics), students used AI strictly as an editor—not author—on technical projects involving random number generation and social network analysis. AI use was constrained to prevent suggesting new methods, code, or interpretations. Students submitted structured AI Use Documentation describing prompts, decisions, and perceived limitations. In PSTAT 190 (Undergraduate Learning Assistant Training), students used AI as a simulated confused learner, prompting it to roleplay a struggling student who responds only to effective guiding questions. This provided low-stakes practice in asking rather than telling, a key pedagogical skill. Across all three contexts, students must bring their own understanding to every AI interaction. AI serves as a thinking partner, not a thinking replacement. This study was conducted at a large public R1 research university. Participants ranged from sophomore-level students in a calculus-based statistics course to upper-division majors and undergraduate peer educators. Class sizes ranged from approximately 200 in a large lecture to 50 in an elective to 15 in the training course.